update taskbar apps (#3286)

Include 'builder' in the RECOMMENDED_APP_NAMES array in src/backend/services/apps/RecommendedAppsService.ts so the RecommendedAppsService treats the builder app as a recommended application and surfaces it accordingly.

.github/workflows/backend-tests.yaml

@@ -3,6 +3,8 @@ name: Backend Tests
 on:
   pull_request:
     types: [opened, synchronize, reopened]
+    paths-ignore:
+      - 'src/docs/**'
 
 permissions:
   contents: read

.github/workflows/docs-build.yml

@@ -0,0 +1,27 @@
+name: Docs Build
+
+on:
+  pull_request:
+    paths:
+      - 'src/docs/**'
+
+jobs:
+  build-docs:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '24'
+          cache: 'npm'
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Build docs
+        run: npm run build
+        working-directory: src/docs

AGENTS.md

@@ -64,9 +64,9 @@ Keep comments light. Prefer self-documenting code — clear names, small functio
 - The _why_ is non-obvious (a hidden constraint, a workaround, a subtle invariant).
 - A non-trivial usage detail would otherwise trip the next reader.
 
-Don't restate what the code already says. Don't write comments that reference the current task or PR — those rot.
+Don't restate what the code already says. Don't write comments that reference the current task or PR or version of something — those rot.
 
-**No ticket references in code or test comments.** Don't write `PUT-1234`, `AUTH-3`, `GUI-2`, `// fix for FOO-99`, `// per ROLLOUT-1`, or section headers like `// -- PUT-1020: ... --`. The ticket exists in git history and the issue tracker; the code should describe the *why* in domain terms ("logout cascade transparently invalidates...") not project-management terms ("the cascade from AUTH-3..."). Same rule for test-block names and section dividers — describe what's being tested, not which ticket asked for it.
+**No ticket references in code or test comments.** Don't write `PUT-1234`, `AUTH-3`, `GUI-2`, `// fix for FOO-99`, `// per ROLLOUT-1`, or section headers like `// -- PUT-1020: ... --`. The ticket exists in git history and the issue tracker; the code should describe the _why_ in domain terms ("logout cascade transparently invalidates...") not project-management terms ("the cascade from AUTH-3..."). Same rule for test-block names and section dividers — describe what's being tested, not which ticket asked for it.
 
 **Use plain ASCII `-` in comment section dividers, not box-drawing characters.** Write `// -- Section name --`, not `// ── Section name ──`. The `─` (U+2500) character renders inconsistently in some editors and diff tools and adds nothing over `-`.
 

config.template.jsonc

@@ -260,6 +260,10 @@
         // ─ Chat / completion ─
         "claude": { "apiKey": "" },
         "openai-completion": { "apiKey": "" },
+        "azure-openai": {
+            "apiKey": "",
+            "apiURL": ""
+        },
         "gemini": { "apiKey": "" },
         "groq": { "apiKey": "" },
         "deepseek": { "apiKey": "" },

doc/i18n/README.jp.md

@@ -104,7 +104,7 @@ Puterは[**puter.com**](https://puter.com)でホストサービスとして利
 
 メンテナーやコミュニティと以下のチャンネルを通じてつながりましょう：
 
-- バグ報告や機能リクエストがありますか？ [issueを開く](https://github.com/HeyPuter/puter/issues/new/choose) してください。
+- バグ報告や機能リクエストは [新規Issue](https://github.com/HeyPuter/puter/issues/new/choose) まで
 - Discord: [discord.com/invite/PQcx7Teh8u](https://discord.com/invite/PQcx7Teh8u)
 - X (Twitter): [x.com/HeyPuter](https://x.com/HeyPuter)
 - Reddit: [reddit.com/r/puter/](https://www.reddit.com/r/puter/)

doc/self-hosting.md

@@ -199,6 +199,32 @@ Drop the resulting `fullchain.pem` and `privkey.pem` into `./puter/tls/`.
     "s3": { "s3Config": { "publicEndpoint": "https://s3.puter.local", ... } }
     ```
 
+## Running behind your own reverse proxy
+
+The bundled `puter-nginx` mirrors production and is the supported default — it terminates TLS (Step 3) and forwards every Host to Puter. But plenty of self-hosters already run their own edge proxy (Caddy, Traefik, HAProxy, a cloud load balancer, another nginx). You can put Puter behind it instead; Puter doesn't terminate TLS itself in any setup, so "your proxy does TLS" is just a matter of pointing it at the Puter container and getting a handful of details right.
+
+You can keep the bundled nginx as the single hop your proxy talks to, or bypass it and forward straight to the Puter container on port `4100` (uncomment the `4100:4100` mapping under the `puter` service in [docker-compose.yml](../docker-compose.yml), or attach your proxy to the compose network). Either works — what matters is the rules below.
+
+**The rules that actually matter** (getting any of these wrong is what causes the redirect loops and "Invalid Host header" failures people hit):
+
+1. **Don't rewrite the `Host` header.** Puter routes entirely on Host — `api.<domain>`, `site.<domain>`, `app.<domain>` are all distinguished by the incoming host. Forward the original host through unchanged. Do **not** rewrite external hostnames to an internal name like `puter.local`; that forces you to remap every subdomain by hand and still breaks signed-URL and CORS checks. Most proxies preserve Host by default — just don't override it.
+
+2. **The external domain must equal `domain` in your `config.json`.** If users reach the box at `puter.example.com`, then `domain` must be `puter.example.com` and the hosting domains must be its real subdomains (`site.puter.example.com`, `app.puter.example.com`, …) — exactly as the installer/`config.json` lays them out. Puter rejects hosts it doesn't recognize with `Invalid Host header`.
+
+3. **Set `protocol` to match the public scheme.** When your proxy terminates TLS, set `"protocol": "https"` (installer: `PUTER_PROTOCOL=https`). Puter builds its origins, redirects, signed S3 URLs, and OIDC callback URLs from this — leave it `http` behind an HTTPS proxy and you get redirect loops, broken logins, and mixed-content errors.
+
+4. **Forward the standard proxy headers.** Puter needs:
+    - `Host` — the original request host (rule 1).
+    - `X-Forwarded-Proto` — the **external** scheme (`https`), so Puter knows TLS was terminated upstream.
+    - `X-Forwarded-For` — the real client IP (rate limiting + audit logs).
+    - `Upgrade` / `Connection` — passed through for WebSocket / socket.io upgrades, or the realtime connection silently fails.
+
+5. **Set `trust_proxy` to the number of hops in front of Puter.** One external proxy talking directly to Puter → `"trust_proxy": 1` (installer: `PUTER_TRUST_PROXY=1`). A second proxy in front (e.g. Cloudflare → your proxy → Puter) → `2`. This is the count of proxies between the client and Puter, **including** the bundled nginx if you keep it. Too low and `req.ip` becomes a proxy address (breaks rate limiting); never set `true` (it trusts every hop and makes `X-Forwarded-For` forgeable).
+
+6. **Route the wildcard to Puter.** Your proxy must forward `*.<domain>` (covering `api.*`, `app.*`, and `s3.<domain>`) and `*.site.<domain>` (and any other hosting domains) to Puter — same wildcard DNS as Step 2. Puter does the per-subdomain routing internally; the proxy just needs to hand it the traffic with the Host intact.
+
+With those in place the rest of the stack is unchanged — `docker compose up -d` as below.
+
 ## Step 4 — Bring it up
 
 ```bash

docker-compose.yml

@@ -264,7 +264,7 @@ services:
       # Persistent runtime data (anything your config points at /var/puter).
       - ./puter/data/puter:/var/puter:z
     healthcheck:
-      test: wget --no-verbose --tries=1 --spider http://puter.localhost:4100/test || exit 1
+      test: wget --no-verbose --tries=1 --spider http://localhost:4100/ || exit 1
       interval: 30s
       timeout: 3s
       retries: 3

extensions/installedApps.ts

@@ -71,6 +71,6 @@ export const handleInstalledApps = async (
 
 extension.get(
     '/installedApps',
-    { subdomain: 'api', requireUserActor: true },
+    { subdomain: 'api', requireUserActor: true, allowFullAccessToken: true },
     handleInstalledApps,
 );

extensions/metering.ts

@@ -1,4 +1,3 @@
-import type { Request, Response } from 'express';
 import { Context } from '@heyputer/backend/src/core';
 import { HttpError } from '@heyputer/backend/src/core/http';
 import {
@@ -6,6 +5,7 @@ import {
     driversContainers,
 } from '@heyputer/backend/src/exports';
 import { extension } from '@heyputer/backend/src/extensions';
+import type { Request, Response } from 'express';
 
 const services = extension.import('service');
 const clients = extension.import('client');

extensions/whoami.ts

@@ -72,6 +72,9 @@ export const handleWhoami = async (
         unconfirmed_email: user.email,
         email_confirmed: user.email_confirmed || user.username === 'admin',
         requires_email_confirmation: user.requires_email_confirmation,
+        phone: user.phone,
+        requires_phone_verification: user.requires_phone_verification,
+        requires_card_verification: user.requires_card_verification,
         desktop_bg_url: user.desktop_bg_url,
         desktop_bg_color: user.desktop_bg_color,
         desktop_bg_fit: user.desktop_bg_fit,

install.sh

@@ -16,11 +16,20 @@
 # the compose file + brings the stack up. Set PUTER_FORCE=1 to overwrite.
 #
 # Tunable env vars:
-#   PUTER_DIR     install directory                       (default: ./puter-selfhosted)
-#   PUTER_URL     base URL to fetch docker-compose.yml    (default: GitHub raw, main branch)
-#   PUTER_DOMAIN  domain Puter will serve on              (default: puter.localhost)
-#   PUTER_PORT    HTTP port for nginx                     (default: 80)
-#   PUTER_FORCE   set to 1 to overwrite existing .env / config.json
+#   PUTER_DIR         install directory                   (default: ./puter-selfhosted)
+#   PUTER_URL         base URL to fetch docker-compose.yml (default: GitHub raw, main branch)
+#   PUTER_DOMAIN      domain Puter will serve on           (default: puter.localhost)
+#   PUTER_PORT        HTTP port for nginx                  (default: 80)
+#   PUTER_PROTOCOL    public scheme: http | https          (default: http)
+#                     Set https when a TLS-terminating reverse proxy (Caddy,
+#                     Traefik, nginx, a cloud LB) sits in front — Puter then
+#                     builds https:// origins, S3 URLs, and redirects. See
+#                     "Running behind your own reverse proxy" in doc/self-hosting.md.
+#   PUTER_TRUST_PROXY number of reverse-proxy hops in front (default: 1)
+#                     1 = the bundled nginx (or a single external proxy);
+#                     2 = two hops (e.g. Cloudflare → your proxy → Puter).
+#   PUTER_ENV         prod | dev                            (default: prod)
+#   PUTER_FORCE       set to 1 to overwrite existing .env / config.json
 
 set -eu
 
@@ -28,6 +37,9 @@ PUTER_DIR="${PUTER_DIR:-puter-selfhosted}"
 PUTER_URL="${PUTER_URL:-https://raw.githubusercontent.com/HeyPuter/puter/main}"
 PUTER_DOMAIN="${PUTER_DOMAIN:-puter.localhost}"
 PUTER_PORT="${PUTER_PORT:-80}"
+PUTER_PROTOCOL="${PUTER_PROTOCOL:-http}"
+PUTER_TRUST_PROXY="${PUTER_TRUST_PROXY:-1}"
+PUTER_ENV="${PUTER_ENV:-prod}"
 PUTER_FORCE="${PUTER_FORCE:-0}"
 
 log()  { printf '\033[1;36m[puter-install]\033[0m %s\n' "$*"; }
@@ -117,9 +129,9 @@ EOF
     cat > puter/config/config.json <<EOF
 {
     "domain": "$PUTER_DOMAIN",
-    "protocol": "http",
+    "protocol": "$PUTER_PROTOCOL",
     "pub_port": $PUTER_PORT,
-    "env": "prod",
+    "env": "$PUTER_ENV",
 
     "static_hosting_domain": "site.$PUTER_DOMAIN",
     "static_hosting_domain_alt": "host.$PUTER_DOMAIN",
@@ -159,7 +171,7 @@ EOF
     "s3": {
         "s3Config": {
             "endpoint": "http://s3:9000",
-            "publicEndpoint": "http://s3.$PUTER_DOMAIN",
+            "publicEndpoint": "$PUTER_PROTOCOL://s3.$PUTER_DOMAIN",
             "accessKeyId": "puter",
             "secretAccessKey": "$S3_SECRET_KEY",
             "region": "us-east-1",
@@ -173,7 +185,7 @@ EOF
         "ollama": { "enabled": false }
     },
 
-    "trust_proxy": 1
+    "trust_proxy": $PUTER_TRUST_PROXY
 }
 EOF
 fi
@@ -187,6 +199,6 @@ log "stack starting. first boot takes ~30s while MariaDB initialises."
 log "follow puter logs:"
 log "    cd $PUTER_DIR && docker compose logs -f puter"
 log ""
-log "open http://$PUTER_DOMAIN:$PUTER_PORT once the puter container is healthy."
+log "open $PUTER_PROTOCOL://$PUTER_DOMAIN:$PUTER_PORT once the puter container is healthy."
 log "first-boot admin password is logged once — grab it with:"
 log "    cd $PUTER_DIR && docker compose logs puter | grep password"

package.json

@@ -44,18 +44,18 @@
     "yaml": "^2.8.1"
   },
   "scripts": {
-    "test:backend": "vitest run --config src/backend/vitest.config.ts ",
-    "test:backend:postgres": "PUTER_TEST_DB_ENGINE=postgres vitest run --config src/backend/vitest.config.ts ",
+    "test:backend": "npm run setupExtensions && vitest run --config src/backend/vitest.config.ts ",
+    "test:backend:postgres": "npm run setupExtensions && PUTER_TEST_DB_ENGINE=postgres vitest run --config src/backend/vitest.config.ts ",
     "start=gui": "nodemon --exec \"node dev-server.js\" ",
     "start": "node --enable-source-maps -r ./dist/src/backend/telemetry.js ./dist/src/backend/index.js",
-    "prestart": "npm run build:ts",
+    "prestart": "npm run setupExtensions && npm run build:ts",
     "dev": "npm start",
-    "build": "npm run build:ts && cd src/gui && node ./build.js && cd ../puter-js && npm run build",
+    "build": "npm run setupExtensions && npm run build:ts && cd src/gui && node ./build.js && cd ../puter-js && npm run build",
     "build:workerLib": "cd src/puter-js && npm run build && cd ../worker && npm run build",
     "check-translations": "node tools/check-translations.js",
     "prepare": "husky",
     "build:ts": "tsc -p tsconfig.build.json && node ./tools/write-dist-package-json.mjs",
-    "postinstall": "node ./tools/extensionSetup.mjs"
+    "setupExtensions": "node ./tools/extensionSetup.mjs"
   },
   "workspaces": [
     "src/*",
@@ -77,9 +77,16 @@
     "ai": "^6.0.73",
     "dedent": "^1.5.3",
     "javascript-time-ago": "^2.5.11",
+    "libphonenumber-js": "1.13.6",
     "open": "^10.1.0"
   },
   "engines": {
     "node": ">=24.0.0"
+  },
+  "allowScripts": {
+    "[email protected]": true,
+    "[email protected]": true,
+    "@google/genai": false,
+    "[email protected]": true
   }
 }

src/backend/clients/database/SqliteDatabaseClient.ts

@@ -85,6 +85,9 @@ const AVAILABLE_MIGRATIONS: [number, string[]][] = [
     [50, ['0054_sessions_workers.sql']],
     [50, ['0055_username_nocase_unique.sql']],
     [51, ['0056_sessions_kind_worker.sql']],
+    [52, ['0057_add_user_reputation.sql']],
+    [53, ['0058_add_phone_verification.sql']],
+    [54, ['0059_add_card_verification.sql']],
 ];
 
 export class SqliteDatabaseClient extends AbstractDatabaseClient {

src/backend/clients/database/migrations/mysql/mysql_mig_1.sql

@@ -24,7 +24,9 @@
 --
 
 --
--- Idempotent column-ensure helper (used by CALLs below; dropped at end of file)
+-- Idempotent column-ensure helper. Used by the CALLs below AND by later
+-- migration files, which CALL it as a one-line idempotent column add. Left
+-- resident (not dropped at end of file) so it outlives this migration.
 --
 
 DROP PROCEDURE IF EXISTS _puter_add_col;
@@ -1333,7 +1335,10 @@ CALL _puter_add_col('user_update_audit', 'new_username', '`new_username` varchar
 CALL _puter_add_col('user_update_audit', 'reason', '`reason` varchar(255) COLLATE utf8mb4_unicode_ci DEFAULT NULL');
 
 
-DROP PROCEDURE IF EXISTS _puter_add_col;
+-- _puter_add_col is intentionally NOT dropped here: later migration files CALL
+-- it as a one-line idempotent column add, and migrations replay on every boot
+-- with no applied-state tracking, so the helper must outlive this file. The
+-- DROP-before-CREATE at the top keeps this migration itself replay-safe.
 /*!40103 SET TIME_ZONE=@OLD_TIME_ZONE */;
 
 /*!40101 SET SQL_MODE=@OLD_SQL_MODE */;

src/backend/clients/database/migrations/mysql/mysql_mig_13.sql

@@ -0,0 +1,47 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- SMS phone verification columns. Mirrors SQLite migration 0058. `phone` is the
+-- E.164 number collected during verification (indexed like `email`);
+-- `requires_phone_verification` gates account use for low-reputation signups
+-- (not indexed, mirroring `requires_email_confirmation`).
+--
+-- Idempotent: column adds use _puter_add_col (defined in mig_1, which leaves it
+-- resident for later migrations); the index add is guarded against
+-- INFORMATION_SCHEMA.STATISTICS so the directory replays safely.
+
+CALL _puter_add_col('user', 'phone', '`phone` varchar(20) DEFAULT NULL');
+CALL _puter_add_col('user', 'requires_phone_verification', '`requires_phone_verification` tinyint(1) NOT NULL DEFAULT ''0''');
+
+DROP PROCEDURE IF EXISTS _puter_add_user_phone_index;
+DELIMITER //
+CREATE PROCEDURE _puter_add_user_phone_index()
+BEGIN
+  IF NOT EXISTS (
+    SELECT 1 FROM INFORMATION_SCHEMA.STATISTICS
+    WHERE TABLE_SCHEMA = DATABASE()
+      AND TABLE_NAME = 'user'
+      AND INDEX_NAME = 'idx_user_phone'
+  ) THEN
+    ALTER TABLE `user` ADD INDEX `idx_user_phone` (`phone`);
+  END IF;
+END//
+DELIMITER ;
+
+CALL _puter_add_user_phone_index();
+
+DROP PROCEDURE IF EXISTS _puter_add_user_phone_index;

src/backend/clients/database/migrations/mysql/mysql_mig_14.sql

@@ -0,0 +1,26 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Credit-card verification column. Mirrors SQLite migration 0059.
+-- `requires_card_verification` gates account use for low-reputation signups;
+-- the card itself never touches our DB, so this is the only column (not
+-- indexed, mirroring `requires_phone_verification`).
+--
+-- Idempotent: the column add uses _puter_add_col (from mig_1) so the
+-- directory replays safely.
+
+CALL _puter_add_col('user', 'requires_card_verification', '`requires_card_verification` tinyint(1) NOT NULL DEFAULT ''0''');

src/backend/clients/database/migrations/mysql/mysql_mig_15.sql

@@ -0,0 +1,47 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Card fingerprint column. Mirrors SQLite migration 0060. `card_fingerprint`
+-- is the Stripe card fingerprint (stable per card number) recorded when a user
+-- clears card verification — the card sibling of `phone`, indexed like it so
+-- admin tooling can find the accounts that verified with a given card. The card
+-- itself never touches our DB, only Stripe's fingerprint for it.
+--
+-- Idempotent: the column add uses _puter_add_col (defined in mig_1, which
+-- leaves it resident for later migrations); the index add is guarded against
+-- INFORMATION_SCHEMA.STATISTICS so the directory replays safely.
+
+CALL _puter_add_col('user', 'card_fingerprint', '`card_fingerprint` varchar(128) DEFAULT NULL');
+
+DROP PROCEDURE IF EXISTS _puter_add_user_card_fingerprint_index;
+DELIMITER //
+CREATE PROCEDURE _puter_add_user_card_fingerprint_index()
+BEGIN
+  IF NOT EXISTS (
+    SELECT 1 FROM INFORMATION_SCHEMA.STATISTICS
+    WHERE TABLE_SCHEMA = DATABASE()
+      AND TABLE_NAME = 'user'
+      AND INDEX_NAME = 'idx_user_card_fingerprint'
+  ) THEN
+    ALTER TABLE `user` ADD INDEX `idx_user_card_fingerprint` (`card_fingerprint`);
+  END IF;
+END//
+DELIMITER ;
+
+CALL _puter_add_user_card_fingerprint_index();
+
+DROP PROCEDURE IF EXISTS _puter_add_user_card_fingerprint_index;

src/backend/clients/database/migrations/postgres/postgres_mig_2.sql

@@ -0,0 +1,26 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- SMS phone verification columns. Mirrors SQLite migration 0058. `phone` is the
+-- E.164 number collected during verification (indexed like `email`);
+-- `requires_phone_verification` gates account use for low-reputation signups
+-- (not indexed, mirroring `requires_email_confirmation`).
+-- Idempotent via IF NOT EXISTS.
+
+ALTER TABLE "user" ADD COLUMN IF NOT EXISTS phone varchar(20);
+ALTER TABLE "user" ADD COLUMN IF NOT EXISTS requires_phone_verification boolean NOT NULL DEFAULT FALSE;
+CREATE INDEX IF NOT EXISTS idx_user_phone ON "user" (phone);

src/backend/clients/database/migrations/postgres/postgres_mig_3.sql

@@ -0,0 +1,24 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Credit-card verification column. Mirrors SQLite migration 0059.
+-- `requires_card_verification` gates account use for low-reputation signups;
+-- the card itself never touches our DB, so this is the only column (not
+-- indexed, mirroring `requires_phone_verification`).
+-- Idempotent via IF NOT EXISTS.
+
+ALTER TABLE "user" ADD COLUMN IF NOT EXISTS requires_card_verification boolean NOT NULL DEFAULT FALSE;

src/backend/clients/database/migrations/postgres/postgres_mig_4.sql

@@ -0,0 +1,26 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Card fingerprint column. Mirrors SQLite migration 0060. `card_fingerprint`
+-- is the Stripe card fingerprint (stable per card number) recorded when a user
+-- clears card verification — the card sibling of `phone`, indexed like it so
+-- admin tooling can find the accounts that verified with a given card. The card
+-- itself never touches our DB, only Stripe's fingerprint for it.
+-- Idempotent via IF NOT EXISTS.
+
+ALTER TABLE "user" ADD COLUMN IF NOT EXISTS card_fingerprint varchar(128);
+CREATE INDEX IF NOT EXISTS idx_user_card_fingerprint ON "user" (card_fingerprint);

src/backend/clients/database/migrations/sqlite/0057_add_user_reputation.sql

@@ -0,0 +1,21 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Abuse v2 reputation score (0–100+) recorded at signup. Mirrors the
+-- `reputation smallint DEFAULT 100` column already present in the mysql and
+-- postgres baseline migrations.
+ALTER TABLE `user` ADD COLUMN `reputation` SMALLINT DEFAULT 100;

src/backend/clients/database/migrations/sqlite/0058_add_phone_verification.sql

@@ -0,0 +1,25 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- SMS phone verification. `phone` holds the E.164 number collected during
+-- verification (indexed like `email`). `requires_phone_verification` gates
+-- account use until verified — the abuse v2 harness sets it for low-reputation
+-- signups instead of blocking them. (Not indexed, mirroring
+-- `requires_email_confirmation`.)
+ALTER TABLE `user` ADD COLUMN `phone` varchar(20) DEFAULT NULL;
+ALTER TABLE `user` ADD COLUMN `requires_phone_verification` tinyint(1) NOT NULL DEFAULT 0;
+CREATE INDEX IF NOT EXISTS idx_user_phone ON `user` (`phone`);

src/backend/clients/database/migrations/sqlite/0059_add_card_verification.sql

@@ -0,0 +1,23 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Credit-card verification ($0 authorization handled by a payments
+-- extension). `requires_card_verification` gates account use until verified —
+-- the abuse v2 harness sets it for low-reputation signups instead of blocking
+-- them. The card itself never touches our DB, so this is the only column.
+-- (Not indexed, mirroring `requires_phone_verification`.)
+ALTER TABLE `user` ADD COLUMN `requires_card_verification` tinyint(1) NOT NULL DEFAULT 0;

src/backend/clients/database/migrations/sqlite/0060_add_card_fingerprint.sql

@@ -0,0 +1,25 @@
+-- Copyright (C) 2024-present Puter Technologies Inc.
+--
+-- This file is part of Puter.
+--
+-- Puter is free software: you can redistribute it and/or modify
+-- it under the terms of the GNU Affero General Public License as published
+-- by the Free Software Foundation, either version 3 of the License, or
+-- (at your option) any later version.
+--
+-- This program is distributed in the hope that it will be useful,
+-- but WITHOUT ANY WARRANTY; without even the implied warranty of
+-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+-- GNU Affero General Public License for more details.
+--
+-- You should have received a copy of the GNU Affero General Public License
+-- along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+-- Card fingerprint column. `card_fingerprint` is the Stripe card fingerprint
+-- (stable per card number) recorded when a user clears card verification — the
+-- card sibling of `phone`. Indexed like `phone` so admin tooling can find the
+-- accounts that verified with a given card. The card itself never touches our
+-- DB, only Stripe's fingerprint for it. Server-only (not on the whoami
+-- allowlist), so it never reaches clients.
+ALTER TABLE `user` ADD COLUMN `card_fingerprint` varchar(128) DEFAULT NULL;
+CREATE INDEX IF NOT EXISTS idx_user_card_fingerprint ON `user` (`card_fingerprint`);

src/backend/clients/event/types.ts

@@ -14,6 +14,10 @@
  *   the SocketService fans out to user-scoped channels.
  */
 
+import type {
+    Request as ExpressRequest,
+    Response as ExpressResponse,
+} from 'express';
 import { Actor } from '../../core';
 import { FSEntry } from '../../stores/fs/FSEntry';
 
@@ -126,7 +130,20 @@ export type EventMap = {
         req?: unknown;
         data?: unknown;
         abuse?: unknown;
-        custom?: unknown;
+        trail?: Array<string>;
+        /**
+         * Set by the abuse harness for flagged signups — the id under which the
+         * decision trail is persisted to KV (`abuse:trail:<id>`), shared back on
+         * the request for log / support correlation.
+         */
+        trail_id?: string;
+        /** Device signals forwarded verbatim from the signup request body. */
+        fingerprint?: string | null;
+        dfp_telemetry_id?: string | null;
+        /** Set by the abuse harness — require SMS phone verification post-signup. */
+        requires_phone_verification?: boolean;
+        /** Set by the abuse harness — require card verification post-signup. */
+        requires_card_verification?: boolean;
         [key: string]: unknown;
     };
     'puter.signup.success': {
@@ -135,6 +152,9 @@ export type EventMap = {
         email: string;
         username: string;
         ip?: string | null;
+        fingerprint?: string | null;
+        /** True when the created account is a temp user (no email/password). */
+        is_temp?: boolean;
         [key: string]: unknown;
     };
     'email.validate': {
@@ -154,12 +174,93 @@ export type EventMap = {
         user_uid: string;
         email: string;
     };
+    // Phone-reuse cap is pure mechanism here — the abuse extension answers
+    // (emitted via `emitAndWait`) by counting how many OTHER accounts have
+    // already verified this number and flipping `allowed` to false when the
+    // cross-account limit is hit. No extension listening → `allowed` stays
+    // true and the send proceeds.
+    'puter.phone-verification.check': {
+        user_id: number;
+        user_uid: string;
+        phone: string;
+        // Client-supplied device fingerprint for this request, or null. Lets
+        // the abuse extension cap sends per device (across accounts) so a
+        // device farm can't dodge the per-account cap with fresh signups.
+        device_fingerprint: string | null;
+        allowed: boolean;
+        reason: string | null;
+        [key: string]: unknown;
+    };
+    // Fire-and-forget signal that a code was actually sent — the abuse
+    // extension bumps its per-number / per-account / per-device send-velocity
+    // counters off this. No-op with no extension listening.
+    'puter.phone-verification.sent': {
+        user_id: number;
+        user_uid: string;
+        phone: string;
+        device_fingerprint: string | null;
+    };
+    'user.phone-verified': {
+        user_id: number;
+        user_uid: string;
+        phone: string;
+    };
+    // Card verification is pure mechanism here — a payments extension fills
+    // these in (emitted via `emitAndWait`). `enabled` stays null when no
+    // extension is installed; the extension always sets it (true/false) so
+    // the endpoints can distinguish "disabled" from "not installed".
+    'puter.card-verification.setup': {
+        user_id: number;
+        user_uid: string;
+        ip?: string | null;
+        // Client-supplied device fingerprint for this request, or null. Lets
+        // the abuse extension cap card-verification setups per device (across
+        // accounts) before any Stripe SetupIntent is created.
+        device_fingerprint: string | null;
+        enabled: boolean | null;
+        // Set false by the extension to refuse this setup (e.g. the per-device
+        // setup-velocity cap); `reason` carries the opaque code. Stays true
+        // when allowed or with no extension listening.
+        allowed: boolean;
+        reason: string | null;
+        client_secret: string | null;
+        publishable_key: string | null;
+        [key: string]: unknown;
+    };
+    'puter.card-verification.confirm': {
+        user_id: number;
+        user_uid: string;
+        setup_intent_id: string;
+        enabled: boolean | null;
+        verified: boolean;
+        reason: string | null;
+        fingerprint: string | null;
+        funding: string | null;
+        country: string | null;
+        [key: string]: unknown;
+    };
+    'user.card-verified': {
+        user_id: number;
+        user_uid: string;
+        fingerprint: string | null;
+        funding: string | null;
+        country: string | null;
+    };
     'user.username-changed': {
         user_id: number;
         old_username: string;
         new_username: string;
     };
     'user.email-changed': { user_id: number; new_email: string };
+    // Fired after an account is torn down (self-serve, admin, or temp-user
+    // logout cleanup). Listeners purge external state tied to the account —
+    // e.g. the marketplace extension cancels the user's Stripe subscriptions.
+    // The row is already gone by emit time, so identifiers ride the payload.
+    'user.delete': {
+        user_id: number;
+        user_uuid?: string;
+        stripe_customer_id?: string | null;
+    };
 
     // ---- Filesystem ----
     'fs.copy.node': {
@@ -211,6 +312,8 @@ export type EventMap = {
         entry: unknown;
         host: string;
         requestPath: string;
+        requestUrl?: string;
+        requestHash?: string;
         mime: string;
     };
 
@@ -257,6 +360,94 @@ export type EventMap = {
     // SocketService re-emits each fanout-eligible event under
     // `sent-to-user.<wireName>` so per-user channels can subscribe by wire name.
     [K in `sent-to-user.${string}`]: { user_id: number; response: unknown };
+} & {
+    // Generic per-driver-method lifecycle, emitted by DriverController for
+    // EVERY driver call under a key scoped to the interface + method:
+    // `driver.<iface>.<method>.before|after|error|reject`. Wildcard
+    // subscribers can listen to `driver.*` for everything, `driver.<iface>.*`
+    // for one interface, or the exact key for one method. The `.before` phase
+    // is emitted via `emitAndWait`, so a listener may set `allow = false`
+    // (with an optional `rejectReason`) to veto the call before it runs — a
+    // vetoed call emits `.reject` instead of running.
+    [K in `driver.${string}`]: DriverMethodLifecycleEvent;
+} & {
+    // Generic per-route-endpoint lifecycle, emitted by the route materializer
+    // for EVERY non-middleware route under a key scoped to the HTTP method +
+    // normalized path: `route.<method>.<path>.before|after|error|reject`. Same
+    // wildcard + veto semantics as the driver lifecycle above.
+    [K in `route.${string}`]: RouteLifecycleEvent;
+};
+
+/**
+ * Phase of a request/method lifecycle. `reject` is emitted when a `before`
+ * listener vetoes the call (sets `allow = false`); the call never runs and no
+ * `after`/`error` follows.
+ */
+export type LifecyclePhase = 'before' | 'after' | 'error' | 'reject';
+
+/**
+ * Payload for `driver.<iface>.<method>.<phase>` events.
+ *
+ * One shape across all phases; read `phase` (or the key suffix) to
+ * branch. `allow`/`rejectReason` are only meaningful on the `before` phase
+ * (emitted via `emitAndWait`).
+ */
+export type DriverMethodLifecycleEvent = {
+    phase: LifecyclePhase;
+    iface: string;
+    method: string;
+    /** Resolved concrete driver name. */
+    driver: string;
+    /** Full actor object, if the request is authenticated. */
+    actor?: Actor;
+    /** Stable actor id (see `actorUid`), if the request is authenticated. */
+    actorUid?: string;
+    /** Call arguments. Present on every phase. */
+    args?: unknown;
+    /** Return value. Present on `after`. */
+    result?: unknown;
+    /** Thrown error. Present on `error`. */
+    error?: unknown;
+    /** Wall-clock duration of the invocation. Present on `after`/`error`. */
+    durationMs?: number;
+    /** Veto channel for `before`: set `false` to block the call. */
+    allow?: boolean;
+    /** Optional human-readable reason surfaced to the caller when vetoed. */
+    rejectReason?: string;
+};
+
+/**
+ * Payload for `route.<method>.<path>.<phase>` events. Mirrors
+ * {@link DriverMethodLifecycleEvent} for HTTP endpoints.
+ */
+export type RouteLifecycleEvent = {
+    phase: LifecyclePhase;
+    /** HTTP method, lowercased (`get`, `post`, ...). */
+    method: string;
+    /** Full route path including the controller prefix. */
+    path: string;
+    /**
+     * The live express request/response for this call. Present on every phase.
+     * On `before` a listener can read the parsed body/headers or write its own
+     * response; on terminal phases they're useful for logging. These are real
+     * in-process objects — never serialize or forward them across nodes.
+     */
+    req: ExpressRequest;
+    res: ExpressResponse;
+    /** Full actor object, if the request is authenticated. */
+    actor?: Actor;
+    /** Stable actor id (see `actorUid`), if the request is authenticated. */
+    actorUid?: string;
+    /** Response status code. Present on `after`/`error`. */
+    statusCode?: number;
+    /** Wall-clock duration from `before` to terminal phase. */
+    durationMs?: number;
+    /** Set when the request did not complete normally (abort / >=500). */
+    error?: unknown;
+    /** Veto channel for `before`: set `false` to block the request. */
+    allow?: boolean;
+    /** Optional human-readable reason surfaced to the caller when vetoed. */
+    rejectReason?: string;
 };
 
 export type EventKey = keyof EventMap & string;

src/backend/clients/index.ts

@@ -24,6 +24,7 @@ import { EventClient } from './event/EventClient';
 import { DDBClient } from './dynamodb/DDBClient';
 import { RedisClient } from './redis/RedisClient';
 import { S3Client } from './s3/S3Client';
+import { PreludeClient } from './prelude/PreludeClient';
 import type { IPuterClientRegistry } from './types';
 
 export const puterClients = {
@@ -34,4 +35,5 @@ export const puterClients = {
     dynamo: DDBClient,
     redis: RedisClient,
     s3: S3Client,
+    prelude: PreludeClient,
 } satisfies IPuterClientRegistry;

src/backend/clients/prelude/PreludeClient.test.ts

@@ -0,0 +1,173 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import { PreludeClient } from './PreludeClient';
+import type { IConfig } from '../../types';
+
+const makeClient = (apiKey?: string) =>
+    new PreludeClient({
+        prelude: apiKey ? { apiKey } : undefined,
+    } as unknown as IConfig);
+
+const okJson = (body: unknown) =>
+    ({
+        ok: true,
+        status: 200,
+        json: async () => body,
+    }) as unknown as Response;
+
+describe('PreludeClient', () => {
+    let fetchMock: ReturnType<typeof vi.fn>;
+
+    beforeEach(() => {
+        fetchMock = vi.fn();
+        vi.stubGlobal('fetch', fetchMock);
+    });
+    afterEach(() => {
+        vi.unstubAllGlobals();
+    });
+
+    it('isConfigured reflects whether an apiKey is set', () => {
+        expect(makeClient('sk_test').isConfigured()).toBe(true);
+        expect(makeClient().isConfigured()).toBe(false);
+    });
+
+    describe('isCountrySupported (€0.07 cap)', () => {
+        const client = makeClient('sk_test');
+
+        it('allows revenue markets up to the cap (incl. the priciest)', () => {
+            expect(client.isCountrySupported('US')).toBe(true); // €0.0043
+            expect(client.isCountrySupported('DE')).toBe(true); // €0.0598
+            expect(client.isCountrySupported('SA')).toBe(true); // €0.0638
+            expect(client.isCountrySupported('us')).toBe(true); // case-insensitive
+        });
+
+        it('rejects countries above the cap, with no SMS, or unknown', () => {
+            expect(client.isCountrySupported('PK')).toBe(false); // €0.3548
+            expect(client.isCountrySupported('ID')).toBe(false); // €0.2430
+            expect(client.isCountrySupported('LI')).toBe(false); // null (no SMS)
+            expect(client.isCountrySupported('ZZ')).toBe(false); // unknown
+            expect(client.isCountrySupported(undefined)).toBe(false);
+        });
+
+        it('honors a configured maxSmsCostEur override', () => {
+            const strict = new PreludeClient({
+                prelude: { apiKey: 'sk', maxSmsCostEur: 0.01 },
+            } as unknown as IConfig);
+            expect(strict.isCountrySupported('US')).toBe(true); // €0.0043
+            expect(strict.isCountrySupported('DE')).toBe(false); // €0.0598 > 0.01
+
+            const loose = new PreludeClient({
+                prelude: { apiKey: 'sk', maxSmsCostEur: 0.5 },
+            } as unknown as IConfig);
+            expect(loose.isCountrySupported('PK')).toBe(true); // €0.3548 <= 0.5
+        });
+    });
+
+    it('createVerification POSTs the phone target + ip signal with bearer auth', async () => {
+        fetchMock.mockResolvedValue(
+            okJson({ id: 'vrf_1', status: 'success' }),
+        );
+        const client = makeClient('sk_test');
+
+        const res = await client.createVerification('+14155550123', {
+            ip: '203.0.113.7',
+        });
+
+        expect(res).toEqual({ id: 'vrf_1', status: 'success' });
+        const [url, init] = fetchMock.mock.calls[0];
+        expect(url).toBe('https://api.prelude.dev/v2/verification');
+        expect(init.method).toBe('POST');
+        expect(init.headers.Authorization).toBe('Bearer sk_test');
+        expect(JSON.parse(init.body)).toEqual({
+            target: { type: 'phone_number', value: '+14155550123' },
+            // Defaults to RCS (cheaper); Prelude falls back to SMS.
+            options: { code_size: 6, preferred_channel: 'rcs' },
+            signals: { ip: '203.0.113.7' },
+        });
+    });
+
+    it('forwards device_id and user_agent signals when supplied', async () => {
+        fetchMock.mockResolvedValue(okJson({ id: 'v', status: 'success' }));
+        const client = makeClient('sk_test');
+
+        await client.createVerification('+14155550123', {
+            ip: '203.0.113.7',
+            device_id: 'thumb_abc123',
+            user_agent: 'Mozilla/5.0',
+        });
+
+        const [, init] = fetchMock.mock.calls[0];
+        expect(JSON.parse(init.body).signals).toEqual({
+            ip: '203.0.113.7',
+            device_id: 'thumb_abc123',
+            user_agent: 'Mozilla/5.0',
+        });
+    });
+
+    it('omits the signals object entirely when none are supplied', async () => {
+        fetchMock.mockResolvedValue(okJson({ id: 'v', status: 'success' }));
+        const client = makeClient('sk_test');
+
+        await client.createVerification('+14155550123');
+
+        expect(JSON.parse(fetchMock.mock.calls[0][1].body)).not.toHaveProperty(
+            'signals',
+        );
+    });
+
+    it('includes a configured template_id + sender_id + preferred channel', async () => {
+        fetchMock.mockResolvedValue(okJson({ id: 'v', status: 'success' }));
+        const client = new PreludeClient({
+            prelude: {
+                apiKey: 'sk_test',
+                templateId: 'tmpl_puter',
+                senderId: 'Puter',
+                preferredChannel: 'sms',
+            },
+        } as unknown as IConfig);
+
+        await client.createVerification('+14155550123');
+
+        const [, init] = fetchMock.mock.calls[0];
+        expect(JSON.parse(init.body).options).toEqual({
+            code_size: 6,
+            preferred_channel: 'sms', // config override beats the rcs default
+            template_id: 'tmpl_puter',
+            sender_id: 'Puter',
+        });
+    });
+
+    it('checkVerification POSTs target + code and returns the status', async () => {
+        fetchMock.mockResolvedValue(okJson({ status: 'success' }));
+        const client = makeClient('sk_test');
+
+        const res = await client.checkVerification('+14155550123', '123456');
+
+        expect(res).toEqual({ status: 'success' });
+        const [url, init] = fetchMock.mock.calls[0];
+        expect(url).toBe('https://api.prelude.dev/v2/verification/check');
+        expect(JSON.parse(init.body)).toEqual({
+            target: { type: 'phone_number', value: '+14155550123' },
+            code: '123456',
+        });
+    });
+
+    it('throws (does not call fetch) when not configured', async () => {
+        const client = makeClient();
+        await expect(
+            client.createVerification('+14155550123'),
+        ).rejects.toThrow(/not configured/i);
+        expect(fetchMock).not.toHaveBeenCalled();
+    });
+
+    it('throws on a non-2xx Prelude response', async () => {
+        fetchMock.mockResolvedValue({
+            ok: false,
+            status: 400,
+            json: async () => ({ message: 'bad target' }),
+        } as unknown as Response);
+
+        await expect(
+            makeClient('sk_test').checkVerification('+1', 'x'),
+        ).rejects.toThrow(/Prelude .* failed: 400/);
+    });
+});

src/backend/clients/prelude/PreludeClient.ts

@@ -0,0 +1,222 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import type { IConfig } from '../../types';
+import { PuterClient } from '../types';
+import { COUNTRY_SMS_PRICES } from './countries.js';
+
+const PRELUDE_API_BASE = 'https://api.prelude.dev/v2';
+const REQUEST_TIMEOUT_MS = 8000;
+/** OTP length — matches the 6-box code UI. Prelude allows 4–8. */
+const PRELUDE_CODE_SIZE = 6;
+/**
+ * Channel Prelude prioritizes for delivery. RCS is far cheaper than SMS, so we
+ * prefer it by default; Prelude routes to the next reachable channel (SMS) when
+ * RCS isn't available on the destination. (Requires an RCS agent provisioned in
+ * the Prelude account to actually use RCS — otherwise it just falls through to
+ * SMS.) Override with `config.prelude.preferredChannel`.
+ */
+const DEFAULT_PREFERRED_CHANNEL: PreludeChannel = 'rcs';
+
+/** Delivery channels Prelude can prioritize via `options.preferred_channel`. */
+export type PreludeChannel =
+    | 'sms'
+    | 'rcs'
+    | 'whatsapp'
+    | 'viber'
+    | 'zalo'
+    | 'telegram';
+/**
+ * Default per-SMS cost ceiling (EUR). Countries whose Prelude SMS rate exceeds
+ * this — or that have no SMS channel — are not offered phone verification. The
+ * cap covers every realistic revenue market (priciest are Germany €0.0598 and
+ * Saudi Arabia €0.0638) while excluding the expensive, high-fraud long tail.
+ * Override per-deployment with `config.prelude.maxSmsCostEur`.
+ */
+const DEFAULT_MAX_SMS_COST_EUR = 0.07;
+
+/** Status returned by Prelude when creating/retrying a verification. */
+export type PreludeCreateStatus =
+    | 'success'
+    | 'retry'
+    | 'challenged'
+    | 'blocked'
+    | 'shadow_blocked';
+
+/** Status returned by Prelude when checking a code. */
+export type PreludeCheckStatus =
+    | 'success'
+    | 'failure'
+    | 'expired_or_not_found'
+    | 'transaction_missing'
+    | 'transaction_mismatch';
+
+/**
+ * Prelude Verify v2 client (https://docs.prelude.so/verify/v2). Sends and checks
+ * SMS one-time codes — Prelude generates, delivers, and validates the code, so
+ * we never store one ourselves. No-ops with a warning when no API key is
+ * configured (so dev environments don't crash); `isConfigured()` lets callers
+ * surface a clean "phone verification unavailable" error instead.
+ */
+export class PreludeClient extends PuterClient {
+    private apiKey: string | null = null;
+
+    constructor(config: IConfig) {
+        super(config);
+        this.apiKey = config.prelude?.apiKey ?? null;
+    }
+
+    override onServerStart(): void {
+        if (!this.apiKey) {
+            console.warn(
+                '[prelude] no apiKey configured — SMS phone verification is disabled',
+            );
+        }
+    }
+
+    /** True when an API key is configured and verification can be attempted. */
+    isConfigured(): boolean {
+        return !!this.apiKey;
+    }
+
+    /** ISO region used to parse local-format numbers (config-driven). */
+    get defaultCountry(): string | undefined {
+        return this.config.prelude?.defaultCountry;
+    }
+
+    /** Per-SMS cost ceiling in EUR (config override or the default cap). */
+    get maxSmsCostEur(): number {
+        return this.config.prelude?.maxSmsCostEur ?? DEFAULT_MAX_SMS_COST_EUR;
+    }
+
+    /**
+     * Whether SMS verification should be offered for a country. False when the
+     * country is unknown, has no SMS channel, or its rate exceeds the cost cap.
+     * @param iso ISO-3166 alpha-2 (e.g. 'US') — from the parsed phone number.
+     */
+    isCountrySupported(iso: string | undefined): boolean {
+        if (!iso) return false; // couldn't determine country → can't price it
+        const price = COUNTRY_SMS_PRICES[iso.toUpperCase()];
+        if (!price || price.sms == null) return false;
+        return price.sms <= this.maxSmsCostEur;
+    }
+
+    /**
+     * Create (or retry) a verification: Prelude sends an OTP to `target`.
+     * @param target E.164 phone number, e.g. "+14155550123".
+     * @param signals Optional anti-fraud signals. The more we forward, the
+     *   better Prelude's fraud model scores the attempt (per their docs IP +
+     *   device + platform are the highest-value). We pass what the backend
+     *   already has: the request `ip`, the client `device_id` (ThumbmarkJS
+     *   hash), and the `user_agent` (Prelude infers platform/model/OS from it).
+     *   Richer client-collected signals require Prelude's frontend SDK, which
+     *   produces a `dispatch_id` the backend would forward here.
+     */
+    async createVerification(
+        target: string,
+        signals: {
+            ip?: string;
+            device_id?: string;
+            user_agent?: string;
+        } = {},
+    ): Promise<{ id?: string; status: PreludeCreateStatus }> {
+        // Match the 6-box code UI (UIWindowPhoneVerificationRequired). Without
+        // code_size Prelude uses the dashboard default (4). preferred_channel
+        // prioritizes RCS (cheaper); Prelude falls back to SMS when unavailable.
+        const options: Record<string, unknown> = {
+            code_size: PRELUDE_CODE_SIZE,
+            preferred_channel:
+                this.config.prelude?.preferredChannel ??
+                DEFAULT_PREFERRED_CHANNEL,
+        };
+        // Branding lives in the Prelude dashboard (the message text is a
+        // template); these just select a Puter-branded template / sender when
+        // configured. See IPreludeConfig.
+        const { templateId, senderId } = this.config.prelude ?? {};
+        if (templateId) options.template_id = templateId;
+        if (senderId) options.sender_id = senderId;
+
+        const body: Record<string, unknown> = {
+            target: { type: 'phone_number', value: target },
+            options,
+        };
+        // Only attach signals we actually have — Prelude treats the object as
+        // optional and an empty one adds nothing.
+        const sig: Record<string, string> = {};
+        if (signals.ip) sig.ip = signals.ip;
+        if (signals.device_id) sig.device_id = signals.device_id;
+        if (signals.user_agent) sig.user_agent = signals.user_agent;
+        if (Object.keys(sig).length > 0) body.signals = sig;
+        return this.#post('/verification', body) as Promise<{
+            id?: string;
+            status: PreludeCreateStatus;
+        }>;
+    }
+
+    /**
+     * Check a code the user entered against the active verification for `target`.
+     * @returns `{ status }` — `'success'` means verified.
+     */
+    async checkVerification(
+        target: string,
+        code: string,
+    ): Promise<{ status: PreludeCheckStatus }> {
+        return this.#post('/verification/check', {
+            target: { type: 'phone_number', value: target },
+            code,
+        }) as Promise<{ status: PreludeCheckStatus }>;
+    }
+
+    async #post(
+        path: string,
+        body: Record<string, unknown>,
+    ): Promise<Record<string, unknown>> {
+        if (!this.apiKey) {
+            throw new Error('Prelude is not configured');
+        }
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), REQUEST_TIMEOUT_MS);
+        try {
+            const res = await fetch(`${PRELUDE_API_BASE}${path}`, {
+                method: 'POST',
+                headers: {
+                    Authorization: `Bearer ${this.apiKey}`,
+                    'Content-Type': 'application/json',
+                    Accept: 'application/json',
+                },
+                body: JSON.stringify(body),
+                signal: controller.signal,
+            });
+            const json = (await res.json().catch(() => ({}))) as Record<
+                string,
+                unknown
+            >;
+            if (!res.ok) {
+                throw new Error(
+                    `Prelude ${path} failed: ${res.status} ${
+                        (json as { message?: string })?.message ?? ''
+                    }`.trim(),
+                );
+            }
+            return json;
+        } finally {
+            clearTimeout(timer);
+        }
+    }
+}

src/backend/clients/prelude/countries.ts

@@ -0,0 +1,952 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+/** Per-country Prelude SMS pricing, hardcoded from the Prelude global price
+ * sheet. `sms` is the average cost in EUR, or null when SMS is unavailable
+ * there. Used to cap phone verification to affordable countries (see
+ * PreludeClient.isCountrySupported). Regenerate from the price sheet when
+ * Prelude updates rates. */
+export interface CountryPrice {
+    name: string;
+    /** Prelude SMS average in EUR, or null when SMS is not available. */
+    sms: number | null;
+}
+
+export const COUNTRY_SMS_PRICES: Record<string, CountryPrice> = {
+    AD: {
+        name: 'Andorra',
+        sms: 0.0515,
+    },
+    AE: {
+        name: 'United Arab Emirates',
+        sms: 0.0145,
+    },
+    AF: {
+        name: 'Afghanistan',
+        sms: 0.2409,
+    },
+    AG: {
+        name: 'Antigua and Barbuda',
+        sms: 0.0755,
+    },
+    AI: {
+        name: 'Anguilla',
+        sms: 0.0765,
+    },
+    AL: {
+        name: 'Albania',
+        sms: 0.0345,
+    },
+    AM: {
+        name: 'Armenia',
+        sms: 0.118,
+    },
+    AN: {
+        name: 'Netherlands Antilles',
+        sms: null,
+    },
+    AO: {
+        name: 'Angola',
+        sms: 0.0347,
+    },
+    AR: {
+        name: 'Argentina',
+        sms: 0.0614,
+    },
+    AS: {
+        name: 'American Samoa',
+        sms: 0.077,
+    },
+    AT: {
+        name: 'Austria',
+        sms: 0.014,
+    },
+    AU: {
+        name: 'Australia',
+        sms: 0.0091,
+    },
+    AW: {
+        name: 'Aruba',
+        sms: 0.0617,
+    },
+    AZ: {
+        name: 'Azerbaijan',
+        sms: 0.0969,
+    },
+    BA: {
+        name: 'Bosnia and Herzegovina',
+        sms: 0.0801,
+    },
+    BB: {
+        name: 'Barbados',
+        sms: 0.2021,
+    },
+    BD: {
+        name: 'Bangladesh',
+        sms: 0.1844,
+    },
+    BE: {
+        name: 'Belgium',
+        sms: 0.0297,
+    },
+    BF: {
+        name: 'Burkina Faso',
+        sms: 0.063,
+    },
+    BG: {
+        name: 'Bulgaria',
+        sms: 0.0799,
+    },
+    BH: {
+        name: 'Bahrain',
+        sms: 0.014,
+    },
+    BI: {
+        name: 'Burundi',
+        sms: 0.217,
+    },
+    BJ: {
+        name: 'Benin',
+        sms: 0.13,
+    },
+    BM: {
+        name: 'Bermuda',
+        sms: 0.21,
+    },
+    BN: {
+        name: 'Brunei',
+        sms: 0.035,
+    },
+    BO: {
+        name: 'Bolivia',
+        sms: 0.0708,
+    },
+    BQ: {
+        name: 'Caribbean Netherlands',
+        sms: 0.0565,
+    },
+    BR: {
+        name: 'Brazil',
+        sms: 0.0086,
+    },
+    BS: {
+        name: 'Bahamas',
+        sms: 0.0672,
+    },
+    BT: {
+        name: 'Bhutan',
+        sms: null,
+    },
+    BW: {
+        name: 'Botswana',
+        sms: 0.0235,
+    },
+    BY: {
+        name: 'Belarus',
+        sms: 0.118,
+    },
+    BZ: {
+        name: 'Belize',
+        sms: 0.1529,
+    },
+    CA: {
+        name: 'Canada',
+        sms: 0.0052,
+    },
+    CD: {
+        name: 'Congo RDC',
+        sms: 0.1112,
+    },
+    CF: {
+        name: 'Central African Republic',
+        sms: 0.2,
+    },
+    CG: {
+        name: 'Congo',
+        sms: 0.2507,
+    },
+    CH: {
+        name: 'Switzerland',
+        sms: 0.0163,
+    },
+    CI: {
+        name: "Cote d'Ivoire",
+        sms: 0.19,
+    },
+    CK: {
+        name: 'Cook Islands',
+        sms: 0.1041,
+    },
+    CL: {
+        name: 'Chile',
+        sms: 0.0027,
+    },
+    CM: {
+        name: 'Cameroon',
+        sms: 0.1504,
+    },
+    CN: {
+        name: 'China',
+        sms: 0.0053,
+    },
+    CO: {
+        name: 'Colombia',
+        sms: 0.0008,
+    },
+    CR: {
+        name: 'Costa Rica',
+        sms: 0.0045,
+    },
+    CU: {
+        name: 'Cuba',
+        sms: 0.0512,
+    },
+    CV: {
+        name: 'Cabo Verde',
+        sms: 0.2031,
+    },
+    CW: {
+        name: 'Curacao',
+        sms: 0.0138,
+    },
+    CY: {
+        name: 'Northern Cyprus',
+        sms: 0.0065,
+    },
+    CZ: {
+        name: 'Czech Republic',
+        sms: 0.0299,
+    },
+    DE: {
+        name: 'Germany',
+        sms: 0.0598,
+    },
+    DJ: {
+        name: 'Djibouti',
+        sms: 0.0897,
+    },
+    DK: {
+        name: 'Denmark',
+        sms: 0.0301,
+    },
+    DM: {
+        name: 'Dominica',
+        sms: 0.0824,
+    },
+    DO: {
+        name: 'Dominican Republic',
+        sms: 0.0351,
+    },
+    DZ: {
+        name: 'Algeria',
+        sms: 0.196,
+    },
+    EC: {
+        name: 'Ecuador',
+        sms: 0.1041,
+    },
+    EE: {
+        name: 'Estonia',
+        sms: 0.0234,
+    },
+    EG: {
+        name: 'Egypt',
+        sms: 0.1561,
+    },
+    ER: {
+        name: 'Eritrea',
+        sms: 0.0659,
+    },
+    ES: {
+        name: 'Spain',
+        sms: 0.0193,
+    },
+    ET: {
+        name: 'Ethiopia',
+        sms: 0.2741,
+    },
+    FI: {
+        name: 'Finland',
+        sms: 0.043,
+    },
+    FJ: {
+        name: 'Fiji',
+        sms: 0.069,
+    },
+    FK: {
+        name: 'Falkland Islands',
+        sms: 0.0713,
+    },
+    FM: {
+        name: 'Micronesia',
+        sms: 0.0118,
+    },
+    FO: {
+        name: 'Faroe Islands',
+        sms: 0.0319,
+    },
+    FR: {
+        name: 'France',
+        sms: 0.03,
+    },
+    GA: {
+        name: 'Gabon',
+        sms: 0.15,
+    },
+    GB: {
+        name: 'United Kingdom',
+        sms: 0.026,
+    },
+    GD: {
+        name: 'Grenada',
+        sms: null,
+    },
+    GE: {
+        name: 'Georgia',
+        sms: 0.0788,
+    },
+    GF: {
+        name: 'French Guiana',
+        sms: 0.05,
+    },
+    GG: {
+        name: 'Guernsey',
+        sms: 0.025,
+    },
+    GH: {
+        name: 'Ghana',
+        sms: 0.15,
+    },
+    GI: {
+        name: 'Gibraltar',
+        sms: 0.0134,
+    },
+    GL: {
+        name: 'Greenland',
+        sms: 0.0048,
+    },
+    GM: {
+        name: 'Gambia',
+        sms: 0.1283,
+    },
+    GN: {
+        name: 'Guinea',
+        sms: 0.2,
+    },
+    GP: {
+        name: 'Guadeloupe',
+        sms: 0.0455,
+    },
+    GQ: {
+        name: 'Equatorial Guinea',
+        sms: 0.099,
+    },
+    GR: {
+        name: 'Greece',
+        sms: 0.0296,
+    },
+    GT: {
+        name: 'Guatemala',
+        sms: 0.1202,
+    },
+    GU: {
+        name: 'Guam',
+        sms: 0.02,
+    },
+    GW: {
+        name: 'Guinea-Bissau',
+        sms: 0.1566,
+    },
+    GY: {
+        name: 'Guyana',
+        sms: 0.2178,
+    },
+    HK: {
+        name: 'Hong Kong',
+        sms: 0.038,
+    },
+    HN: {
+        name: 'Honduras',
+        sms: 0.162,
+    },
+    HR: {
+        name: 'Croatia',
+        sms: 0.03,
+    },
+    HT: {
+        name: 'Haiti',
+        sms: 0.09,
+    },
+    HU: {
+        name: 'Hungary',
+        sms: 0.029,
+    },
+    ID: {
+        name: 'Indonesia',
+        sms: 0.243,
+    },
+    IE: {
+        name: 'Ireland',
+        sms: 0.0305,
+    },
+    IL: {
+        name: 'Israel',
+        sms: 0.01,
+    },
+    IM: {
+        name: 'Isle of Man',
+        sms: 0.0385,
+    },
+    IN: {
+        name: 'India',
+        sms: 0.0375,
+    },
+    IQ: {
+        name: 'Iraq',
+        sms: 0.151,
+    },
+    IR: {
+        name: 'Iran',
+        sms: 0.14,
+    },
+    IS: {
+        name: 'Iceland',
+        sms: 0.0493,
+    },
+    IT: {
+        name: 'Italy',
+        sms: 0.0245,
+    },
+    JE: {
+        name: 'Jersey',
+        sms: 0.025,
+    },
+    JM: {
+        name: 'Jamaica',
+        sms: 0.1525,
+    },
+    JO: {
+        name: 'Jordan',
+        sms: 0.2094,
+    },
+    JP: {
+        name: 'Japan',
+        sms: 0.017,
+    },
+    KE: {
+        name: 'Kenya',
+        sms: 0.125,
+    },
+    KG: {
+        name: 'Kyrgyzstan',
+        sms: 0.15,
+    },
+    KH: {
+        name: 'Cambodia',
+        sms: 0.1529,
+    },
+    KI: {
+        name: 'Kiribati',
+        sms: 0.025,
+    },
+    KM: {
+        name: 'Comoros',
+        sms: 0.15,
+    },
+    KN: {
+        name: 'Saint Kitts and Nevis',
+        sms: 0.1295,
+    },
+    KR: {
+        name: 'South Korea',
+        sms: 0.006,
+    },
+    KW: {
+        name: 'Kuwait',
+        sms: 0.145,
+    },
+    KY: {
+        name: 'Cayman Islands',
+        sms: 0.2172,
+    },
+    KZ: {
+        name: 'Kazakhstan',
+        sms: 0.202,
+    },
+    LA: {
+        name: 'Laos',
+        sms: 0.15,
+    },
+    LB: {
+        name: 'Lebanon',
+        sms: 0.153,
+    },
+    LC: {
+        name: 'Saint Lucia',
+        sms: 0.0892,
+    },
+    LI: {
+        name: 'Liechtenstein',
+        sms: null,
+    },
+    LK: {
+        name: 'Sri Lanka',
+        sms: 0.3593,
+    },
+    LR: {
+        name: 'Liberia',
+        sms: 0.1448,
+    },
+    LS: {
+        name: 'Lesotho',
+        sms: 0.0301,
+    },
+    LT: {
+        name: 'Lithuania',
+        sms: 0.0261,
+    },
+    LU: {
+        name: 'Luxembourg',
+        sms: 0.034,
+    },
+    LV: {
+        name: 'Latvia',
+        sms: 0.028,
+    },
+    LY: {
+        name: 'Libya',
+        sms: 0.1898,
+    },
+    MA: {
+        name: 'Morocco',
+        sms: 0.105,
+    },
+    MC: {
+        name: 'Monaco',
+        sms: 0.1219,
+    },
+    MD: {
+        name: 'Moldova',
+        sms: 0.065,
+    },
+    ME: {
+        name: 'Montenegro',
+        sms: 0.07,
+    },
+    MG: {
+        name: 'Madagascar',
+        sms: 0.24,
+    },
+    MH: {
+        name: '',
+        sms: 0.032,
+    },
+    MK: {
+        name: 'Macedonia',
+        sms: 0.0096,
+    },
+    ML: {
+        name: 'Mali',
+        sms: 0.136,
+    },
+    MM: {
+        name: 'Myanmar',
+        sms: 0.3037,
+    },
+    MN: {
+        name: 'Mongolia',
+        sms: 0.183,
+    },
+    MO: {
+        name: 'Macao',
+        sms: 0.005,
+    },
+    MP: {
+        name: 'Northern Mariana Islands',
+        sms: 0.0669,
+    },
+    MQ: {
+        name: 'Martinique',
+        sms: 0.0455,
+    },
+    MR: {
+        name: 'Mauritania',
+        sms: 0.161,
+    },
+    MS: {
+        name: 'Montserrat',
+        sms: 0.0725,
+    },
+    MT: {
+        name: 'Malta',
+        sms: 0.0343,
+    },
+    MU: {
+        name: 'Mauritius',
+        sms: 0.1431,
+    },
+    MV: {
+        name: 'Maldives',
+        sms: 0.145,
+    },
+    MW: {
+        name: 'Malawi',
+        sms: 0.19,
+    },
+    MX: {
+        name: 'Mexico',
+        sms: 0.0021,
+    },
+    MY: {
+        name: 'Malaysia',
+        sms: 0.08,
+    },
+    MZ: {
+        name: 'Mozambique',
+        sms: 0.2266,
+    },
+    NA: {
+        name: 'Namibia',
+        sms: 0.0173,
+    },
+    NC: {
+        name: 'New Caledonia',
+        sms: 0.052,
+    },
+    NE: {
+        name: 'Niger',
+        sms: 0.143,
+    },
+    NG: {
+        name: 'Nigeria',
+        sms: 0.198,
+    },
+    NI: {
+        name: 'Nicaragua',
+        sms: 0.1033,
+    },
+    NL: {
+        name: 'Netherlands',
+        sms: 0.046,
+    },
+    NO: {
+        name: 'Norway',
+        sms: 0.03,
+    },
+    NP: {
+        name: 'Nepal',
+        sms: 0.1645,
+    },
+    NR: {
+        name: 'Nauru',
+        sms: null,
+    },
+    NU: {
+        name: 'Niue',
+        sms: null,
+    },
+    NZ: {
+        name: 'New Zealand',
+        sms: 0.0372,
+    },
+    OM: {
+        name: 'Oman',
+        sms: 0.0713,
+    },
+    PA: {
+        name: 'Panama',
+        sms: 0.07,
+    },
+    PE: {
+        name: 'Peru',
+        sms: 0.13,
+    },
+    PF: {
+        name: 'French Polynesia',
+        sms: 0.0518,
+    },
+    PG: {
+        name: 'Papua New Guinea',
+        sms: 0.15,
+    },
+    PH: {
+        name: 'Philippines',
+        sms: 0.1237,
+    },
+    PK: {
+        name: 'Pakistan',
+        sms: 0.3548,
+    },
+    PL: {
+        name: 'Poland',
+        sms: 0.011,
+    },
+    PM: {
+        name: 'Saint Pierre and Miquelon',
+        sms: 0.0905,
+    },
+    PR: {
+        name: 'Puerto Rico',
+        sms: 0.02,
+    },
+    PS: {
+        name: 'Palestine',
+        sms: 0.2541,
+    },
+    PT: {
+        name: 'Portugal',
+        sms: 0.009,
+    },
+    PW: {
+        name: 'Palau',
+        sms: null,
+    },
+    PY: {
+        name: 'Paraguay',
+        sms: 0.0236,
+    },
+    QA: {
+        name: 'Qatar',
+        sms: 0.1393,
+    },
+    RE: {
+        name: 'Reunion',
+        sms: 0.034,
+    },
+    RO: {
+        name: 'Romania',
+        sms: 0.0255,
+    },
+    RS: {
+        name: 'Serbia',
+        sms: 0.1935,
+    },
+    RU: {
+        name: 'Russia',
+        sms: 0.2028,
+    },
+    RW: {
+        name: 'Rwanda',
+        sms: 0.1177,
+    },
+    SA: {
+        name: 'Saudi Arabia',
+        sms: 0.0638,
+    },
+    SB: {
+        name: 'Solomon Islands',
+        sms: 0.0357,
+    },
+    SC: {
+        name: 'Seychelles',
+        sms: 0.0404,
+    },
+    SD: {
+        name: 'Sudan',
+        sms: 0.224,
+    },
+    SE: {
+        name: 'Sweden',
+        sms: 0.026,
+    },
+    SG: {
+        name: 'Singapore',
+        sms: 0.0256,
+    },
+    SI: {
+        name: 'Slovenia',
+        sms: 0.1,
+    },
+    SK: {
+        name: 'Slovakia',
+        sms: 0.0245,
+    },
+    SL: {
+        name: 'Sierra Leone',
+        sms: 0.2322,
+    },
+    SM: {
+        name: 'San Marino',
+        sms: null,
+    },
+    SN: {
+        name: 'Senegal',
+        sms: 0.1151,
+    },
+    SO: {
+        name: 'Somalia',
+        sms: 0.06,
+    },
+    SR: {
+        name: 'Suriname',
+        sms: 0.119,
+    },
+    SS: {
+        name: 'South Sudan',
+        sms: 0.15,
+    },
+    ST: {
+        name: 'Sao Tome and Principe',
+        sms: 0.0133,
+    },
+    SV: {
+        name: 'El Salvador',
+        sms: 0.06,
+    },
+    SX: {
+        name: 'Sint Maarten',
+        sms: 0.0623,
+    },
+    SY: {
+        name: 'Syria',
+        sms: 0.221,
+    },
+    SZ: {
+        name: 'Eswatini',
+        sms: 0.12,
+    },
+    TC: {
+        name: 'Turks and Caicos Islands',
+        sms: null,
+    },
+    TD: {
+        name: 'Chad',
+        sms: 0.1711,
+    },
+    TG: {
+        name: 'Togo',
+        sms: 0.1806,
+    },
+    TH: {
+        name: 'Thailand',
+        sms: 0.003,
+    },
+    TJ: {
+        name: 'Tajikistan',
+        sms: 0.2626,
+    },
+    TL: {
+        name: 'Timor-Leste',
+        sms: 0.0675,
+    },
+    TM: {
+        name: 'Turkmenistan',
+        sms: 0.1677,
+    },
+    TN: {
+        name: 'Tunisia',
+        sms: 0.225,
+    },
+    TO: {
+        name: 'Tonga',
+        sms: 0.0902,
+    },
+    TR: {
+        name: 'Turkey',
+        sms: 0.0008,
+    },
+    TT: {
+        name: 'Trinidad and Tobago',
+        sms: 0.1556,
+    },
+    TW: {
+        name: 'Taiwan',
+        sms: 0.0165,
+    },
+    TZ: {
+        name: 'Tanzania',
+        sms: 0.2753,
+    },
+    UA: {
+        name: 'Ukraine',
+        sms: 0.094,
+    },
+    UG: {
+        name: 'Uganda',
+        sms: 0.1707,
+    },
+    US: {
+        name: 'United States',
+        sms: 0.0043,
+    },
+    UY: {
+        name: 'Uruguay',
+        sms: 0.0174,
+    },
+    UZ: {
+        name: 'Uzbekistan',
+        sms: 0.292,
+    },
+    VC: {
+        name: 'Saint Vincent and The Grenadines',
+        sms: 0.1216,
+    },
+    VE: {
+        name: 'Venezuela',
+        sms: 0.0427,
+    },
+    VG: {
+        name: 'British Virgin Islands',
+        sms: 0.083,
+    },
+    VI: {
+        name: 'Virgin Island',
+        sms: 0.0017,
+    },
+    VN: {
+        name: 'Vietnam',
+        sms: 0.0885,
+    },
+    VU: {
+        name: 'Vanuatu',
+        sms: 0.1347,
+    },
+    WF: {
+        name: 'Wallis and Futuna',
+        sms: 0.09,
+    },
+    WS: {
+        name: 'Samoa',
+        sms: 0.2,
+    },
+    XK: {
+        name: 'Kosovo',
+        sms: 0.1394,
+    },
+    YE: {
+        name: 'Yemen',
+        sms: 0.1486,
+    },
+    YT: {
+        name: 'Mayotte',
+        sms: 0.06,
+    },
+    ZA: {
+        name: 'South Africa',
+        sms: 0.039,
+    },
+    ZM: {
+        name: 'Zambia',
+        sms: 0.21,
+    },
+    ZW: {
+        name: 'Zimbabwe',
+        sms: 0.13,
+    },
+};

src/backend/controllers/apps/AppController.js

@@ -56,6 +56,7 @@ export class AppController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const apps = await this.appDriver.select({
@@ -71,6 +72,7 @@ export class AppController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const name = req.query?.name;
@@ -185,6 +187,7 @@ export class AppController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const raw = req.params.name;

src/backend/controllers/auth/AuthController.ts

@@ -47,6 +47,7 @@ import {
 } from '../../services/auth/OTPUtil.js';
 import { sessionCookieFlags } from '../../util/cookieFlags.js';
 import { cleanEmail, isBlockedEmail } from '../../util/email.js';
+import { parsePhone } from '../../util/phone.js';
 import { generate_identifier } from '../../util/identifier.js';
 import { getTaskbarItems } from '../../util/taskbarItems.js';
 import {
@@ -57,6 +58,8 @@ import { PuterController } from '../types.js';
 
 const USERNAME_REGEX = /^\w{1,}$/;
 const USERNAME_MAX_LENGTH = 45;
+const FINGERPRINT_MAX_LENGTH = 128;
+const DFP_TELEMETRY_ID_MAX_LENGTH = 64;
 const RESERVED_USERNAMES = new Set([
     'admin',
     'administrator',
@@ -99,7 +102,16 @@ export class AuthController extends PuterController {
 
     @Post('/login', {
         captcha: true,
-        rateLimit: { scope: 'login', limit: 10, window: 15 * 60_000 },
+        // Two limits: per-fingerprint keeps users behind a shared IP
+        // (offices, campuses) from throttling each other, while the
+        // coarser per-IP backstop stops an attacker from minting fresh
+        // fingerprint buckets by rotating client-controlled headers
+        // (User-Agent etc.). Same pattern on the other unauthenticated
+        // credential endpoints below.
+        rateLimit: [
+            { scope: 'login', limit: 10, window: 15 * 60_000 },
+            { scope: 'login-ip', limit: 50, window: 15 * 60_000, key: 'ip' },
+        ],
     })
     async handleLogin(req: Request, res: Response): Promise<void> {
         const { username, email, password } = req.body;
@@ -206,11 +218,15 @@ export class AuthController extends PuterController {
 
     @Post('/login/otp', {
         captcha: true,
-        rateLimit: {
-            scope: 'login-otp',
-            limit: 15,
-            window: 30 * 60_000,
-        },
+        rateLimit: [
+            { scope: 'login-otp', limit: 15, window: 30 * 60_000 },
+            {
+                scope: 'login-otp-ip',
+                limit: 60,
+                window: 30 * 60_000,
+                key: 'ip',
+            },
+        ],
     })
     async handleLoginOtp(req: Request, res: Response): Promise<void> {
         const { token, code } = req.body;
@@ -266,11 +282,15 @@ export class AuthController extends PuterController {
 
     @Post('/login/recovery-code', {
         captcha: true,
-        rateLimit: {
-            scope: 'login-recovery',
-            limit: 10,
-            window: 60 * 60_000,
-        },
+        rateLimit: [
+            { scope: 'login-recovery', limit: 10, window: 60 * 60_000 },
+            {
+                scope: 'login-recovery-ip',
+                limit: 40,
+                window: 60 * 60_000,
+                key: 'ip',
+            },
+        ],
     })
     async handleLoginRecoveryCode(req: Request, res: Response): Promise<void> {
         const { token, code } = req.body;
@@ -339,7 +359,10 @@ export class AuthController extends PuterController {
 
     @Post('/signup', {
         captcha: true,
-        rateLimit: { scope: 'signup', limit: 10, window: 15 * 60_000 },
+        rateLimit: [
+            { scope: 'signup', limit: 10, window: 15 * 60_000 },
+            { scope: 'signup-ip', limit: 50, window: 15 * 60_000, key: 'ip' },
+        ],
     })
     async handleSignup(req: Request, res: Response): Promise<void> {
         const body = req.body ?? {};
@@ -355,6 +378,43 @@ export class AuthController extends PuterController {
             return;
         }
 
+        // Optional device signals (browser fingerprint hash, DFP telemetry
+        // id). Core only enforces shape and forwards the values verbatim —
+        // signup-abuse policy built on them lives in extensions. Checked
+        // before the reauth short-circuit so a malformed value is rejected
+        // on every /signup path.
+        if (body.fingerprint !== undefined && body.fingerprint !== null) {
+            if (typeof body.fingerprint !== 'string')
+                throw new HttpError(400, 'fingerprint must be a string.', {
+                    legacyCode: 'bad_request',
+                });
+            if (body.fingerprint.length > FINGERPRINT_MAX_LENGTH)
+                throw new HttpError(
+                    400,
+                    `fingerprint cannot be longer than ${FINGERPRINT_MAX_LENGTH} characters.`,
+                    { legacyCode: 'bad_request' },
+                );
+        }
+        if (
+            body.dfp_telemetry_id !== undefined &&
+            body.dfp_telemetry_id !== null
+        ) {
+            if (typeof body.dfp_telemetry_id !== 'string')
+                throw new HttpError(400, 'dfp_telemetry_id must be a string.', {
+                    legacyCode: 'bad_request',
+                });
+            if (body.dfp_telemetry_id.length > DFP_TELEMETRY_ID_MAX_LENGTH)
+                throw new HttpError(
+                    400,
+                    `dfp_telemetry_id cannot be longer than ${DFP_TELEMETRY_ID_MAX_LENGTH} characters.`,
+                    { legacyCode: 'bad_request' },
+                );
+        }
+        // Empty strings are treated as absent — a signal that wasn't
+        // collected, not a malformed request.
+        const fingerprint: string | null = body.fingerprint || null;
+        const dfp_telemetry_id: string | null = body.dfp_telemetry_id || null;
+
         // Temp-user reauth short-circuit: when an existing temp user is
         // forced through the reauth flow, the GUI re-submits /signup with
         // is_temp=true plus the server-signed reauth_token from the 401.
@@ -534,23 +594,45 @@ export class AuthController extends PuterController {
             allow: true,
             no_temp_user: false,
             requires_email_confirmation: false,
+            // Set by the abuse harness for low-reputation signups: the account is
+            // created + logged in but gated behind SMS phone verification (in
+            // addition to email confirmation) instead of being blocked.
+            requires_phone_verification: false,
+            // Same idea, one rung up the ladder: gate the account behind
+            // credit-card verification (a $0 auth handled by an extension).
+            requires_card_verification: false,
             message: null,
             code: null,
             user_agent: req?.headers?.['user-agent'] ?? null,
+            fingerprint,
+            dfp_telemetry_id,
+            // Populated by the abuse extension's v2 harness; persisted to the
+            // user row below so the signup-time reputation is referable later.
+            reputation: null as number | null,
+            // Stamped by the abuse harness for flagged signups — the id keying
+            // the `abuse:trail:<id>` decision trail (carrying both the live and
+            // shadow trails). Surfaced to a blocked user as the Request Code so
+            // the code they quote support leads straight to their trail.
+            trail_id: undefined as string | undefined,
         };
         try {
             await this.clients.event?.emitAndWait(
-                'puter.signup.validate' as never,
-                validateEvent as never,
+                'puter.signup.validate',
+                validateEvent,
                 {},
             );
         } catch (e) {
             console.warn('[signup] validate hook failed:', e);
         }
         if (!validateEvent.allow) {
+            // Pass the trail id back to a blocked user as the Request Code (when
+            // the harness stamped one), embedded in the message so the existing
+            // signup-block UI surfaces it without a GUI change.
+            const requestCode = validateEvent.trail_id;
             throw new HttpError(
                 403,
-                validateEvent.message ?? 'Signup blocked',
+                (validateEvent.message ?? 'Signup blocked') +
+                    (requestCode ? ` Request Code: ${requestCode}` : ''),
                 {
                     ...(validateEvent.code
                         ? { legacyCode: validateEvent.code as never }
@@ -573,6 +655,17 @@ export class AuthController extends PuterController {
         const force_email_confirmation = Boolean(
             validateEvent.requires_email_confirmation,
         );
+        const force_phone_verification =
+            Boolean(validateEvent.requires_phone_verification) ||
+            // Test/QA switch: force the SMS gate on every signup regardless of
+            // reputation (see config.always_require_phone_verification).
+            Boolean(this.config.always_require_phone_verification);
+        const force_card_verification = Boolean(
+            validateEvent.requires_card_verification ||
+            // Test/QA switch: force the card gate on every signup regardless of
+            // reputation (see config.always_require_card_verification).
+            this.config.always_require_card_verification,
+        );
 
         // Prepare shared fields
         const user_uuid = uuidv4();
@@ -597,11 +690,13 @@ export class AuthController extends PuterController {
                 email_confirm_code,
                 email_confirm_token,
                 email_confirmed: 0,
-                // Pseudo claims always require email confirmation — the
-                // validate hook can only tighten, not loosen, so `1`
-                // stays hardcoded here.
                 requires_email_confirmation: 1,
                 last_activity_ts: signupSqlTs,
+                ...(validateEvent.reputation != null
+                    ? { reputation: validateEvent.reputation }
+                    : {}),
+                requires_phone_verification: force_phone_verification ? 1 : 0,
+                requires_card_verification: force_card_verification ? 1 : 0,
             });
 
             // Move from temp group to regular user group
@@ -650,6 +745,8 @@ export class AuthController extends PuterController {
                     ip_fwd: proxyIpChain,
                     user_agent: req.headers?.['user-agent'],
                     origin: req.headers?.origin,
+                    fingerprint,
+                    dfp_telemetry_id,
                 },
                 signup_ip: clientIp,
                 signup_ip_forwarded: proxyIpChain,
@@ -658,6 +755,11 @@ export class AuthController extends PuterController {
                 signup_server: (this.config as { serverId?: string }).serverId,
                 referrer: req.body.referrer ?? null,
                 last_activity_ts: signupSqlTs,
+                reputation: validateEvent.reputation,
+                // Phone collected later in the verification dialog (null now).
+                phone: null,
+                requires_phone_verification: force_phone_verification,
+                requires_card_verification: force_card_verification,
             } as never);
 
             // Add to default group
@@ -728,6 +830,11 @@ export class AuthController extends PuterController {
                     user_uuid: user!.uuid,
                     email: user!.email,
                     username: user!.username,
+                    fingerprint,
+                    // Reflects the row that was actually created/claimed —
+                    // a pseudo-user claim ends up with credentials, so it
+                    // reports false here. Same signal completeLogin uses.
+                    is_temp: user!.password === null && user!.email === null,
                     ip:
                         (req?.headers?.['x-forwarded-for'] as
                             | string
@@ -928,6 +1035,507 @@ export class AuthController extends PuterController {
         res.json({ email_confirmed: true, original_client_socket_id });
     }
 
+    // -- Phone verification (SMS via Prelude) ------------------------
+
+    @Post('/send-confirm-phone', {
+        subdomain: ['api', ''],
+        requireUserActor: true,
+        allowUnconfirmed: true,
+        rateLimit: {
+            scope: 'send-confirm-phone',
+            limit: 10,
+            window: 60 * 60_000,
+            key: 'user',
+        },
+    })
+    async handleSendConfirmPhone(req: Request, res: Response): Promise<void> {
+        const user = await this.stores.user.getById(req.actor!.user.id!, {
+            force: true,
+        });
+        if (!user)
+            throw new HttpError(404, 'User not found.', {
+                legacyCode: 'user_not_found' as never,
+            });
+        if (user.suspended)
+            throw new HttpError(403, 'Account suspended.', {
+                legacyCode: 'account_suspended',
+            });
+        if (!this.clients.prelude?.isConfigured())
+            throw new HttpError(503, 'Phone verification is unavailable.', {
+                legacyCode: 'service_unavailable' as never,
+            });
+
+        // Parse to E.164 (Prelude's required form + the stored form) and the
+        // country, so we can apply the per-country cost cap.
+        const parsed = parsePhone(
+            req.body?.phone,
+            this.clients.prelude.defaultCountry,
+        );
+        if (!parsed)
+            throw new HttpError(400, 'Invalid phone number.', {
+                legacyCode: 'bad_request',
+            });
+
+        // Cost cap: skip countries with no SMS channel or rates above the cap
+        // (see PreludeClient / countries.ts). Avoids paying exorbitant per-SMS
+        // rates in low-revenue, high-fraud geographies.
+        if (!this.clients.prelude.isCountrySupported(parsed.country))
+            throw new HttpError(
+                400,
+                'Phone verification is not available for this country.',
+                { legacyCode: 'phone_country_not_supported' as never },
+            );
+
+        // Abuse caps live ENTIRELY in a listening abuse extension, consulted
+        // via `puter.phone-verification.check`. The backend ships no thresholds
+        // or detection of its own (so none of it is readable in the open-source
+        // repo): it forwards the user / number / ip, and the extension decides
+        // `allowed` plus an opaque `reason` (per-account + per-number send
+        // velocity, cross-account reuse, …). With no extension listening
+        // `allowed` stays true. Fail-open on a hook error — this is abuse/cost
+        // control, not a security boundary (the route rate limit and the country
+        // cost cap remain), so a flaky hook must not lock signups out.
+        const abuseCheck = {
+            user_id: user.id,
+            user_uid: user.uuid,
+            phone: parsed.e164,
+            device_fingerprint: req.deviceFingerprint ?? null,
+            allowed: true,
+            reason: null as string | null,
+        };
+        try {
+            await this.clients.event?.emitAndWait(
+                'puter.phone-verification.check',
+                abuseCheck,
+                {},
+            );
+        } catch (e) {
+            console.warn('[send-confirm-phone] abuse-check hook failed:', e);
+        }
+        // Forward the verdict verbatim: a generic 429 plus the opaque reason for
+        // the client to message on. The backend never interprets the reason —
+        // its meaning lives in the extension (which sets it) and the GUI (which
+        // displays it), so no abuse semantics leak into the OSS repo.
+        if (abuseCheck.allowed === false)
+            throw new HttpError(
+                429,
+                'Phone verification is unavailable for this number right now.',
+                {
+                    legacyCode: 'phone_verification_unavailable' as never,
+                    fields: abuseCheck.reason
+                        ? { reason: abuseCheck.reason }
+                        : {},
+                },
+            );
+
+        // Stage the parsed number as pending in KV (NOT on the user row) so a
+        // never-confirmed number is never written to the indexed `phone`
+        // column. /confirm-phone reads it back and persists it to the row only
+        // once Prelude confirms the code. ~1h TTL covers the code's lifetime.
+        // Stored before the send so we never dispatch an SMS we couldn't later
+        // confirm against.
+        const pendingPhoneKey = `phone-verify-pending:${user.id}`;
+        try {
+            await this.stores.kv.set({
+                key: pendingPhoneKey,
+                value: parsed.e164,
+                expireAt: Math.floor(Date.now() / 1000) + 60 * 60,
+            });
+        } catch (e) {
+            console.warn('[send-confirm-phone] pending-store failed:', e);
+            throw new HttpError(503, 'Could not start phone verification.', {
+                legacyCode: 'service_unavailable' as never,
+            });
+        }
+
+        const ip = req.ip || req.socket?.remoteAddress || undefined;
+        const userAgent =
+            typeof req.headers['user-agent'] === 'string'
+                ? req.headers['user-agent']
+                : undefined;
+        try {
+            const result = await this.clients.prelude.createVerification(
+                parsed.e164,
+                {
+                    ip,
+                    device_id: req.deviceFingerprint ?? undefined,
+                    user_agent: userAgent,
+                },
+            );
+            // Prelude rejected the attempt as abusive — surface as rate-limit.
+            if (
+                result.status === 'blocked' ||
+                result.status === 'shadow_blocked'
+            ) {
+                throw new HttpError(
+                    429,
+                    'Phone verification is temporarily unavailable for this number.',
+                    { legacyCode: 'too_many_requests' as never },
+                );
+            }
+        } catch (e) {
+            if (e instanceof HttpError) throw e;
+            console.warn('[send-confirm-phone] createVerification failed:', e);
+            throw new HttpError(502, 'Could not send verification code.', {
+                legacyCode: 'upstream_error' as never,
+            });
+        }
+
+        // Tell the abuse extension a code was actually sent, so it can bump its
+        // send-velocity counters (per number + per account). Fire-and-forget;
+        // the backend keeps no send counts of its own. Only reached after a
+        // successful send, so an upstream error never burns quota.
+        try {
+            this.clients.event?.emit(
+                'puter.phone-verification.sent' as never,
+                {
+                    user_id: user.id,
+                    user_uid: user.uuid,
+                    phone: parsed.e164,
+                    device_fingerprint: req.deviceFingerprint ?? null,
+                } as never,
+                {},
+            );
+        } catch {
+            // ignore — best-effort velocity signal
+        }
+        res.json({});
+    }
+
+    @Post('/confirm-phone', {
+        subdomain: ['api', ''],
+        requireUserActor: true,
+        allowUnconfirmed: true,
+        rateLimit: {
+            scope: 'confirm-phone',
+            limit: 10,
+            window: 10 * 60_000,
+            key: 'user',
+        },
+    })
+    async handleConfirmPhone(req: Request, res: Response): Promise<void> {
+        const { code, original_client_socket_id } = req.body ?? {};
+        if (!code)
+            throw new HttpError(400, 'Missing `code`.', {
+                legacyCode: 'bad_request',
+            });
+
+        const user = await this.stores.user.getById(req.actor!.user.id!, {
+            force: true,
+        });
+        if (!user)
+            throw new HttpError(404, 'User not found.', {
+                legacyCode: 'not_found',
+            });
+        if (!user.requires_phone_verification) {
+            res.json({ phone_verified: true, original_client_socket_id });
+            return;
+        }
+        // The number being verified is the one staged at send time (KV
+        // pending), not the user row — we don't persist an unverified number.
+        // Fall back to a row value for accounts that already have one on file
+        // (legacy / a number persisted by a prior verified flow).
+        const pendingPhoneKey = `phone-verify-pending:${user.id}`;
+        let pendingPhone: string | null = null;
+        try {
+            const { res: staged } = await this.stores.kv.get({
+                key: pendingPhoneKey,
+            });
+            if (typeof staged === 'string' && staged) pendingPhone = staged;
+        } catch (e) {
+            console.warn('[confirm-phone] pending read failed:', e);
+        }
+        if (!pendingPhone) pendingPhone = user.phone ?? null;
+        if (!pendingPhone)
+            throw new HttpError(
+                400,
+                'No phone number on file. Request a code first.',
+                { legacyCode: 'bad_request' },
+            );
+        if (!this.clients.prelude?.isConfigured())
+            throw new HttpError(503, 'Phone verification is unavailable.', {
+                legacyCode: 'service_unavailable' as never,
+            });
+
+        let status;
+        try {
+            ({ status } = await this.clients.prelude.checkVerification(
+                pendingPhone,
+                String(code),
+            ));
+        } catch (e) {
+            console.warn('[confirm-phone] checkVerification failed:', e);
+            throw new HttpError(502, 'Could not verify code.', {
+                legacyCode: 'upstream_error' as never,
+            });
+        }
+
+        if (status !== 'success') {
+            res.json({ phone_verified: false, original_client_socket_id });
+            return;
+        }
+
+        // Verified — persist the number now (and only now) and clear the gate.
+        await this.stores.user.update(user.id, {
+            requires_phone_verification: 0,
+            phone: pendingPhone,
+        });
+
+        try {
+            this.clients.event?.emit(
+                'user.phone-verified' as never,
+                {
+                    user_id: user.id,
+                    user_uid: user.uuid,
+                    phone: pendingPhone,
+                } as never,
+                {},
+            );
+        } catch {
+            // ignore — event is a side-channel signal, not load-bearing
+        }
+        // Notify other tabs/devices for this user so they refresh + drop the gate.
+        try {
+            await this.services.socket?.send(
+                { room: user.id },
+                'user.phone_verified',
+                { original_client_socket_id },
+            );
+        } catch {
+            // ignore — best-effort
+        }
+
+        res.json({ phone_verified: true, original_client_socket_id });
+    }
+
+    // -- Card verification ($0 auth via a payments extension) --------
+
+    /**
+     * Start card verification for the calling user. Pure mechanism: the
+     * endpoint emits `puter.card-verification.setup` and a payments
+     * extension fills in the client credentials — the OSS backend holds
+     * no provider knowledge or config. Phone verification (when required)
+     * must be completed first; the ordering is enforced here so a client
+     * can't skip the cheaper gate.
+     */
+    @Post('/card-verification/setup', {
+        subdomain: ['api', ''],
+        requireUserActor: true,
+        allowUnconfirmed: true,
+        rateLimit: {
+            scope: 'card-verification-setup',
+            limit: 5,
+            window: 60 * 60_000,
+            key: 'user',
+        },
+    })
+    async handleCardVerificationSetup(
+        req: Request,
+        res: Response,
+    ): Promise<void> {
+        const user = await this.stores.user.getById(req.actor!.user.id!, {
+            force: true,
+        });
+        if (!user)
+            throw new HttpError(404, 'User not found.', {
+                legacyCode: 'user_not_found' as never,
+            });
+        if (user.suspended)
+            throw new HttpError(403, 'Account suspended.', {
+                legacyCode: 'account_suspended',
+            });
+        if (!user.requires_card_verification) {
+            res.json({ card_verified: true });
+            return;
+        }
+        if (user.requires_phone_verification)
+            throw new HttpError(
+                409,
+                'Phone verification must be completed first.',
+                { legacyCode: 'conflict' },
+            );
+
+        // `enabled` stays null when no extension is listening; an installed
+        // extension always sets it (true/false) before doing any work.
+        const setupEvent = {
+            user_id: user.id,
+            user_uid: user.uuid,
+            ip: (req.ip || req.socket?.remoteAddress || null) as string | null,
+            device_fingerprint: req.deviceFingerprint ?? null,
+            enabled: null as boolean | null,
+            allowed: true,
+            reason: null as string | null,
+            client_secret: null as string | null,
+            publishable_key: null as string | null,
+        };
+        try {
+            await this.clients.event?.emitAndWait(
+                'puter.card-verification.setup',
+                setupEvent,
+                {},
+            );
+        } catch (e) {
+            console.warn('[card-verification/setup] setup hook failed:', e);
+        }
+
+        // Abuse veto (e.g. per-device setup-velocity cap): the extension refused
+        // before any Stripe work. Forward the opaque reason verbatim as a 429,
+        // same as the phone gate — the backend never interprets it.
+        if (setupEvent.allowed === false)
+            throw new HttpError(
+                429,
+                'Card verification is unavailable right now.',
+                {
+                    legacyCode: 'too_many_requests' as never,
+                    fields: setupEvent.reason
+                        ? { reason: setupEvent.reason }
+                        : {},
+                },
+            );
+
+        // Kill switch: the extension reports the feature disabled — unstick
+        // any user still carrying the flag instead of dead-ending them.
+        if (setupEvent.enabled === false) {
+            await this.stores.user.update(user.id, {
+                requires_card_verification: 0,
+            });
+            res.json({ card_verified: true, disabled: true });
+            return;
+        }
+        if (!setupEvent.client_secret || !setupEvent.publishable_key)
+            throw new HttpError(503, 'Card verification is not available.', {
+                legacyCode: 'service_unavailable' as never,
+            });
+
+        res.json({
+            client_secret: setupEvent.client_secret,
+            publishable_key: setupEvent.publishable_key,
+        });
+    }
+
+    /**
+     * Complete card verification. The client confirms the setup intent with
+     * the payment provider directly, then posts the resulting id here; the
+     * payments extension checks it (and applies its own abuse limits) via
+     * `puter.card-verification.confirm`. On success the gate clears exactly
+     * like `/confirm-phone` clears the phone gate.
+     */
+    @Post('/card-verification/confirm', {
+        subdomain: ['api', ''],
+        requireUserActor: true,
+        allowUnconfirmed: true,
+        rateLimit: {
+            scope: 'card-verification-confirm',
+            limit: 10,
+            window: 10 * 60_000,
+            key: 'user',
+        },
+    })
+    async handleCardVerificationConfirm(
+        req: Request,
+        res: Response,
+    ): Promise<void> {
+        const { setup_intent_id, original_client_socket_id } = req.body ?? {};
+        if (
+            typeof setup_intent_id !== 'string' ||
+            setup_intent_id.length === 0 ||
+            setup_intent_id.length > 255
+        )
+            throw new HttpError(400, 'Invalid `setup_intent_id`.', {
+                legacyCode: 'bad_request',
+            });
+
+        const user = await this.stores.user.getById(req.actor!.user.id!, {
+            force: true,
+        });
+        if (!user)
+            throw new HttpError(404, 'User not found.', {
+                legacyCode: 'not_found',
+            });
+        if (!user.requires_card_verification) {
+            res.json({ card_verified: true });
+            return;
+        }
+        if (user.requires_phone_verification)
+            throw new HttpError(
+                409,
+                'Phone verification must be completed first.',
+                { legacyCode: 'conflict' },
+            );
+
+        const confirmEvent = {
+            user_id: user.id,
+            user_uid: user.uuid,
+            setup_intent_id,
+            enabled: null as boolean | null,
+            verified: false,
+            reason: null as string | null,
+            fingerprint: null as string | null,
+            funding: null as string | null,
+            country: null as string | null,
+        };
+        try {
+            await this.clients.event?.emitAndWait(
+                'puter.card-verification.confirm',
+                confirmEvent,
+                {},
+            );
+        } catch (e) {
+            console.warn('[card-verification/confirm] confirm hook failed:', e);
+        }
+
+        // Kill switch — same semantics as /card-verification/setup.
+        if (confirmEvent.enabled === false) {
+            await this.stores.user.update(user.id, {
+                requires_card_verification: 0,
+            });
+            res.json({ card_verified: true, disabled: true });
+            return;
+        }
+        // No extension listening — nothing could have verified anything.
+        if (confirmEvent.enabled === null)
+            throw new HttpError(503, 'Card verification is not available.', {
+                legacyCode: 'service_unavailable' as never,
+            });
+
+        if (confirmEvent.verified !== true) {
+            res.json({ card_verified: false, reason: confirmEvent.reason });
+            return;
+        }
+
+        await this.stores.user.update(user.id, {
+            requires_card_verification: 0,
+        });
+
+        try {
+            this.clients.event?.emit(
+                'user.card-verified' as never,
+                {
+                    user_id: user.id,
+                    user_uid: user.uuid,
+                    fingerprint: confirmEvent.fingerprint,
+                    funding: confirmEvent.funding,
+                    country: confirmEvent.country,
+                } as never,
+                {},
+            );
+        } catch {
+            // ignore — event is a side-channel signal, not load-bearing
+        }
+        // Notify other tabs/devices for this user so they refresh + drop the gate.
+        try {
+            await this.services.socket?.send(
+                { room: user.id },
+                'user.card_verified',
+                { original_client_socket_id },
+            );
+        } catch {
+            // ignore — best-effort
+        }
+
+        res.json({ card_verified: true });
+    }
+
     // -- Password recovery -------------------------------------------
 
     @Post('/send-pass-recovery-email', {
@@ -1129,6 +1737,12 @@ export class AuthController extends PuterController {
         }
         await this.stores.user.invalidateById(user.id);
 
+        // A password reset is the "I think someone else has access" flow —
+        // evict every interactive session so a hijacked one doesn't survive.
+        await this.services.auth.revokeInteractiveSessionsForUserId(
+            user.id as number,
+        );
+
         res.send('Password successfully updated.');
     }
 
@@ -1165,6 +1779,10 @@ export class AuthController extends PuterController {
             change_email_confirm_token: null,
         });
 
+        // Sign out every other web session (cascading to their derived
+        // rows); only the session that changed the password survives.
+        await this.services.auth.revokeAllSessions(req.actor!);
+
         if (this.clients.email && user.email) {
             try {
                 await this.clients.email.send(
@@ -2140,12 +2758,15 @@ export class AuthController extends PuterController {
     // -- Access tokens -----------------------------------------------
 
     async handleMigrateToken(req: Request, res: Response): Promise<void> {
-        // 1. Origin lock. Reject anything that isn't same-origin to
-        // `config.origin` or in the explicit per-deployment allowlist.
-        // No Origin header → reject (this endpoint is browser-only by
-        // design; server-side callers should re-auth properly).
+        // 1. Browser-only gate. No Origin header → reject (server-side
+        // callers should re-auth properly). The exchange itself is open
+        // to any browser origin: puter.js apps live on arbitrary
+        // third-party domains, the v1 bearer token is the credential,
+        // and cookies are never read here — so a hostile page learns
+        // nothing it doesn't already hold. Origin *trust* only decides
+        // whether the `puter_token_v2` companion cookie is set below.
         const reqOrigin = req.headers.origin;
-        if (!reqOrigin || !this.#isMigrateTokenOriginAllowed(reqOrigin)) {
+        if (!reqOrigin) {
             throw new HttpError(403, 'Origin not allowed', {
                 legacyCode: 'forbidden',
             });
@@ -2175,12 +2796,19 @@ export class AuthController extends PuterController {
                     : undefined,
         });
         // `puter_token_v2` is the cookie companion to v2 app tokens —
-        // the app runs in-browser (we already gated on Origin above) so
-        // the GUI's cookie-only middleware can authenticate subsequent
-        // calls from the same iframe without the client having to
-        // forward Authorization headers. Access tokens are programmatic
-        // (no browser cookie surface) so we deliberately skip them here.
-        if (result.kind === 'app') {
+        // it lets the GUI's cookie-only middleware authenticate
+        // subsequent same-origin calls without the client forwarding
+        // Authorization headers. Issuing it is gated on the trusted
+        // origin allowlist: an arbitrary attacker page must not be able
+        // to plant a session cookie on the GUI origin (login CSRF).
+        // Untrusted origins still get the token in the JSON body — the
+        // cookie is only consumed same-origin (see authProbe) so they
+        // lose nothing. Access tokens are programmatic (no browser
+        // cookie surface) so we deliberately skip them here.
+        if (
+            result.kind === 'app' &&
+            this.#isMigrateTokenOriginAllowed(reqOrigin)
+        ) {
             res.cookie('puter_token_v2', result.token, {
                 ...sessionCookieFlags(this.config),
                 httpOnly: true,
@@ -2212,13 +2840,25 @@ export class AuthController extends PuterController {
         requireAuth: true,
     })
     async handleCreateAccessToken(req: Request, res: Response): Promise<void> {
-        const { permissions, expiresIn } = req.body;
+        const { permissions, expiresIn, label } = req.body;
         if (!Array.isArray(permissions) || permissions.length === 0) {
             throw new HttpError(400, 'Missing or empty `permissions` array', {
                 legacyCode: 'bad_request',
             });
         }
 
+        // Optional user-facing name for the manage-sessions UI. Trim and clamp
+        // to the same 64-char limit the rename endpoint enforces.
+        let normalizedLabel: string | null = null;
+        if (label !== undefined && label !== null) {
+            if (typeof label !== 'string') {
+                throw new HttpError(400, '`label` must be a string', {
+                    legacyCode: 'bad_request',
+                });
+            }
+            normalizedLabel = label.trim().slice(0, 64) || null;
+        }
+
         // Normalize specs: string → [string], [string] → [string, {}], [string, extra] → as-is
         const normalized = permissions.map((spec) => {
             if (typeof spec === 'string') return [spec];
@@ -2233,7 +2873,10 @@ export class AuthController extends PuterController {
         const token = await this.services.auth.createAccessToken(
             req.actor!,
             normalized as never,
-            expiresIn ? { expiresIn } : {},
+            {
+                ...(expiresIn ? { expiresIn } : {}),
+                ...(normalizedLabel ? { label: normalizedLabel } : {}),
+            },
         );
         res.json({ token });
     }
@@ -2470,6 +3113,9 @@ export class AuthController extends PuterController {
             });
 
         await this.stores.group.addUsers(uid, users);
+        // New members inherit the group's permissions immediately, not
+        // after the permission-cache TTL.
+        await this.services.permission.bumpPermissionCacheForUsernames(users);
         res.json({});
     }
 
@@ -2499,6 +3145,9 @@ export class AuthController extends PuterController {
             });
 
         await this.stores.group.removeUsers(uid, users);
+        // Removed members must lose the group's permissions immediately,
+        // not after the permission-cache TTL.
+        await this.services.permission.bumpPermissionCacheForUsernames(users);
         res.json({});
     }
 
@@ -2805,6 +3454,22 @@ export class AuthController extends PuterController {
     // -- Private helpers ----------------------------------------------
 
     async #cascadeDeleteUser(userId: number): Promise<void> {
+        // Capture the identifiers downstream teardown needs before the row is
+        // gone — the marketplace extension cancels the user's Stripe
+        // subscriptions off `user.delete`, keyed by uuid / customer id.
+        let userUuid: string | undefined;
+        let stripeCustomerId: string | null = null;
+        try {
+            const rows = (await this.clients.db.read(
+                'SELECT `uuid`, `stripe_customer_id` FROM `user` WHERE `id` = ?',
+                [userId],
+            )) as Array<{ uuid?: string; stripe_customer_id?: string | null }>;
+            userUuid = rows[0]?.uuid;
+            stripeCustomerId = rows[0]?.stripe_customer_id ?? null;
+        } catch (e) {
+            console.warn('[cascade-delete-user] identifier lookup failed:', e);
+        }
+
         try {
             await this.services.fs.removeAllForUser(userId);
         } catch (e) {
@@ -2822,6 +3487,24 @@ export class AuthController extends PuterController {
             userId,
         ]);
         await this.stores.user.invalidateById(userId);
+
+        // Fire-and-forget: let listeners purge external state tied to the
+        // account (Stripe subscriptions are cancelled immediately, without
+        // proration). Emitted after the row delete — listeners key off the
+        // payload, not the DB row.
+        try {
+            this.clients.event?.emit(
+                'user.delete',
+                {
+                    user_id: userId,
+                    user_uuid: userUuid,
+                    stripe_customer_id: stripeCustomerId,
+                },
+                {},
+            );
+        } catch {
+            // ignore — event emission shouldn't block deletion
+        }
     }
 
     async #generateRandomUsername(): Promise<string> {
@@ -2981,6 +3664,9 @@ export class AuthController extends PuterController {
             password?: string | null;
             email_confirmed?: number | boolean;
             requires_email_confirmation?: number | boolean;
+            phone?: string | null;
+            requires_phone_verification?: number | boolean;
+            requires_card_verification?: number | boolean;
         },
     ): Promise<void> {
         const meta = {
@@ -3029,6 +3715,9 @@ export class AuthController extends PuterController {
                 email: user.email,
                 email_confirmed: user.email_confirmed,
                 requires_email_confirmation: user.requires_email_confirmation,
+                phone: user.phone,
+                requires_phone_verification: user.requires_phone_verification,
+                requires_card_verification: user.requires_card_verification,
                 is_temp: user.password === null && user.email === null,
                 taskbar_items,
             },

src/backend/controllers/desktop/DesktopController.js

@@ -51,6 +51,7 @@ export class DesktopController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const { url, color, fit } = req.body ?? {};
@@ -105,6 +106,7 @@ export class DesktopController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const { items } = req.body ?? {};
@@ -130,6 +132,7 @@ export class DesktopController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const { item_uid, item_path, layout } = req.body ?? {};
@@ -156,6 +159,7 @@ export class DesktopController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             },
             async (req, res) => {
                 const { item_uid, item_path, sort_by, sort_order } =

src/backend/controllers/drivers/DriverController.test.ts

@@ -21,7 +21,8 @@ import { Readable } from 'node:stream';
 import type { Request, RequestHandler, Response } from 'express';
 import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
 import { v4 as uuidv4 } from 'uuid';
-import type { Actor } from '../../core/actor.js';
+import { actorUid, type Actor } from '../../core/actor.js';
+import type { DriverMethodLifecycleEvent } from '../../clients/event/types.js';
 import { runWithContext } from '../../core/context.js';
 import { PuterServer } from '../../server.js';
 import { setupTestServer } from '../../testUtil.js';
@@ -460,6 +461,129 @@ describe('DriverController.#handleCall (via captured router)', () => {
     });
 });
 
+describe('DriverController driver-method lifecycle events', () => {
+    let routes: Captured;
+    beforeAll(() => {
+        routes = captureRoutes(controller);
+    });
+
+    const grantKvPerm = async (actor: Actor) => {
+        await server.stores.permission.setFlatUserPerm(
+            actor.user!.id!,
+            'service:puter-kvstore:ii:puter-kvstore',
+            {
+                permission: 'service:puter-kvstore:ii:puter-kvstore',
+                deleted: false,
+                issuer_user_id: actor.user!.id!,
+            } as never,
+        );
+    };
+
+    it('emits before then after around a successful driver method', async () => {
+        const actor = await makeUserActor();
+        await grantKvPerm(actor);
+
+        // Correlate on the actor: each test creates a fresh user, so this
+        // isolates our call from any other traffic on the shared event bus.
+        const who = actorUid(actor);
+        const before: DriverMethodLifecycleEvent[] = [];
+        const after: DriverMethodLifecycleEvent[] = [];
+        server.clients.event.on(
+            'driver.puter-kvstore.set.before',
+            (_k, data) => {
+                if (data.actorUid === who) before.push(data);
+            },
+        );
+        server.clients.event.on(
+            'driver.puter-kvstore.set.after',
+            (_k, data) => {
+                if (data.actorUid === who) after.push(data);
+            },
+        );
+
+        const key = `lc-${uuidv4()}`;
+        const req = makeReq(
+            {
+                interface: 'puter-kvstore',
+                method: 'set',
+                args: { key, value: 'v' },
+            },
+            actor,
+        );
+        await runWithContext({ actor }, () =>
+            routes['POST /call'](
+                req,
+                makeRes() as unknown as Response,
+                () => {},
+            ),
+        );
+
+        expect(before).toHaveLength(1);
+        expect(before[0]).toMatchObject({
+            phase: 'before',
+            iface: 'puter-kvstore',
+            method: 'set',
+            actor,
+            actorUid: who,
+        });
+        expect(after).toHaveLength(1);
+        expect(after[0].phase).toBe('after');
+        expect(after[0].args).toMatchObject({ key, value: 'v' });
+        expect(typeof after[0].durationMs).toBe('number');
+    });
+
+    it('emits reject and answers 403 when a before listener vetoes', async () => {
+        const actor = await makeUserActor();
+        await grantKvPerm(actor);
+
+        // Scope the veto to a sentinel key so it doesn't affect other calls
+        // sharing this server's event bus.
+        const key = `veto-${uuidv4()}`;
+        const reject: DriverMethodLifecycleEvent[] = [];
+        server.clients.event.on(
+            'driver.puter-kvstore.set.before',
+            (_k, data) => {
+                if ((data.args as { key?: string })?.key === key) {
+                    data.allow = false;
+                    data.rejectReason = 'blocked in test';
+                }
+            },
+        );
+        server.clients.event.on(
+            'driver.puter-kvstore.set.reject',
+            (_k, data) => {
+                if (data.rejectReason === 'blocked in test') reject.push(data);
+            },
+        );
+
+        const req = makeReq(
+            {
+                interface: 'puter-kvstore',
+                method: 'set',
+                args: { key, value: 'v' },
+            },
+            actor,
+        );
+        await expect(
+            runWithContext({ actor }, () =>
+                routes['POST /call'](
+                    req,
+                    makeRes() as unknown as Response,
+                    () => {},
+                ),
+            ),
+        ).rejects.toMatchObject({ statusCode: 403 });
+
+        expect(reject).toHaveLength(1);
+        expect(reject[0]).toMatchObject({
+            phase: 'reject',
+            method: 'set',
+            args: { key, value: 'v' },
+            rejectReason: 'blocked in test',
+        });
+    });
+});
+
 describe('DriverController.#handleListInterfaces', () => {
     let routes: Captured;
     beforeAll(() => {

src/backend/controllers/drivers/DriverController.ts

@@ -18,6 +18,7 @@
  */
 
 import type { Request, Response } from 'express';
+import { actorUid } from '../../core/actor.js';
 import { Context } from '../../core/context.js';
 import { Controller } from '../../core/http/decorators.js';
 import { HttpError, isHttpError } from '../../core/http/HttpError.js';
@@ -34,6 +35,7 @@ import {
     resolveDriverMethodRateLimit,
 } from '../../drivers/meta.js';
 import type { PermissionService } from '../../services/permission/PermissionService.js';
+import { PermissionUtil } from '../../services/permission/permissionUtil.js';
 import type { WithLifecycle } from '../../types';
 import { PuterController } from '../types.js';
 
@@ -221,7 +223,16 @@ export class DriverController extends PuterController {
                 | PermissionService
                 | undefined;
             if (permService) {
-                const permKey = `service:${resolvedDriverName}:ii:${ifaceName}`;
+                // Build via PermissionUtil.join so any `:` in a driver or
+                // interface name is escaped — raw interpolation would let a
+                // crafted name shift permission-segment boundaries and match
+                // a broader/narrower parent than intended in the scan logic.
+                const permKey = PermissionUtil.join(
+                    'service',
+                    String(resolvedDriverName),
+                    'ii',
+                    ifaceName,
+                );
                 const hasPermission = await permService.check(
                     req.actor,
                     permKey,
@@ -324,14 +335,91 @@ export class DriverController extends PuterController {
         // a stale value from a prior call.
         Context.set('driverName', requestedDriver);
 
+        // Per-method lifecycle events, scoped to `driver.<iface>.<method>`.
+        // Subscribers can listen on `driver.*`, `driver.<iface>.*`, or the
+        // exact key. `before` is emitted via `emitAndWait` so a listener may
+        // veto the call by setting `allow = false` (emits `reject`, throws
+        // 403); otherwise `after`/`error` carry the result/error + duration.
+        const actor = req.actor ? actorUid(req.actor) : undefined;
+        const resolved = String(resolvedDriverName);
+        const beforeEvent = {
+            phase: 'before' as const,
+            iface: ifaceName,
+            method,
+            driver: resolved,
+            actor: req.actor,
+            actorUid: actor,
+            args,
+            allow: true as boolean,
+            rejectReason: undefined as string | undefined,
+        };
+        await this.clients.event?.emitAndWait(
+            `driver.${ifaceName}.${method}.before`,
+            beforeEvent,
+            {},
+        );
+        if (beforeEvent.allow === false) {
+            this.clients.event?.emit(
+                `driver.${ifaceName}.${method}.reject`,
+                {
+                    phase: 'reject',
+                    iface: ifaceName,
+                    method,
+                    driver: resolved,
+                    actor: req.actor,
+                    actorUid: actor,
+                    args,
+                    rejectReason: beforeEvent.rejectReason,
+                },
+                {},
+            );
+            throw new HttpError(
+                403,
+                beforeEvent.rejectReason ??
+                    `Blocked by policy: ${ifaceName}:${method}`,
+                { legacyCode: 'forbidden' },
+            );
+        }
+
         // Drivers read actor/context via the Context API — no drilled args.
+        const startedAt = Date.now();
         let result;
         try {
             // eslint-disable-next-line @typescript-eslint/no-explicit-any
             result = await (fn as (...x: unknown[]) => any).call(driver, args);
         } catch (e) {
+            this.clients.event?.emit(
+                `driver.${ifaceName}.${method}.error`,
+                {
+                    phase: 'error',
+                    iface: ifaceName,
+                    method,
+                    driver: resolved,
+                    actor: req.actor,
+                    actorUid: actor,
+                    args,
+                    error: e,
+                    durationMs: Date.now() - startedAt,
+                },
+                {},
+            );
             throw translateProviderError(e);
         }
+        this.clients.event?.emit(
+            `driver.${ifaceName}.${method}.after`,
+            {
+                phase: 'after',
+                iface: ifaceName,
+                method,
+                driver: resolved,
+                actor: req.actor,
+                actorUid: actor,
+                args,
+                result,
+                durationMs: Date.now() - startedAt,
+            },
+            {},
+        );
 
         if (isDriverStreamResult(result)) {
             res.setHeader('Content-Type', result.content_type);

src/backend/controllers/fs/FSController.test.ts

@@ -18,6 +18,7 @@
  */
 
 import type { Request, Response } from 'express';
+import type { Readable } from 'node:stream';
 import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
 import { v4 as uuidv4 } from 'uuid';
 import type { Actor } from '../../core/actor.js';
@@ -124,6 +125,14 @@ const makeRes = () => {
 const withActor = async <T>(actor: Actor, fn: () => Promise<T>): Promise<T> =>
     runWithContext({ actor }, fn);
 
+const streamToString = async (stream: Readable): Promise<string> => {
+    const chunks: Buffer[] = [];
+    for await (const chunk of stream) {
+        chunks.push(Buffer.from(chunk as Buffer));
+    }
+    return Buffer.concat(chunks).toString('utf8');
+};
+
 // ── /startBatchWrite ────────────────────────────────────────────────
 
 describe('FSController.startBatchWrites', () => {
@@ -434,6 +443,65 @@ describe('FSController.completeBatchWrites', () => {
         expect(finalized?.wasOverwrite).toBe(true);
     });
 
+    // Regression: a signed (direct-to-S3) upload could declare a tiny size
+    // and PUT far more — the presigned URL doesn't bound the body. The
+    // completion path must reconcile the recorded size against the object's
+    // true size, or storage accounting is permanently understated and the
+    // quota is bypassable.
+    it('reconciles the recorded size to the real uploaded bytes (ignores under-declared size)', async () => {
+        const { actor, userId } = await makeUser();
+        const username = actor.user!.username!;
+        const target = `/${username}/Documents/under-declared.bin`;
+
+        // Declare 1 byte.
+        const start = makeRes();
+        await withActor(actor, () =>
+            controller.startBatchWrites(
+                makeReq<SignedWriteRequest[]>({
+                    body: [{ fileMetadata: { path: target, size: 1 } }],
+                    actor,
+                }),
+                start.res,
+            ),
+        );
+        const [started] = start.captured.body as SignedWriteResponse[];
+
+        // Actually upload 4096 bytes to the session's object key (simulating
+        // a client that PUTs more than it declared via the signed URL).
+        const realBytes = Buffer.alloc(4096, 0x41);
+        await server.stores.s3Object.uploadFromServer(
+            {
+                bucket: started!.bucket,
+                objectKey: started!.objectKey,
+                contentType: 'application/octet-stream',
+                body: realBytes,
+                contentLength: realBytes.byteLength,
+            },
+            started!.bucketRegion,
+        );
+
+        const complete = makeRes();
+        await withActor(actor, () =>
+            controller.completeBatchWrites(
+                makeReq<CompleteWriteRequest[]>({
+                    body: [{ uploadId: started!.sessionId }],
+                    actor,
+                }),
+                complete.res,
+            ),
+        );
+
+        const entry = await server.stores.fsEntry.getEntryByPath(target);
+        expect(entry).not.toBeNull();
+        // Recorded size is the true 4096 bytes, not the declared 1.
+        expect(entry?.size).toBe(4096);
+
+        // And the user's accounted usage reflects the real bytes.
+        const allowance =
+            await server.stores.fsEntry.getUserStorageAllowance(userId);
+        expect(allowance.curr).toBeGreaterThanOrEqual(4096);
+    });
+
     it('emits updated events with GUI metadata when overwriting via batch completion', async () => {
         const { actor, userId } = await makeUser();
         const username = actor.user!.username!;
@@ -490,9 +558,7 @@ describe('FSController.completeBatchWrites', () => {
             .body as SignedWriteResponse[];
 
         const emitSpy = vi.spyOn(server.clients.event, 'emit');
-        let updatedCall:
-            | (typeof emitSpy.mock.calls)[number]
-            | undefined;
+        let updatedCall: (typeof emitSpy.mock.calls)[number] | undefined;
         try {
             await withActor(actor, () =>
                 controller.completeBatchWrites(
@@ -667,9 +733,9 @@ describe('FSController.statEntry', () => {
             ...makeReq({ body: { path: '/x' }, actor }),
             actor: undefined,
         } as unknown as Request;
-        await expect(
-            controller.statEntry(req, res),
-        ).rejects.toMatchObject({ statusCode: 401 });
+        await expect(controller.statEntry(req, res)).rejects.toMatchObject({
+            statusCode: 401,
+        });
     });
 });
 
@@ -839,9 +905,7 @@ describe('FSController.searchEntries', () => {
             ).toBe(true);
         }
         expect(
-            results.some(
-                (r) => r.path === `/${username}/Documents/${needle}`,
-            ),
+            results.some((r) => r.path === `/${username}/Documents/${needle}`),
         ).toBe(false);
     });
 
@@ -1259,10 +1323,6 @@ describe('FSController.copyEntry', () => {
 // ── /read (readEntry, full read) ────────────────────────────────────
 
 describe('FSController.readEntry (file streaming)', () => {
-    // makeRes here adds a real Writable surface so that
-    // `pipeline(download.body, res)` inside the controller can pipe
-    // the in-memory S3 stream into the test's response and capture
-    // bytes for assertions.
     const makeStreamingRes = () => {
         const captured = {
             statusCode: 200,
@@ -1270,7 +1330,8 @@ describe('FSController.readEntry (file streaming)', () => {
             bodyChunks: [] as Buffer[],
         };
         // eslint-disable-next-line @typescript-eslint/no-require-imports
-        const { Writable } = require('node:stream') as typeof import('node:stream');
+        const { Writable } =
+            require('node:stream') as typeof import('node:stream');
         const writable = new Writable({
             write(chunk: Buffer, _enc, cb) {
                 captured.bodyChunks.push(chunk);
@@ -1330,7 +1391,9 @@ describe('FSController.readEntry (file streaming)', () => {
         // Pipeline awaits the stream-end on success.
         expect(captured.statusCode).toBe(200);
         expect(captured.headers['Content-Type']).toMatch(/text\/plain/);
-        expect(captured.headers['Content-Length']).toBe(String(body.byteLength));
+        expect(captured.headers['Content-Length']).toBe(
+            String(body.byteLength),
+        );
         expect(captured.headers['Content-Disposition']).toMatch(
             /inline; filename=/,
         );
@@ -1774,3 +1837,158 @@ describe('FSController.mkshortcutEntry', () => {
         expect(body.isShortcut).toBe(true);
     });
 });
+
+describe('FSController metadata.objectKey injection', () => {
+    const writeFile = async (
+        actor: Actor,
+        path: string,
+        content: string,
+        metadata?: Record<string, unknown>,
+    ) => {
+        await withActor(actor, () =>
+            controller.write(
+                makeReq({
+                    body: {
+                        fileMetadata: {
+                            path,
+                            size: Buffer.byteLength(content),
+                            contentType: 'text/plain',
+                            overwrite: true,
+                            ...(metadata ? { metadata } : {}),
+                        },
+                        fileContent: content,
+                        encoding: 'utf8',
+                    },
+                    actor,
+                }) as unknown as Request<
+                    Record<string, never>,
+                    null,
+                    import('./requestTypes.js').WriteRequest
+                >,
+                makeRes().res,
+            ),
+        );
+        const entry = await server.stores.fsEntry.getEntryByPath(path, {
+            skipCache: true,
+        });
+        if (!entry) throw new Error(`entry not found after write: ${path}`);
+        return entry;
+    };
+
+    it("does not stream another user's file when a client injects metadata.objectKey on write", async () => {
+        const victim = await makeUser();
+        const attacker = await makeUser();
+        const victimSecret = 'VICTIM-TOP-SECRET-PAYLOAD';
+        const attackerDecoy = 'attacker-own-decoy-bytes';
+
+        const victimEntry = await writeFile(
+            victim.actor,
+            `/${victim.actor.user!.username}/Documents/secret.txt`,
+            victimSecret,
+        );
+
+        const victimRead = await server.services.fs.readContent(victimEntry);
+        expect(await streamToString(victimRead.body)).toBe(victimSecret);
+
+        const attackerEntry = await writeFile(
+            attacker.actor,
+            `/${attacker.actor.user!.username}/Documents/loot.txt`,
+            attackerDecoy,
+            { objectKey: victimEntry.uuid },
+        );
+
+        const persisted = attackerEntry.metadata
+            ? (JSON.parse(attackerEntry.metadata) as Record<string, unknown>)
+            : {};
+        expect(persisted.objectKey).toBeUndefined();
+
+        const attackerRead =
+            await server.services.fs.readContent(attackerEntry);
+        const got = await streamToString(attackerRead.body);
+        expect(got).toBe(attackerDecoy);
+        expect(got).not.toBe(victimSecret);
+    });
+
+    it('read path ignores a divergent metadata.objectKey on an already-poisoned row', async () => {
+        const victim = await makeUser();
+        const attacker = await makeUser();
+        const victimSecret = 'VICTIM-SECRET-FOR-POISON-TEST';
+        const attackerDecoy = 'attacker-decoy-for-poison-test';
+
+        const victimEntry = await writeFile(
+            victim.actor,
+            `/${victim.actor.user!.username}/Documents/secret2.txt`,
+            victimSecret,
+        );
+        const attackerEntry = await writeFile(
+            attacker.actor,
+            `/${attacker.actor.user!.username}/Documents/loot2.txt`,
+            attackerDecoy,
+        );
+
+        await server.stores.fsEntry.updateEntry(attackerEntry.uuid, {
+            metadata: JSON.stringify({ objectKey: victimEntry.uuid }),
+        });
+        const poisoned = await server.stores.fsEntry.getEntryByPath(
+            attackerEntry.path,
+            { skipCache: true },
+        );
+        if (!poisoned) throw new Error('poisoned entry not found');
+        expect(
+            (JSON.parse(poisoned.metadata!) as { objectKey: string }).objectKey,
+        ).toBe(victimEntry.uuid);
+
+        const read = await server.services.fs.readContent(poisoned);
+        expect(await streamToString(read.body)).toBe(attackerDecoy);
+    });
+
+    it('scrubs objectKey from move newMetadata while preserving legit trash metadata', async () => {
+        const victim = await makeUser();
+        const attacker = await makeUser();
+        const victimSecret = 'VICTIM-SECRET-FOR-MOVE-TEST';
+        const attackerDecoy = 'attacker-decoy-for-move-test';
+        const username = attacker.actor.user!.username!;
+
+        const victimEntry = await writeFile(
+            victim.actor,
+            `/${victim.actor.user!.username}/Documents/secret3.txt`,
+            victimSecret,
+        );
+        const attackerEntry = await writeFile(
+            attacker.actor,
+            `/${username}/Documents/loot3.txt`,
+            attackerDecoy,
+        );
+        const documents = await server.stores.fsEntry.getEntryByPath(
+            `/${username}/Documents`,
+            { skipCache: true },
+        );
+        if (!documents) throw new Error('Documents dir not found');
+
+        const moved = await withActor(attacker.actor, () =>
+            server.services.fs.move(attacker.userId, {
+                source: attackerEntry,
+                destinationParent: documents,
+                newName: 'loot3-moved.txt',
+                newMetadata: {
+                    original_path: `/${username}/Documents/loot3.txt`,
+                    trashed_ts: 1700000000,
+                    objectKey: victimEntry.uuid,
+                },
+            }),
+        );
+
+        const persisted = JSON.parse(moved.metadata!) as Record<
+            string,
+            unknown
+        >;
+        expect(persisted.objectKey).toBeUndefined();
+        expect(persisted.original_path).toBe(
+            `/${username}/Documents/loot3.txt`,
+        );
+        expect(persisted.trashed_ts).toBe(1700000000);
+
+        const read = await server.services.fs.readContent(moved);
+        expect(await streamToString(read.body)).toBe(attackerDecoy);
+    });
+});

src/backend/controllers/fs/FSController.ts

@@ -37,6 +37,7 @@ import {
     runWithConcurrencyLimit,
     runWithConcurrencyLimitSettled,
 } from '../../util/concurrency.js';
+import { applyInlineContentSecurity } from '../../util/inlineContentSecurity.js';
 import { PuterController } from '../types.js';
 import { FS_COSTS } from './costs.js';
 import type {
@@ -964,8 +965,13 @@ export class FSController extends PuterController {
             range,
         });
 
-        if (download.contentType)
+        if (download.contentType) {
             res.setHeader('Content-Type', download.contentType);
+            // Stored type is uploader-controlled and served inline on the
+            // api origin — sandbox active-document types (HTML/SVG/XML) so
+            // embedded scripts can't run here. Inert types are untouched.
+            applyInlineContentSecurity(res, download.contentType);
+        }
         if (download.contentLength !== null)
             res.setHeader('Content-Length', String(download.contentLength));
         if (download.contentRange)

src/backend/controllers/fs/LegacyFSController.ts

@@ -29,6 +29,7 @@ import type { PuterRouter } from '../../core/http/PuterRouter.js';
 import type { ACLService } from '../../services/acl/ACLService.js';
 import type { SignedFile } from '../../util/fileSigning.js';
 import { verifySignature } from '../../util/fileSigning.js';
+import { applyInlineContentSecurity } from '../../util/inlineContentSecurity.js';
 import { PuterController } from '../types.js';
 import { FS_COSTS } from './costs.js';
 import {
@@ -45,14 +46,6 @@ import {
 } from './legacyFsHelpers.js';
 import { RouteOptions } from '../../core/http/index.js';
 
-/**
- * Legacy FS routes, implemented as thin shims over `FSService`.
- *
- * Each shim parses the request shape (FSNodeParam-style `{ path, uid, id }`
- * or `{ parent, name }`), invokes the service method, and returns the
- * snake_case response clients expect.
- */
-
 type RouterCache = Map<string, RequestHandler | null>;
 
 const additionalRoutePaths: Record<string, string> = {};
@@ -146,6 +139,11 @@ export class LegacyFSController extends PuterController {
             {
                 subdomain: ['api', ''],
                 requireUserActor: true,
+                // The user's own credential may download their files; apps and
+                // scoped tokens stay blocked. antiCsrf still protects the
+                // cookie-authed GUI path — a bearer-token PAT is exempt below
+                // (CSRF can't forge a header-credentialed request).
+                allowFullAccessToken: true,
                 requireVerified: true,
                 antiCsrf: true,
             },
@@ -1352,8 +1350,15 @@ export class LegacyFSController extends PuterController {
             query.download === '1' ||
             query.download === true;
 
-        if (download.contentType)
+        if (download.contentType) {
             res.setHeader('Content-Type', download.contentType);
+            // Uploader-controlled type served inline on the file origin —
+            // sandbox active-document types so an uploaded HTML/SVG can't
+            // execute scripts here. Mirrors FSController /fs/read.
+            if (!wantsAttachment) {
+                applyInlineContentSecurity(res, download.contentType);
+            }
+        }
         if (download.contentLength !== null)
             res.setHeader('Content-Length', String(download.contentLength));
         if (download.contentRange)
@@ -1618,17 +1623,6 @@ export class LegacyFSController extends PuterController {
         download.body.pipe(res);
     };
 
-    // Helpers for writeFile
-    // -- GUI event emission -------------------------------------------
-    //
-    // Fire-and-forget `outer.gui.item.*` events so SocketService,
-    // BroadcastService, WorkerDriver (hot-reload), and cache-invalidation
-    // listeners pick up mutations made through the legacy (bare-path) routes.
-    // FSController (v2-native /fs/* routes) emits these from its own handlers;
-    // LegacyFSController delegates to the same FSService but needs its
-    // own emissions because the service layer deliberately doesn't emit GUI
-    // events (that's a controller concern).
-
     async #emitGuiEvent(
         eventName:
             | 'outer.gui.item.added'

src/backend/controllers/hosting/HostingController.js

@@ -44,6 +44,7 @@ export class HostingController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
                 requireVerified: true,
             },
             async (req, res) => {

src/backend/controllers/notification/NotificationController.ts

@@ -38,7 +38,11 @@ export class NotificationController extends PuterController {
      * POST /notif/mark-ack — user dismissed a notification.
      * Sets `acknowledged` timestamp; pushes ack event to sockets.
      */
-    @Post('/mark-ack', { subdomain: 'api', requireUserActor: true })
+    @Post('/mark-ack', {
+        subdomain: 'api',
+        requireUserActor: true,
+        allowFullAccessToken: true,
+    })
     async markAck(req: Request, res: Response): Promise<void> {
         const uid = req.body?.uid;
         if (typeof uid !== 'string' || uid.length === 0) {
@@ -77,7 +81,11 @@ export class NotificationController extends PuterController {
      * POST /notif/mark-read — user saw a notification.
      * Sets `shown` timestamp; pushes ack event to sockets.
      */
-    @Post('/mark-read', { subdomain: 'api', requireUserActor: true })
+    @Post('/mark-read', {
+        subdomain: 'api',
+        requireUserActor: true,
+        allowFullAccessToken: true,
+    })
     async markRead(req: Request, res: Response): Promise<void> {
         const uid = req.body?.uid;
         if (typeof uid !== 'string' || uid.length === 0) {

src/backend/controllers/peer/PeerController.ts

@@ -17,6 +17,7 @@
  * along with this program.  If not, see <https://www.gnu.org/licenses/>.
  */
 
+import { createHmac, randomBytes, timingSafeEqual } from 'node:crypto';
 import type { Request, Response } from 'express';
 import type { Actor } from '../../core/actor.js';
 import { HttpError } from '../../core/http/HttpError.js';
@@ -24,6 +25,21 @@ import type { PuterRouter } from '../../core/http/PuterRouter.js';
 import { PuterController } from '../types.js';
 import { PEER_COSTS } from './costs.js';
 
+/**
+ * Constant-time secret comparison for the internal-auth header. HMAC both
+ * sides under a random per-process key to a fixed 32-byte digest first:
+ * this avoids leaking length via an early-return and sidesteps
+ * `timingSafeEqual`'s equal-length requirement for arbitrary-length
+ * inputs. The key need not persist — it only has to be unknown to the
+ * attacker for the duration of the comparison.
+ */
+const COMPARE_KEY = randomBytes(32);
+const secretsEqual = (a: string, b: string): boolean => {
+    const ha = createHmac('sha256', COMPARE_KEY).update(a).digest();
+    const hb = createHmac('sha256', COMPARE_KEY).update(b).digest();
+    return timingSafeEqual(ha, hb);
+};
+
 /**
  * Encode a UUID (or `app-<uuid>` UID) as base64url with no padding.
  * Strips an `app-` prefix and dashes, then reinterprets the hex bytes.
@@ -170,7 +186,11 @@ export class PeerController extends PuterController {
         }
         const expectedSecret = cfg.internal_auth_secret;
         const header = req.headers['x-puter-internal-auth'];
-        if (!expectedSecret || header !== expectedSecret) {
+        if (
+            !expectedSecret ||
+            typeof header !== 'string' ||
+            !secretsEqual(header, expectedSecret)
+        ) {
             throw new HttpError(403, 'Forbidden', { legacyCode: 'forbidden' });
         }
 

src/backend/controllers/puterai/PuterAIController.test.ts

@@ -188,10 +188,11 @@ describe('PuterAIController.registerRoutes', () => {
             ]),
         );
 
-        // The four upstream-proxy routes are user-only: app actors must
-        // call puter-chat-completion directly. The `requireUserActor`
-        // route option wires up the gate that enforces this, so each
-        // proxy route gets the assertion (not just chat/completions).
+        // The four upstream-proxy routes reject third-party apps (they must
+        // call puter-chat-completion directly), but admit the user's own
+        // full-access personal access token: `requireUserActor` keeps apps out
+        // and `allowFullAccessToken` opens the gate to a full-access PAT. Each
+        // proxy route carries both flags.
         const userOnlyPaths = [
             '/puterai/openai/v1/chat/completions',
             '/puterai/openai/v1/completions',
@@ -203,6 +204,7 @@ describe('PuterAIController.registerRoutes', () => {
             expect(route?.opts).toEqual({
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
             });
         }
         const modelsRoute = calls.find(
@@ -1504,6 +1506,212 @@ describe('PuterAIController.anthropicMessages streaming + helpers', () => {
     });
 });
 
+// ── Inline compaction ───────────────────────────────────────────────
+
+describe('PuterAIController inline compaction', () => {
+    // Extract the single `event: compaction\ndata: {...}` SSE frame.
+    const compactionFrame = (out: string): string | null => {
+        const m = out.match(/event: compaction\ndata: [^\n]*\n\n/);
+        return m ? m[0] : null;
+    };
+
+    it('forwards the compaction opt-in to the driver (/responses)', async () => {
+        const spy = stubChatComplete({
+            message: { role: 'assistant', content: 'ok' },
+            finish_reason: 'stop',
+        });
+        await controller.openaiResponses(
+            makeReq({
+                body: { model: 'gpt-test', input: 'hi', compaction: true },
+                actor: makeUserActor(),
+            }),
+            makeRes().res,
+        );
+        expect(spy.mock.calls[0]![0].compaction).toBe(true);
+    });
+
+    it('round-trips a compaction `input` item into a messages compaction item', async () => {
+        const spy = stubChatComplete({
+            message: { role: 'assistant', content: 'ack' },
+            finish_reason: 'stop',
+        });
+        await controller.openaiResponses(
+            makeReq({
+                body: {
+                    model: 'gpt-test',
+                    input: [
+                        {
+                            type: 'compaction',
+                            id: 'cmpct_1',
+                            encrypted_content: 'ENC',
+                        },
+                    ],
+                },
+                actor: makeUserActor(),
+            }),
+            makeRes().res,
+        );
+        expect(spy.mock.calls[0]![0].messages).toContainEqual({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+    });
+
+    it('emits a canonical compaction SSE event and output item (/responses stream)', async () => {
+        stubChatComplete({
+            dataType: 'stream',
+            content_type: 'application/x-ndjson',
+            stream: ndjsonStreamFrom([
+                { type: 'text', text: 'hi' },
+                { type: 'compaction', id: 'cmpct_9', encrypted_content: 'ENC9' },
+                { type: 'usage', usage: { prompt_tokens: 1, completion_tokens: 1 } },
+            ]),
+        });
+        const { res, captured } = makeRes();
+        await controller.openaiResponses(
+            makeReq({
+                body: {
+                    model: 'gpt-test',
+                    input: 'hi',
+                    stream: true,
+                    compaction: true,
+                },
+                actor: makeUserActor(),
+            }),
+            res,
+        );
+        await new Promise<void>((resolve) => setImmediate(resolve));
+
+        const out = captured.written.join('');
+        expect(compactionFrame(out)).toBe(
+            'event: compaction\ndata: {"type":"compaction","id":"cmpct_9","encrypted_content":"ENC9"}\n\n',
+        );
+        // Native shape also lands in the final response.completed output[].
+        const completed = out
+            .split('event: response.completed\n')[1]
+            ?.split('\n\n')[0];
+        expect(completed).toContain('"type":"compaction"');
+        expect(completed).toContain('"encrypted_content":"ENC9"');
+    });
+
+    it('emits the compaction item in non-streaming /responses output', async () => {
+        stubChatComplete({
+            message: { role: 'assistant', content: 'done' },
+            finish_reason: 'stop',
+            compaction: { id: 'cmpct_n', encrypted_content: 'ENCN' },
+        });
+        const { res, captured } = makeRes();
+        await controller.openaiResponses(
+            makeReq({
+                body: { model: 'gpt-test', input: 'hi', compaction: true },
+                actor: makeUserActor(),
+            }),
+            res,
+        );
+        const body = captured.body as { output: Array<Record<string, unknown>> };
+        expect(body.output).toContainEqual(
+            expect.objectContaining({
+                type: 'compaction',
+                encrypted_content: 'ENCN',
+            }),
+        );
+    });
+
+    it('emits an identical canonical compaction SSE event on the Anthropic surface', async () => {
+        // /responses frame
+        stubChatComplete({
+            dataType: 'stream',
+            content_type: 'application/x-ndjson',
+            stream: ndjsonStreamFrom([
+                { type: 'compaction', id: 'cmpct_x', encrypted_content: 'ENCX' },
+            ]),
+        });
+        const r1 = makeRes();
+        await controller.openaiResponses(
+            makeReq({
+                body: { model: 'gpt-test', input: 'hi', stream: true },
+                actor: makeUserActor(),
+            }),
+            r1.res,
+        );
+        await new Promise<void>((resolve) => setImmediate(resolve));
+
+        // /anthropic/v1/messages frame
+        stubChatComplete({
+            dataType: 'stream',
+            content_type: 'application/x-ndjson',
+            stream: ndjsonStreamFrom([
+                { type: 'compaction', id: 'cmpct_x', encrypted_content: 'ENCX' },
+            ]),
+        });
+        const r2 = makeRes();
+        await controller.anthropicMessages(
+            makeReq({
+                body: {
+                    model: 'claude-test',
+                    messages: [{ role: 'user', content: 'hi' }],
+                    stream: true,
+                },
+                actor: makeUserActor(),
+            }),
+            r2.res,
+        );
+        await new Promise<void>((resolve) => setImmediate(resolve));
+
+        const a = compactionFrame(r1.captured.written.join(''));
+        const b = compactionFrame(r2.captured.written.join(''));
+        expect(a).not.toBeNull();
+        expect(a).toBe(b); // byte-identical streaming shape across providers
+    });
+
+    it('renders a native compaction content block in non-streaming /messages', async () => {
+        stubChatComplete({
+            message: { role: 'assistant', content: 'done' },
+            finish_reason: 'stop',
+            compaction: { id: 'cmpct_m', encrypted_content: 'ENCM' },
+        });
+        const { res, captured } = makeRes();
+        await controller.anthropicMessages(
+            makeReq({
+                body: {
+                    model: 'claude-test',
+                    messages: [{ role: 'user', content: 'hi' }],
+                },
+                actor: makeUserActor(),
+            }),
+            res,
+        );
+        const body = captured.body as { content: Array<Record<string, unknown>> };
+        expect(body.content).toContainEqual({
+            type: 'compaction',
+            id: 'cmpct_m',
+            encrypted_content: 'ENCM',
+        });
+    });
+
+    it('does not emit compaction frames for a normal stream (regression)', async () => {
+        stubChatComplete({
+            dataType: 'stream',
+            content_type: 'application/x-ndjson',
+            stream: ndjsonStreamFrom([
+                { type: 'text', text: 'hello' },
+                { type: 'usage', usage: { prompt_tokens: 1, completion_tokens: 1 } },
+            ]),
+        });
+        const { res, captured } = makeRes();
+        await controller.openaiResponses(
+            makeReq({
+                body: { model: 'gpt-test', input: 'hi', stream: true },
+                actor: makeUserActor(),
+            }),
+            res,
+        );
+        await new Promise<void>((resolve) => setImmediate(resolve));
+        expect(captured.written.join('')).not.toContain('event: compaction');
+    });
+});
+
 // ── Model details listing ───────────────────────────────────────────
 
 describe('PuterAIController model details', () => {

src/backend/controllers/puterai/PuterAIController.ts

@@ -52,6 +52,12 @@ export class PuterAIController extends PuterController {
         const apiAuthOpts = {
             subdomain: 'api',
             requireUserActor: true,
+            // These are the user's own programmatic AI surface (OpenAI/Anthropic
+            // wire). `requireUserActor` here exists to keep third-party apps
+            // out, not to block the user's own credential — so admit full-access
+            // personal access tokens (CLI/MCP/scripts). Apps and scoped tokens
+            // stay blocked.
+            allowFullAccessToken: true,
         } as RouteOptions;
         const publicOpts = { subdomain: 'api', requireAuth: false } as const;
 
@@ -581,6 +587,15 @@ export class PuterAIController extends PuterController {
                 ? { metadata: body.metadata as Record<string, string> }
                 : {}),
             ...(body.conversation ? { conversation: body.conversation } : {}),
+            ...(body.context_management !== undefined
+                ? { context_management: body.context_management }
+                : {}),
+            ...(body.compaction !== undefined
+                ? {
+                      compaction:
+                          body.compaction as ICompleteArguments['compaction'],
+                  }
+                : {}),
             ...(body.previous_response_id
                 ? { previous_response_id: String(body.previous_response_id) }
                 : {}),
@@ -744,6 +759,30 @@ export class PuterAIController extends PuterController {
                             output_index: outputIndex,
                             item,
                         });
+                    } else if (ev.type === 'compaction') {
+                        // Native shape in the final `output[]`, plus the
+                        // canonical SSE event shared with the Anthropic surface.
+                        const item = {
+                            type: 'compaction',
+                            ...(ev.id !== undefined ? { id: ev.id } : {}),
+                            encrypted_content: ev.encrypted_content,
+                        };
+                        const outputIndex = output.length;
+                        output.push(item);
+                        sendEvent({
+                            type: 'response.output_item.added',
+                            output_index: outputIndex,
+                            item,
+                        });
+                        sendEvent({
+                            type: 'response.output_item.done',
+                            output_index: outputIndex,
+                            item,
+                        });
+                        writeCompactionEvent(res, {
+                            id: ev.id,
+                            encrypted_content: ev.encrypted_content,
+                        });
                     } else if (ev.type === 'usage') {
                         usage = buildResponsesUsage(
                             ev.usage as Record<string, unknown>,
@@ -863,6 +902,15 @@ export class PuterAIController extends PuterController {
             ...(body.max_tokens !== undefined
                 ? { max_tokens: Number(body.max_tokens) }
                 : {}),
+            ...(body.context_management !== undefined
+                ? { context_management: body.context_management }
+                : {}),
+            ...(body.compaction !== undefined
+                ? {
+                      compaction:
+                          body.compaction as ICompleteArguments['compaction'],
+                  }
+                : {}),
             ...(body.provider
                 ? { provider: toStringOrEmpty(body.provider) }
                 : { provider: DEFAULTS.anthropic }),
@@ -962,6 +1010,14 @@ export class PuterAIController extends PuterController {
                             },
                         });
                         closeBlock();
+                    } else if (ev.type === 'compaction') {
+                        // Close any open content block, then emit the canonical
+                        // compaction SSE event — byte-identical to /responses.
+                        closeBlock();
+                        writeCompactionEvent(res, {
+                            id: ev.id,
+                            encrypted_content: ev.encrypted_content,
+                        });
                     } else if (ev.type === 'usage') {
                         usage = ev.usage as Record<string, unknown>;
                     }
@@ -1020,6 +1076,18 @@ export class PuterAIController extends PuterController {
         if (textContent)
             contentBlocks.push({ type: 'text', text: textContent });
         contentBlocks.push(...toolUseBlocks);
+        // Native Anthropic-shaped compaction block (non-streaming bodies stay
+        // provider-native, unlike the unified streaming event).
+        const compaction = (
+            messageResult as { compaction?: Record<string, unknown> }
+        ).compaction;
+        if (compaction) {
+            contentBlocks.push({
+                type: 'compaction',
+                ...(compaction.id !== undefined ? { id: compaction.id } : {}),
+                encrypted_content: compaction.encrypted_content,
+            });
+        }
         if (contentBlocks.length === 0)
             contentBlocks.push({ type: 'text', text: '' });
 
@@ -1076,6 +1144,25 @@ const setSseHeaders = (res: Response): void => {
     res.setHeader('Connection', 'keep-alive');
 };
 
+/**
+ * Inline-compaction is emitted in a single canonical SSE shape that is
+ * byte-identical across the `/responses` and `/anthropic/v1/messages` streaming
+ * surfaces, so a streaming client parses compaction the same way regardless of
+ * which upstream served the request. (Non-streaming bodies stay provider-native.)
+ */
+const writeCompactionEvent = (
+    res: Response,
+    compaction: { id?: unknown; encrypted_content?: unknown },
+): void => {
+    const payload = {
+        type: 'compaction',
+        ...(compaction.id !== undefined ? { id: compaction.id } : {}),
+        encrypted_content: compaction.encrypted_content,
+    };
+    res.write(`event: compaction\n`);
+    res.write(`data: ${JSON.stringify(payload)}\n\n`);
+};
+
 /**
  * The chat driver returns either a stream-result envelope or a plain
  * message result. Proxy routes invoked with `stream: true` expect the
@@ -1319,6 +1406,17 @@ const responseInputToMessages = (input: unknown): unknown[] => {
         if (!item || typeof item !== 'object') continue;
         const it = item as Record<string, unknown>;
 
+        if (it.type === 'compaction') {
+            // Round-tripped compaction artifact: preserve it as a bare item so
+            // `normalize_single_message` wraps it into an internal compaction
+            // content block (providers map it back to their native input shape).
+            messages.push({
+                type: 'compaction',
+                ...(it.id !== undefined ? { id: it.id } : {}),
+                encrypted_content: it.encrypted_content,
+            });
+            continue;
+        }
         if (it.type === 'function_call_output') {
             messages.push({
                 role: 'tool',
@@ -1410,6 +1508,16 @@ const responseOutputFromResult = (
         });
     }
 
+    const compaction = (result as { compaction?: Record<string, unknown> })
+        .compaction;
+    if (compaction) {
+        output.push({
+            id: (compaction.id as string | undefined) || generateId('cmpct'),
+            type: 'compaction',
+            encrypted_content: compaction.encrypted_content,
+        });
+    }
+
     return output;
 };
 

src/backend/controllers/system/SystemController.js

@@ -77,6 +77,7 @@ export class SystemController extends PuterController {
             {
                 subdomain: 'api',
                 requireUserActor: true,
+                allowFullAccessToken: true,
                 rateLimit: {
                     scope: 'contact-us',
                     limit: 10,

src/backend/controllers/webdav/WebDAVController.test.ts

@@ -4,6 +4,7 @@ import type { Request, Response } from 'express';
 import { Readable } from 'node:stream';
 import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest';
 import { v4 as uuidv4 } from 'uuid';
+import { hash as bcryptHash } from 'bcrypt';
 import { PuterRouter } from '../../core/http/PuterRouter.js';
 import { PuterServer } from '../../server.js';
 import { setupTestServer } from '../../testUtil.js';
@@ -228,6 +229,112 @@ describe('WebDAVController', () => {
         });
     });
 
+    describe('pending-verification gate', () => {
+        // WebDAV must enforce the same gate every other authenticated route
+        // gets from requireVerifiedAccount — it dispatches off a single use()
+        // with no route options, so the middleware is never wired in and it
+        // has to call assertVerifiedAccount itself. Without it, an account
+        // still pending email/phone/card verification could read/write its
+        // whole filesystem over the `dav` subdomain.
+        const gatedActor = (flags: Record<string, unknown>) => ({
+            user: {
+                id: 1,
+                uuid: 'gated-uuid',
+                username: 'gated',
+                ...flags,
+            },
+        });
+
+        it('rejects a session actor pending phone verification with 403', async () => {
+            const { res, captured } = makeRes();
+            await dispatchMiddleware(
+                makeReq({
+                    method: 'PROPFIND',
+                    actor: gatedActor({ requires_phone_verification: true }),
+                }),
+                res,
+                noop,
+            );
+            expect(captured.statusCode).toBe(403);
+        });
+
+        it('rejects a session actor pending card verification with 403', async () => {
+            const { res, captured } = makeRes();
+            await dispatchMiddleware(
+                makeReq({
+                    method: 'GET',
+                    actor: gatedActor({ requires_card_verification: true }),
+                }),
+                res,
+                noop,
+            );
+            expect(captured.statusCode).toBe(403);
+        });
+
+        it('rejects a session actor with an unconfirmed email with 403', async () => {
+            const { res, captured } = makeRes();
+            await dispatchMiddleware(
+                makeReq({
+                    method: 'PROPFIND',
+                    actor: gatedActor({
+                        requires_email_confirmation: true,
+                        email_confirmed: false,
+                    }),
+                }),
+                res,
+                noop,
+            );
+            expect(captured.statusCode).toBe(403);
+        });
+
+        it('enforces the gate on the Basic-auth path (flags carried onto the built actor)', async () => {
+            const username = `webdav-gated-${Math.random()
+                .toString(36)
+                .slice(2, 10)}`;
+            const created = await server.stores.user.create({
+                username,
+                uuid: uuidv4(),
+                password: await bcryptHash('correct-horse', 4),
+                email: `${username}@test.local`,
+                free_storage: 100 * 1024 * 1024,
+                requires_email_confirmation: false,
+            });
+            await server.stores.user.update(created.id, {
+                requires_phone_verification: 1,
+            });
+
+            const { res, captured } = makeRes();
+            await dispatchMiddleware(
+                makeReq({
+                    method: 'PROPFIND',
+                    headers: {
+                        authorization: basicAuth(username, 'correct-horse'),
+                    },
+                }),
+                res,
+                noop,
+            );
+            expect(captured.statusCode).toBe(403);
+        });
+
+        it('lets a fully-verified session actor through the gate', async () => {
+            const { res, captured } = makeRes();
+            await dispatchMiddleware(
+                makeReq({
+                    method: 'OPTIONS',
+                    actor: gatedActor({
+                        requires_phone_verification: false,
+                        requires_card_verification: false,
+                        requires_email_confirmation: false,
+                    }),
+                }),
+                res,
+                noop,
+            );
+            expect(captured.statusCode).toBe(200);
+        });
+    });
+
     describe('unsupported methods', () => {
         it('returns 405 for unknown HTTP methods', async () => {
             const { res, captured } = makeRes();

src/backend/controllers/webdav/WebDAVController.ts

@@ -23,6 +23,7 @@ import { posix as pathPosix } from 'node:path';
 import { EventMap } from '../../clients/event/types.js';
 import type { Actor } from '../../core/actor.js';
 import { HttpError } from '../../core/http/HttpError.js';
+import { assertVerifiedAccount } from '../../core/http/middleware/gates.js';
 import type { PuterRouter } from '../../core/http/PuterRouter.js';
 import { verify as verifyOtp } from '../../services/auth/OTPUtil.js';
 import { expandTildePath } from '../../services/fs/resolveNode.js';
@@ -89,6 +90,16 @@ export class WebDAVController extends PuterController {
         const actor = await this.#resolveActor(req, res);
         if (!actor) return; // 401 already sent
 
+        // Apply the same pending-verification gate every other authenticated
+        // route gets from `requireVerifiedAccount`. WebDAV dispatches every
+        // method off a single `router.use` with no route options, so that
+        // middleware is never inserted into its chain — without this call an
+        // account still pending email / phone / card verification could read,
+        // write, and delete its entire filesystem over the `dav` subdomain,
+        // bypassing the gate. Throws a 403 HttpError, surfaced by the catch in
+        // registerRoutes.
+        assertVerifiedAccount(actor.user);
+
         // Expand `~`/`~/...` against the authenticated actor's username.
         // WebDAV doesn't standardize `~`, but some clients do — and the
         // pre-existing behaviour silently expanded it via the FS store.
@@ -233,6 +244,15 @@ export class WebDAVController extends PuterController {
                 email_confirmed: user.email_confirmed ?? false,
                 requires_email_confirmation:
                     user.requires_email_confirmation ?? false,
+                // Carry the signup-time verification flags so the
+                // assertVerifiedAccount() gate in #dispatch can see them on
+                // the Basic-auth path too (the cookie / `-token` paths get
+                // them from AuthService#actorUserFromRow). Omitting these is
+                // what let phone/card-gated accounts through WebDAV.
+                requires_phone_verification:
+                    user.requires_phone_verification ?? false,
+                requires_card_verification:
+                    user.requires_card_verification ?? false,
             },
         };
     }

src/backend/core/actor.ts

@@ -33,6 +33,15 @@ export interface ActorAccessToken {
     uid: string;
     issuer: Actor;
     authorized?: Actor | null;
+    /**
+     * Full-API-access ("personal access token") flag, set from the signed
+     * `full_access` JWT claim. Such a token may exercise everything its issuing
+     * user can do via the API and is admitted past `requireNonAccessTokenGate`
+     * — but is still rejected by `requireUserActor` / web-session gates, so it
+     * can never manage the account. Normal (scoped) access tokens leave this
+     * false and remain blocked from non-`allowAccessToken` routes.
+     */
+    fullAccess?: boolean;
 }
 
 export interface Actor {

src/backend/core/http/expressAugmentation.ts

@@ -66,6 +66,21 @@ declare global {
             /** True when the request's Host is a custom domain (not one of the configured Puter domains). */
             is_custom_domain?: boolean;
 
+            /**
+             * Coarse server-derived request fingerprint (IP + UA + accept
+             * headers), populated by the global fingerprint middleware. Always
+             * present; the same value the rate limiter keys on.
+             */
+            networkFingerprint?: string;
+
+            /**
+             * Client-supplied device fingerprint (ThumbmarkJS hash) from the
+             * body or `x-puter-device-fingerprint` header, populated by the
+             * global fingerprint middleware. Present only when the client sent a
+             * well-shaped value; spoofable but stable per device across IPs.
+             */
+            deviceFingerprint?: string;
+
             /** Parsed cookies, populated by the global `cookie-parser` middleware. */
             cookies?: Record<string, string>;
         }

src/backend/core/http/middleware/antiCsrf.js

@@ -73,6 +73,19 @@ export const antiCsrf = {
 export function requireAntiCsrf() {
     return async (req, _res, next) => {
         try {
+            // CSRF only applies to ambient credentials the browser attaches
+            // automatically (the session cookie). A full-access personal access
+            // token travels in the Authorization header — it can't be forged
+            // cross-origin — so it needs no anti-CSRF token, which is what makes
+            // routes like `/fs/down` reachable by a PAT. We deliberately scope
+            // this to full-access tokens (the only access tokens routed onto
+            // antiCsrf-protected endpoints): anything else — cookie-authed user
+            // actors, app actors, or a scoped token that somehow reaches here —
+            // still goes through the check below and fails closed.
+            if (req.actor?.accessToken?.fullAccess) {
+                return next();
+            }
+
             const sessionId = req.actor?.user?.uuid;
             if (!sessionId) {
                 return next(

src/backend/core/http/middleware/antiCsrf.test.js

@@ -118,6 +118,51 @@ describe('requireAntiCsrf middleware', () => {
         expect(arg).toBeUndefined();
     });
 
+    it('exempts a FULL-ACCESS access token (no anti_csrf token needed)', async () => {
+        // A full-access PAT is header-credentialed (bearer), so it can't be
+        // CSRF-forged — the middleware lets it through without a token. This is
+        // what makes antiCsrf routes like /fs/down reachable by a PAT.
+        const arg = await runMiddleware({
+            actor: {
+                user: { uuid: 'user-uuid-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'user-uuid-1' } },
+                    fullAccess: true,
+                },
+            },
+            body: {},
+        });
+        expect(arg).toBeUndefined();
+    });
+
+    it('does NOT exempt a scoped (non-full-access) access token — fails closed', async () => {
+        // Scoped tokens are never deliberately routed onto antiCsrf endpoints;
+        // if one reaches here it must still present a token (which it can't get)
+        // rather than getting a free pass.
+        const arg = await runMiddleware({
+            actor: {
+                user: { uuid: 'user-uuid-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'user-uuid-1' } },
+                },
+            },
+            body: {},
+        });
+        expect(isHttpError(arg)).toBe(true);
+        expect(arg.statusCode).toBe(400);
+    });
+
+    it('still enforces the token for cookie-authed user actors (no accessToken)', async () => {
+        const arg = await runMiddleware({
+            actor: { user: { uuid: 'user-uuid-1' } },
+            body: {},
+        });
+        expect(isHttpError(arg)).toBe(true);
+        expect(arg.statusCode).toBe(400);
+    });
+
     it('returns 401 unauthorized when no actor is attached', async () => {
         const arg = await runMiddleware({ body: { anti_csrf: 'whatever' } });
         expect(isHttpError(arg)).toBe(true);

src/backend/core/http/middleware/fingerprint.test.ts

@@ -0,0 +1,93 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import type { Request, Response } from 'express';
+import { describe, expect, it, vi } from 'vitest';
+import { createFingerprintMiddleware } from './fingerprint';
+
+const VALID_FP = 'dev-fingerprint-abc123';
+
+const run = (reqPartial: Partial<Request> & Record<string, unknown>) => {
+    const middleware = createFingerprintMiddleware();
+    const req = {
+        headers: {},
+        ...reqPartial,
+    } as unknown as Request;
+    const next = vi.fn();
+    middleware(req, {} as Response, next);
+    return { req, next };
+};
+
+describe('createFingerprintMiddleware', () => {
+    it('always sets a well-formed networkFingerprint and calls next', () => {
+        const { req, next } = run({ ip: '203.0.113.1' });
+
+        expect(req.networkFingerprint).toMatch(/^[A-Za-z0-9_-]{16}$/);
+        expect(next).toHaveBeenCalledOnce();
+    });
+
+    it('networkFingerprint is stable for identical inputs and varies by UA', () => {
+        const headers = { 'user-agent': 'UA/1.0' };
+        const a = run({ ip: '203.0.113.1', headers }).req.networkFingerprint;
+        const b = run({ ip: '203.0.113.1', headers }).req.networkFingerprint;
+        const c = run({
+            ip: '203.0.113.1',
+            headers: { 'user-agent': 'UA/2.0' },
+        }).req.networkFingerprint;
+
+        expect(a).toBe(b);
+        expect(a).not.toBe(c);
+    });
+
+    it('reads a well-shaped device fingerprint from the body', () => {
+        const { req } = run({ body: { fingerprint: VALID_FP } });
+        expect(req.deviceFingerprint).toBe(VALID_FP);
+    });
+
+    it('falls back to the x-puter-device-fingerprint header', () => {
+        const { req } = run({
+            headers: { 'x-puter-device-fingerprint': VALID_FP },
+        });
+        expect(req.deviceFingerprint).toBe(VALID_FP);
+    });
+
+    it('prefers the body fingerprint over the header', () => {
+        const { req } = run({
+            body: { fingerprint: VALID_FP },
+            headers: { 'x-puter-device-fingerprint': 'other-fingerprint-xyz' },
+        });
+        expect(req.deviceFingerprint).toBe(VALID_FP);
+    });
+
+    it('drops a malformed fingerprint to undefined', () => {
+        expect(run({ body: { fingerprint: 'bad fp!' } }).req.deviceFingerprint)
+            .toBeUndefined();
+        expect(run({ body: { fingerprint: 'short' } }).req.deviceFingerprint)
+            .toBeUndefined();
+        expect(
+            run({ body: { fingerprint: 123 as unknown as string } }).req
+                .deviceFingerprint,
+        ).toBeUndefined();
+    });
+
+    it('leaves deviceFingerprint undefined when nothing was supplied', () => {
+        const { req } = run({ ip: '203.0.113.1' });
+        expect(req.deviceFingerprint).toBeUndefined();
+    });
+});

src/backend/core/http/middleware/fingerprint.ts

@@ -0,0 +1,76 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import type { RequestHandler } from 'express';
+import { computeNetworkFingerprint } from './rateLimit.js';
+import '../expressAugmentation';
+
+/**
+ * Conservative charset/length for the client-supplied device fingerprint. It
+ * becomes part of KV keys downstream (abuse extension), so anything outside
+ * this shape is dropped to `undefined` rather than trusted. Kept in sync with
+ * the abuse extension's `DEVICE_FINGERPRINT_SHAPE`; the core only enforces the
+ * shape, never any abuse policy built on the value.
+ */
+const DEVICE_FINGERPRINT_SHAPE = /^[A-Za-z0-9._-]{8,128}$/;
+
+/** Header the GUI may send the device fingerprint on for non-signup requests. */
+const DEVICE_FINGERPRINT_HEADER = 'x-puter-device-fingerprint';
+
+function readDeviceFingerprint(req: {
+    body?: unknown;
+    headers?: Record<string, unknown>;
+}): string | undefined {
+    // Body first (what /signup already sends), then a header fallback so
+    // authenticated, bodyless-or-different-shape requests can still carry it.
+    const body = req.body as { fingerprint?: unknown } | undefined;
+    const candidate =
+        (typeof body?.fingerprint === 'string'
+            ? body.fingerprint
+            : undefined) ??
+        (typeof req.headers?.[DEVICE_FINGERPRINT_HEADER] === 'string'
+            ? (req.headers[DEVICE_FINGERPRINT_HEADER] as string)
+            : undefined);
+    if (candidate && DEVICE_FINGERPRINT_SHAPE.test(candidate)) return candidate;
+    return undefined;
+}
+
+/**
+ * Stamp request-scoped fingerprints onto `req` so any downstream gate, handler,
+ * or service (via the ALS `Context.get('req')`) can read them without recomputing:
+ *
+ *   - `req.networkFingerprint` — always set; a coarse IP+headers hash (the same
+ *     value the rate limiter keys on). Server-derived, so it can't be forged
+ *     away, but it's coarse (shared behind NAT/VPN, rotates with UA).
+ *   - `req.deviceFingerprint` — set only when the client supplied a well-shaped
+ *     device fingerprint (ThumbmarkJS hash) in the body or the
+ *     `x-puter-device-fingerprint` header; `undefined` otherwise. Client-supplied
+ *     and spoofable, but stable per real device across IP rotation.
+ *
+ * Install AFTER the body parsers (so the body fingerprint is readable) and
+ * before `requestContext` (so the snapshot into ALS already carries them).
+ * Never rejects — a missing/invalid device fingerprint is simply absent.
+ */
+export const createFingerprintMiddleware = (): RequestHandler => {
+    return (req, _res, next) => {
+        req.networkFingerprint = computeNetworkFingerprint(req);
+        req.deviceFingerprint = readDeviceFingerprint(req);
+        next();
+    };
+};

src/backend/core/http/middleware/gates.test.ts

@@ -25,7 +25,8 @@ import {
     adminOnlyGate,
     allowedAppIdsGate,
     requireAuthGate,
-    requireEmailConfirmedGate,
+    requireVerifiedAccount,
+    requireNonAccessTokenGate,
     requireUserActorGate,
     requireVerifiedGate,
     subdomainGate,
@@ -224,6 +225,110 @@ describe('requireUserActorGate', () => {
         const got = runGate(requireUserActorGate(), {});
         expectHttpError(got, 401, 'token_missing');
     });
+
+    it('rejects a FULL-ACCESS access token too — account routes stay closed', () => {
+        // The account wall is actor-type based: even a full-access PAT (which
+        // the resource wall lets through) is rejected here, so it can never
+        // reach change-password/email/2FA/token-minting/etc.
+        const got = runGate(requireUserActorGate(), {
+            actor: {
+                user: { uuid: 'u-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'u-1' } },
+                    fullAccess: true,
+                },
+            },
+        });
+        expectHttpError(got, 403, 'forbidden');
+    });
+
+    // allowFullAccess opt-in: relaxes ONLY the access-token half, for
+    // user-resource/inference routes (e.g. the AI proxy).
+    it('admits a full-access PAT when allowFullAccess is set', () => {
+        const got = runGate(requireUserActorGate({ allowFullAccess: true }), {
+            actor: {
+                user: { uuid: 'u-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'u-1' } },
+                    fullAccess: true,
+                },
+            },
+        });
+        expect(got).toBeUndefined();
+    });
+
+    it('still rejects a SCOPED access token even when allowFullAccess is set', () => {
+        const got = runGate(requireUserActorGate({ allowFullAccess: true }), {
+            actor: {
+                user: { uuid: 'u-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'u-1' } },
+                    // no fullAccess
+                },
+            },
+        });
+        expectHttpError(got, 403, 'forbidden');
+    });
+
+    it('still rejects a third-party app even when allowFullAccess is set', () => {
+        const got = runGate(requireUserActorGate({ allowFullAccess: true }), {
+            actor: { user: { uuid: 'u-1' }, app: { uid: 'app-1' } },
+        });
+        expectHttpError(got, 403, 'forbidden');
+    });
+});
+
+// -- requireNonAccessTokenGate --
+
+describe('requireNonAccessTokenGate', () => {
+    it('passes through for plain user actors', () => {
+        const got = runGate(requireNonAccessTokenGate(), {
+            actor: { user: { uuid: 'u-1' } },
+        });
+        expect(got).toBeUndefined();
+    });
+
+    it('passes through for app-under-user actors (only access tokens are gated)', () => {
+        const got = runGate(requireNonAccessTokenGate(), {
+            actor: { user: { uuid: 'u-1' }, app: { uid: 'app-1' } },
+        });
+        expect(got).toBeUndefined();
+    });
+
+    it('rejects a normal (scoped) access token with 403', () => {
+        const got = runGate(requireNonAccessTokenGate(), {
+            actor: {
+                user: { uuid: 'u-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'u-1' } },
+                },
+            },
+        });
+        expectHttpError(got, 403, 'forbidden');
+    });
+
+    it('admits a FULL-ACCESS access token (the resource-wall carve-out)', () => {
+        const got = runGate(requireNonAccessTokenGate(), {
+            actor: {
+                user: { uuid: 'u-1' },
+                accessToken: {
+                    uid: 'tok-1',
+                    issuer: { user: { uuid: 'u-1' } },
+                    fullAccess: true,
+                },
+            },
+        });
+        expect(got).toBeUndefined();
+    });
+
+    it("falls back to 401 if there's no actor at all (defensive)", () => {
+        const got = runGate(requireNonAccessTokenGate(), {});
+        expectHttpError(got, 401, 'token_missing');
+    });
 });
 
 // ── adminOnlyGate ───────────────────────────────────────────────────
@@ -335,11 +440,11 @@ describe('requireVerifiedGate', () => {
     });
 });
 
-// ── requireEmailConfirmedGate ───────────────────────────────────────
+// ── requireVerifiedAccount ──────────────────────────────────────────
 
-describe('requireEmailConfirmedGate', () => {
+describe('requireVerifiedAccount', () => {
     it('passes through users that do not require confirmation (e.g. legacy/temp)', () => {
-        const got = runGate(requireEmailConfirmedGate(), {
+        const got = runGate(requireVerifiedAccount(), {
             actor: {
                 user: {
                     uuid: 'u-1',
@@ -352,7 +457,7 @@ describe('requireEmailConfirmedGate', () => {
     });
 
     it('passes through confirmed users even when confirmation is required', () => {
-        const got = runGate(requireEmailConfirmedGate(), {
+        const got = runGate(requireVerifiedAccount(), {
             actor: {
                 user: {
                     uuid: 'u-1',
@@ -365,7 +470,7 @@ describe('requireEmailConfirmedGate', () => {
     });
 
     it('returns 403 email_confirmation_required for pending-confirmation users', () => {
-        const got = runGate(requireEmailConfirmedGate(), {
+        const got = runGate(requireVerifiedAccount(), {
             actor: {
                 user: {
                     uuid: 'u-1',
@@ -377,8 +482,51 @@ describe('requireEmailConfirmedGate', () => {
         expectHttpError(got, 403, 'email_confirmation_required');
     });
 
+    it('returns 403 phone_verification_required while the phone gate is set', () => {
+        const got = runGate(requireVerifiedAccount(), {
+            actor: {
+                user: {
+                    uuid: 'u-1',
+                    requires_email_confirmation: false,
+                    email_confirmed: true,
+                    requires_phone_verification: true,
+                },
+            },
+        });
+        expectHttpError(got, 403, 'phone_verification_required');
+    });
+
+    it('returns 403 card_verification_required while the card gate is set', () => {
+        const got = runGate(requireVerifiedAccount(), {
+            actor: {
+                user: {
+                    uuid: 'u-1',
+                    requires_email_confirmation: false,
+                    email_confirmed: true,
+                    requires_card_verification: true,
+                },
+            },
+        });
+        expectHttpError(got, 403, 'card_verification_required');
+    });
+
+    it('passes through once every gate is cleared', () => {
+        const got = runGate(requireVerifiedAccount(), {
+            actor: {
+                user: {
+                    uuid: 'u-1',
+                    requires_email_confirmation: true,
+                    email_confirmed: true,
+                    requires_phone_verification: false,
+                    requires_card_verification: false,
+                },
+            },
+        });
+        expect(got).toBeUndefined();
+    });
+
     it("passes through when there's no actor (auth gate handled it)", () => {
-        const got = runGate(requireEmailConfirmedGate(), {});
+        const got = runGate(requireVerifiedAccount(), {});
         expect(got).toBeUndefined();
     });
 });

src/backend/core/http/middleware/gates.ts

@@ -99,8 +99,18 @@ export const requireAuthGate = (): RequestHandler => {
  * Reject app-under-user and access-token actors with 403. Use on endpoints
  * that should only be exercised by a human session — settings changes,
  * admin-style actions on the user's own account.
+ *
+ * `allowFullAccess` (set per-route via the `allowFullAccessToken` route option)
+ * relaxes ONLY the access-token half: a full-access ("personal access token")
+ * actor is admitted, because it represents the user's own full API reach.
+ * Third-party apps are ALWAYS rejected, and scoped access tokens are always
+ * rejected. This opt-in is for user-resource / inference endpoints (AI proxy,
+ * etc.) that use this gate purely to keep apps out — NEVER for account or
+ * security management, which must stay closed to every access token.
  */
-export const requireUserActorGate = (): RequestHandler => {
+export const requireUserActorGate = (
+    opts: { allowFullAccess?: boolean } = {},
+): RequestHandler => {
     return (req, _res, next) => {
         const actor = req.actor;
         // requireAuth runs first; this gate just narrows the actor type.
@@ -108,7 +118,14 @@ export const requireUserActorGate = (): RequestHandler => {
             next(rejectAuth(req));
             return;
         }
-        if (actor.app || actor.accessToken) {
+        // Third-party apps are never allowed through this gate.
+        const appBlocked = !!actor.app;
+        // Access tokens are blocked unless the route opted in AND this is a
+        // full-access PAT (the user's own credential). Scoped tokens: blocked.
+        const tokenBlocked =
+            !!actor.accessToken &&
+            !(opts.allowFullAccess && actor.accessToken.fullAccess);
+        if (appBlocked || tokenBlocked) {
             next(
                 new HttpError(
                     403,
@@ -129,7 +146,13 @@ export const requireNonAccessTokenGate = (): RequestHandler => {
             next(rejectAuth(req));
             return;
         }
-        if (actor.accessToken) {
+        // Full-access ("personal access token") access tokens are admitted here:
+        // they carry the user's full API reach by design. They remain blocked
+        // from account management because those routes also use
+        // `requireUserActorGate`, which rejects ALL access tokens. Normal
+        // (scoped) access tokens stay blocked from non-`allowAccessToken`
+        // routes.
+        if (actor.accessToken && !actor.accessToken.fullAccess) {
             next(
                 new HttpError(
                     403,
@@ -201,29 +224,84 @@ export const requireVerifiedGate = (strictFlag: boolean): RequestHandler => {
 };
 
 /**
- * Reject authenticated users whose account is pending email confirmation
- * (`requires_email_confirmation && !email_confirmed`). Runs on every
- * authenticated route by default; routes that set `allowUnconfirmed: true`
- * skip this gate.
+ * Reject authenticated users whose account is still pending any signup-time
+ * verification — email confirmation, SMS phone verification, or credit-card
+ * verification. The abuse harness sets the phone/card flags on low-reputation
+ * signups (in place of a hard block), and this gate is what actually keeps
+ * those accounts out of the product until the flag clears: the flags live on
+ * `req.actor.user`, so every authenticated route enforces them, not just the
+ * GUI modal.
  *
- * Returns 403 with `email_confirmation_required` so clients can show the
- * confirmation prompt instead of a generic error.
+ * Runs on every authenticated route by default; routes that set
+ * `allowUnconfirmed: true` opt out (the verification endpoints themselves,
+ * plus essential flows like whoami / logout / save-account so a pending
+ * account can still reach the screens that clear the gate).
+ *
+ * Returns 403 with a per-gate legacy code (`email_confirmation_required` /
+ * `phone_verification_required` / `card_verification_required`) so clients can
+ * show the right prompt instead of a generic error. There is no state where a
+ * user should be allowed in with one verification pending, so any pending gate
+ * rejects.
  */
-export const requireEmailConfirmedGate = (): RequestHandler => {
+export const requireVerifiedAccount = (): RequestHandler => {
     return (req, _res, next) => {
-        const user = req.actor?.user;
-        if (user?.requires_email_confirmation && !user?.email_confirmed) {
-            next(
-                new HttpError(403, 'Please confirm your email to continue', {
-                    legacyCode: 'email_confirmation_required',
-                }),
-            );
+        try {
+            assertVerifiedAccount(req.actor?.user);
+        } catch (err) {
+            next(err);
             return;
         }
         next();
     };
 };
 
+/**
+ * The pending-verification check, factored out of {@link requireVerifiedAccount}
+ * so auth paths that build their own actor outside the route-option machinery
+ * can enforce the exact same gate. The WebDAV controller is the motivating
+ * case: it dispatches every method off a single `router.use`, so
+ * `requireVerifiedAccount` is never wired into its chain — it has to call this
+ * directly. Keeping one implementation is the point: a verification gate added
+ * here is picked up by every caller, so the paths can't drift (which is how
+ * WebDAV came to bypass the phone/card gate to begin with).
+ *
+ * Throws 403 with a per-gate legacy code (`email_confirmation_required` /
+ * `phone_verification_required` / `card_verification_required`) so clients can
+ * show the right prompt instead of a generic error. There is no state where a
+ * user should be let in with any verification pending, so the first pending
+ * gate rejects.
+ */
+export const assertVerifiedAccount = (
+    user:
+        | {
+              requires_email_confirmation?: unknown;
+              email_confirmed?: unknown;
+              requires_phone_verification?: unknown;
+              requires_card_verification?: unknown;
+          }
+        | undefined,
+): void => {
+    if (user?.requires_email_confirmation && !user?.email_confirmed) {
+        throw new HttpError(403, 'Please confirm your email to continue', {
+            legacyCode: 'email_confirmation_required',
+        });
+    }
+    if (user?.requires_phone_verification) {
+        throw new HttpError(
+            403,
+            'Please verify your phone number to continue',
+            {
+                legacyCode: 'phone_verification_required' as never,
+            },
+        );
+    }
+    if (user?.requires_card_verification) {
+        throw new HttpError(403, 'Please verify your card to continue', {
+            legacyCode: 'card_verification_required' as never,
+        });
+    }
+};
+
 /**
  * Reject unless the actor is acting through one of the named apps.
  * App-under-user actors are permitted iff `actor.app.uid` is in the allowList;

src/backend/core/http/middleware/hostRedirects.test.ts

@@ -264,6 +264,79 @@ describe('createUserSubdomainRedirect', () => {
         ]);
     });
 
+    const selfHosted = {
+        domain: 'puter.localhost',
+        static_hosting_domain: 'site.puter.localhost',
+        static_hosting_domain_alt: 'host.puter.localhost',
+        private_app_hosting_domain: 'app.puter.localhost',
+        private_app_hosting_domain_alt: 'dev.puter.localhost',
+    } as IConfig;
+
+    it('still redirects a bare subdomain on the main domain to the hosting domain (self-hosted)', () => {
+        const { out, next } = run(
+            createUserSubdomainRedirect(selfHosted),
+            makeReq({
+                subdomains: ['localhost', 'puter', 'foo'],
+                host: 'foo.puter.localhost',
+                originalUrl: '/bar?x=1',
+                protocol: 'http',
+            }),
+        );
+        expect(out.redirectArgs).toEqual([
+            302,
+            'http://foo.site.puter.localhost/bar?x=1',
+        ]);
+        expect(next).not.toHaveBeenCalled();
+    });
+
+    it('passes through hosts already on the static hosting domain (no redirect loop)', () => {
+        const { out, next } = run(
+            createUserSubdomainRedirect(selfHosted),
+            makeReq({
+                subdomains: ['localhost', 'puter', 'site', 'foo'],
+                host: 'foo.site.puter.localhost',
+                originalUrl: '/',
+            }),
+        );
+        expect(out.redirectArgs).toBeUndefined();
+        expect(next).toHaveBeenCalledTimes(1);
+    });
+
+    it('passes through hosts on the alt / private-app hosting domains too', () => {
+        for (const host of [
+            'foo.host.puter.localhost',
+            'foo.app.puter.localhost',
+            'foo.dev.puter.localhost',
+        ]) {
+            const { out, next } = run(
+                createUserSubdomainRedirect(selfHosted),
+                makeReq({
+                    subdomains: [
+                        'localhost',
+                        'puter',
+                        host.split('.')[1],
+                        'foo',
+                    ],
+                    host,
+                }),
+            );
+            expect(out.redirectArgs).toBeUndefined();
+            expect(next).toHaveBeenCalledTimes(1);
+        }
+    });
+
+    it('passes through the hosting-domain root itself (exact match, no loop)', () => {
+        const { out, next } = run(
+            createUserSubdomainRedirect(selfHosted),
+            makeReq({
+                subdomains: ['localhost', 'puter', 'site'],
+                host: 'site.puter.localhost',
+            }),
+        );
+        expect(out.redirectArgs).toBeUndefined();
+        expect(next).toHaveBeenCalledTimes(1);
+    });
+
     it("passes through when the request host has a port the configured domain doesn't", () => {
         // Edge case worth pinning: the suffix check is exact-`endsWith`,
         // so a port mismatch silently bypasses the redirect. Documenting

src/backend/core/http/middleware/hostRedirects.ts

@@ -88,6 +88,15 @@ export const createUserSubdomainRedirect = (
     if (!domain || !target) {
         return (_req, _res, next) => next();
     }
+
+    const hostingDomains = [
+        config.static_hosting_domain,
+        config.static_hosting_domain_alt,
+        config.private_app_hosting_domain,
+        config.private_app_hosting_domain_alt,
+    ]
+        .map((d) => (d ?? '').toLowerCase().split(':')[0])
+        .filter((d) => d.length > 0);
     return (req, res, next) => {
         const active = (
             req.subdomains?.[req.subdomains.length - 1] ?? ''
@@ -95,6 +104,14 @@ export const createUserSubdomainRedirect = (
         if (active === '' || RESERVED_SUBDOMAINS.has(active)) return next();
 
         const host = (req.headers.host ?? '').toLowerCase();
+        const hostName = host.split(':')[0];
+        if (
+            hostingDomains.some(
+                (d) => hostName === d || hostName.endsWith(`.${d}`),
+            )
+        ) {
+            return next();
+        }
         if (!host.endsWith(domain)) return next();
 
         // host ends in domain — swap the domain suffix for the hosting one,

src/backend/core/http/middleware/puterSite.test.ts

@@ -534,6 +534,51 @@ describe('createPuterSiteMiddleware — file serving', () => {
         expect(piped!.equals(body)).toBe(true);
     });
 
+    it('emits site.htmlServed with the original URL target including query string', async () => {
+        const owner = await makeUserWithHome();
+        const homePath = `/${owner.username}`;
+        const homeEntry = await server.stores.fsEntry.getEntryByPath(homePath);
+        const sub = `serveq-${Math.random().toString(36).slice(2, 8)}`;
+        await server.stores.subdomain.create({
+            userId: owner.id,
+            subdomain: sub,
+            rootDirId: homeEntry!.id,
+        });
+        await writeFile(
+            owner.id,
+            `${homePath}/index.html`,
+            Buffer.from('<html>hi</html>'),
+            'text/html',
+        );
+
+        const emitSpy = vi.spyOn(server.clients.event, 'emit');
+        try {
+            const mw = buildMiddleware();
+            const { res } = makeRes();
+            await mw(
+                makeReq({
+                    hostname: `${sub}.site.puter.localhost`,
+                    path: '/index.html',
+                    originalUrl: '/index.html?brand=paypal',
+                }),
+                res,
+                vi.fn(),
+            );
+
+            const htmlServedCall = emitSpy.mock.calls.find(
+                (call) => call[0] === 'site.htmlServed',
+            );
+            expect(htmlServedCall?.[1]).toMatchObject({
+                subdomain: sub,
+                requestPath: '/index.html',
+                requestUrl: '/index.html?brand=paypal',
+                mime: 'text/html',
+            });
+        } finally {
+            emitSpy.mockRestore();
+        }
+    });
+
     it('returns 206 status when a Range header is supplied', async () => {
         const owner = await makeUserWithHome();
         const homePath = `/${owner.username}`;

src/backend/core/http/middleware/puterSite.ts

@@ -555,8 +555,12 @@ export const createPuterSiteMiddleware = (
         res.setHeader('Content-Type', mime);
 
         // Fire-and-forget signal for downstream extensions
-        if (mime === 'text/html' || mime === 'application/xhtml+xml') {
+        const mimeBase = mime.split(';', 1)[0].trim().toLowerCase();
+        if (mimeBase === 'text/html' || mimeBase === 'application/xhtml+xml') {
             try {
+                const requestUrl = (req.originalUrl || '/').startsWith('/')
+                    ? req.originalUrl || '/'
+                    : `/${req.originalUrl}`;
                 layers.clients.event.emit(
                     'site.htmlServed',
                     {
@@ -564,7 +568,8 @@ export const createPuterSiteMiddleware = (
                         entry,
                         host,
                         requestPath: req.path,
-                        mime,
+                        requestUrl,
+                        mime: mimeBase,
                     },
                     {},
                 );

src/backend/core/http/middleware/rateLimit.js

@@ -392,7 +392,14 @@ function ip(req) {
     return req.ip || req.socket?.remoteAddress || 'unknown';
 }
 
-function fingerprint(req) {
+/**
+ * A coarse network fingerprint for a request: a short hash of the (proxy-aware)
+ * IP plus the headers a client can't trivially vary per-request without also
+ * changing how the request looks. Used as the default rate-limit key here, and
+ * exported so the global fingerprint middleware can stamp the identical value
+ * on `req.networkFingerprint` (one key space shared by both).
+ */
+export function computeNetworkFingerprint(req) {
     const parts = [
         ip(req),
         req.headers?.['user-agent'] || '',
@@ -406,6 +413,10 @@ function fingerprint(req) {
         .slice(0, 16);
 }
 
+function fingerprint(req) {
+    return computeNetworkFingerprint(req);
+}
+
 // -- Route middleware ------------------------------------------------
 
 /**

src/backend/core/http/middleware/rateLimit.test.js

@@ -140,6 +140,67 @@ describe('rateLimitGate — memory backend (default)', () => {
         expect(isHttpError(re)).toBe(true);
     });
 
+    // Stacked gates, mirroring the materializer's handling of a
+    // `rateLimit: [...]` array (each entry is its own gate; a rejection
+    // short-circuits the chain). This is the credential-endpoint pattern:
+    // per-fingerprint budget for fairness on shared IPs, per-IP backstop
+    // against header rotation.
+    const runStack = async (stack, req) => {
+        for (const opts of stack) {
+            const err = await runGate(opts, req);
+            if (err) return err;
+        }
+        return undefined;
+    };
+
+    it('stacked gates: per-IP backstop catches header rotation that escapes the fingerprint key', async () => {
+        const stack = [
+            { limit: 3, window: 60_000, scope: 'stack-fp' },
+            { limit: 6, window: 60_000, key: 'ip', scope: 'stack-ip' },
+        ];
+        // Rotating the User-Agent mints a fresh fingerprint bucket every
+        // request, so the fingerprint gate admits all of these — the IP
+        // backstop must cap the total anyway.
+        for (let i = 0; i < 6; i++) {
+            const arg = await runStack(
+                stack,
+                makeReq({ headers: { 'user-agent': `rotated-ua-${i}` } }),
+            );
+            expect(arg).toBeUndefined();
+        }
+        const rejected = await runStack(
+            stack,
+            makeReq({ headers: { 'user-agent': 'rotated-ua-final' } }),
+        );
+        expect(isHttpError(rejected)).toBe(true);
+        expect(rejected.statusCode).toBe(429);
+    });
+
+    it('stacked gates: clients sharing an IP each get their own fingerprint budget under the IP cap', async () => {
+        const stack = [
+            { limit: 2, window: 60_000, scope: 'stack2-fp' },
+            { limit: 10, window: 60_000, key: 'ip', scope: 'stack2-ip' },
+        ];
+        // Two clients (distinct UAs) behind one NAT IP: each gets the
+        // full per-fingerprint budget, and each is cut off by its own
+        // fingerprint bucket — not starved by the other's traffic.
+        for (const ua of ['alice-browser', 'bob-browser']) {
+            for (let i = 0; i < 2; i++) {
+                expect(
+                    await runStack(
+                        stack,
+                        makeReq({ headers: { 'user-agent': ua } }),
+                    ),
+                ).toBeUndefined();
+            }
+            const rejected = await runStack(
+                stack,
+                makeReq({ headers: { 'user-agent': ua } }),
+            );
+            expect(isHttpError(rejected)).toBe(true);
+        }
+    });
+
     it("falls back to fingerprint when 'user' strategy is selected but no actor present", async () => {
         // The fallback prevents anonymous traffic from sharing a single
         // bucket (which would invite trivial DoS via global rate-limit).

src/backend/core/http/routeLifecycle.test.ts

@@ -0,0 +1,251 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import { EventEmitter } from 'node:events';
+import type { Request, Response } from 'express';
+import { describe, expect, it } from 'vitest';
+import { EventClient } from '../../clients/event/EventClient';
+import type {
+    EventMap,
+    RouteLifecycleEvent,
+} from '../../clients/event/types';
+import type { IConfig } from '../../types';
+import {
+    createRouteLifecycleMiddleware,
+    pathKeySegment,
+    routeEventKeyBase,
+} from './routeLifecycle';
+
+// -- Fakes --
+//
+// EventClient is a pure in-memory bus (no external deps), so we use the real
+// one and register real listeners. The request/response are the boundary we
+// fake: a minimal `res` that is an EventEmitter (for `finish`/`close`) and
+// records status/body.
+
+const makeEvents = () => new EventClient({} as IConfig);
+
+class FakeRes extends EventEmitter {
+    statusCode = 200;
+    writableFinished = false;
+    headersSent = false;
+    body: unknown;
+    status(code: number) {
+        this.statusCode = code;
+        return this;
+    }
+    json(b: unknown) {
+        this.body = b;
+        return this;
+    }
+}
+
+const makeReq = (actorUuid?: string): Request =>
+    ({
+        actor: actorUuid ? { user: { uuid: actorUuid } } : undefined,
+    }) as unknown as Request;
+
+// Collect every emit on a given key into an array for assertions.
+const record = (events: EventClient, key: keyof EventMap) => {
+    const seen: RouteLifecycleEvent[] = [];
+    events.on(key as never, (_k, data) =>
+        seen.push(data as RouteLifecycleEvent),
+    );
+    return seen;
+};
+
+const POST = 'post' as const;
+const PATH = '/fs/completeBatchWrite';
+const BASE = 'route.post.fs.completeBatchWrite';
+
+describe('pathKeySegment', () => {
+    it('joins path parts with dots and drops slashes', () => {
+        expect(pathKeySegment('/fs/completeBatchWrite')).toBe(
+            'fs.completeBatchWrite',
+        );
+    });
+
+    it('strips the leading colon from path params', () => {
+        expect(pathKeySegment('/foo/:id')).toBe('foo.id');
+    });
+
+    it('maps the root path and non-string paths to placeholders', () => {
+        expect(pathKeySegment('/')).toBe('root');
+        expect(pathKeySegment(/^\/x/)).toBe('_');
+    });
+});
+
+describe('routeEventKeyBase', () => {
+    it('scopes the key to method + normalized path', () => {
+        expect(routeEventKeyBase(POST, PATH)).toBe(BASE);
+    });
+});
+
+describe('createRouteLifecycleMiddleware', () => {
+    it('emits before, then after on a clean finish', async () => {
+        const events = makeEvents();
+        const before = record(events, `${BASE}.before` as keyof EventMap);
+        const after = record(events, `${BASE}.after` as keyof EventMap);
+        const error = record(events, `${BASE}.error` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const req = makeReq('u-1');
+        const res = new FakeRes();
+        let nextCalled = false;
+
+        await mw(req, res as unknown as Response, () => {
+            nextCalled = true;
+        });
+
+        expect(nextCalled).toBe(true);
+        expect(before).toHaveLength(1);
+        expect(before[0]).toMatchObject({
+            phase: 'before',
+            method: 'post',
+            path: PATH,
+            actor: { user: { uuid: 'u-1' } },
+            actorUid: 'user:u-1',
+        });
+        // The live req/res are exposed so listeners can read the body or
+        // respond themselves.
+        expect(before[0].req).toBe(req);
+        expect(before[0].res).toBe(res);
+
+        res.statusCode = 200;
+        res.writableFinished = true;
+        res.emit('finish');
+
+        expect(after).toHaveLength(1);
+        expect(after[0]).toMatchObject({ phase: 'after', statusCode: 200 });
+        expect(typeof after[0].durationMs).toBe('number');
+        expect(error).toHaveLength(0);
+    });
+
+    it('emits error (not after) when the response is a 5xx', async () => {
+        const events = makeEvents();
+        const after = record(events, `${BASE}.after` as keyof EventMap);
+        const error = record(events, `${BASE}.error` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const res = new FakeRes();
+        await mw(makeReq(), res as unknown as Response, () => {});
+
+        res.statusCode = 500;
+        res.writableFinished = true;
+        res.emit('finish');
+
+        expect(after).toHaveLength(0);
+        expect(error).toHaveLength(1);
+        expect(error[0]).toMatchObject({ phase: 'error', statusCode: 500 });
+    });
+
+    it('emits error when the connection closes without finishing (abort)', async () => {
+        const events = makeEvents();
+        const error = record(events, `${BASE}.error` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const res = new FakeRes();
+        await mw(makeReq(), res as unknown as Response, () => {});
+
+        res.writableFinished = false;
+        res.emit('close');
+
+        expect(error).toHaveLength(1);
+        expect(error[0].phase).toBe('error');
+        expect(error[0].error).toBeInstanceOf(Error);
+    });
+
+    it('emits a terminal event only once across finish + close', async () => {
+        const events = makeEvents();
+        const after = record(events, `${BASE}.after` as keyof EventMap);
+        const error = record(events, `${BASE}.error` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const res = new FakeRes();
+        await mw(makeReq(), res as unknown as Response, () => {});
+
+        res.statusCode = 200;
+        res.writableFinished = true;
+        res.emit('finish');
+        res.emit('close');
+
+        expect(after).toHaveLength(1);
+        expect(error).toHaveLength(0);
+    });
+
+    it('vetoes the request when a before listener sets allow=false', async () => {
+        const events = makeEvents();
+        events.on(`${BASE}.before` as never, (_k, data) => {
+            (data as RouteLifecycleEvent).allow = false;
+            (data as RouteLifecycleEvent).rejectReason = 'quota exceeded';
+        });
+        const reject = record(events, `${BASE}.reject` as keyof EventMap);
+        const after = record(events, `${BASE}.after` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const res = new FakeRes();
+        let nextCalled = false;
+        await mw(makeReq('u-9'), res as unknown as Response, () => {
+            nextCalled = true;
+        });
+
+        expect(nextCalled).toBe(false);
+        expect(res.statusCode).toBe(403);
+        expect(res.body).toMatchObject({
+            error: { code: 'forbidden', message: 'quota exceeded' },
+        });
+        expect(reject).toHaveLength(1);
+        expect(reject[0]).toMatchObject({
+            phase: 'reject',
+            statusCode: 403,
+            rejectReason: 'quota exceeded',
+        });
+        // A vetoed request never runs, so no terminal after fires.
+        res.emit('finish');
+        expect(after).toHaveLength(0);
+    });
+
+    it('treats a listener-sent response as a terminal after (not a reject) with the real status', async () => {
+        const events = makeEvents();
+        events.on(`${BASE}.before` as never, (_k, data) => {
+            // Listener answers the request itself without vetoing.
+            const e = data as RouteLifecycleEvent;
+            (e.res as unknown as FakeRes).headersSent = true;
+            e.res.status(418).json({ teapot: true });
+        });
+        const reject = record(events, `${BASE}.reject` as keyof EventMap);
+        const after = record(events, `${BASE}.after` as keyof EventMap);
+
+        const mw = createRouteLifecycleMiddleware(events, POST, PATH);
+        const res = new FakeRes();
+        let nextCalled = false;
+        await mw(makeReq(), res as unknown as Response, () => {
+            nextCalled = true;
+        });
+
+        expect(nextCalled).toBe(false);
+        // Listener's own 418 stands; the middleware doesn't clobber it with 403.
+        expect(res.statusCode).toBe(418);
+        expect(res.body).toEqual({ teapot: true });
+        // Not a veto, so no reject — a terminal `after` keyed off the real 418.
+        expect(reject).toHaveLength(0);
+        expect(after).toHaveLength(1);
+        expect(after[0]).toMatchObject({ phase: 'after', statusCode: 418 });
+    });
+});

src/backend/core/http/routeLifecycle.ts

@@ -0,0 +1,162 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import type { Request, RequestHandler } from 'express';
+import { actorUid } from '../actor';
+import type { EventClient } from '../../clients/event/EventClient';
+import type { RouteMethod } from './types';
+
+type RoutePathValue = string | RegExp | Array<string | RegExp> | undefined;
+
+/**
+ * Normalize a route path into a dot-separated key segment so lifecycle events
+ * can be scoped per endpoint. Strips slashes and param markers so the key is
+ * stable (`/drivers/call` -> `drivers.call`, `/foo/:id` -> `foo.id`).
+ * Non-string paths (RegExp / arrays) collapse to a single `_` placeholder.
+ */
+export const pathKeySegment = (fullPath: RoutePathValue): string => {
+    if (typeof fullPath !== 'string') return '_';
+    const seg = fullPath
+        .split('/')
+        .filter(Boolean)
+        .map((p) => p.replace(/^:/, '').replace(/[^A-Za-z0-9_-]/g, '_'))
+        .join('.');
+    return seg || 'root';
+};
+
+/**
+ * Scoped event-key base for a route: `route.<method>.<normalized-path>`.
+ * Callers append `.before` / `.after` / `.error` / `.reject`.
+ */
+export const routeEventKeyBase = (
+    method: RouteMethod,
+    fullPath: RoutePathValue,
+): `route.${string}` => `route.${method}.${pathKeySegment(fullPath)}`;
+
+/**
+ * Build the per-endpoint lifecycle middleware for one route.
+ *
+ * On each request it: emits `before` (awaited, so a listener may veto by
+ * setting `allow = false`), times the handler, and emits a single terminal
+ * event when the response finishes or the connection closes — `after` on a
+ * clean response, `error` on an abort or a >=500, `reject` only when a
+ * listener vetoes (it then answers 403 and the handler never runs). A listener
+ * that writes its own response without vetoing terminates as `after`/`error`
+ * keyed off the real status, not `reject`.
+ *
+ * Subscribers can listen on `route.*`, `route.<method>.*`, or the exact key.
+ */
+export const createRouteLifecycleMiddleware = (
+    events: EventClient,
+    method: RouteMethod,
+    fullPath: RoutePathValue,
+): RequestHandler => {
+    const pathLabel =
+        typeof fullPath === 'string' ? fullPath : String(fullPath);
+    const keyBase = routeEventKeyBase(method, fullPath);
+
+    return async (req: Request, res, next) => {
+        const actor = req.actor ? actorUid(req.actor) : undefined;
+        const startedAt = Date.now();
+        const base = {
+            method,
+            path: pathLabel,
+            req,
+            res,
+            actor: req.actor,
+            actorUid: actor,
+        };
+
+        const beforeEvent = {
+            phase: 'before' as const,
+            ...base,
+            allow: true as boolean,
+            rejectReason: undefined as string | undefined,
+        };
+        await events.emitAndWait(`${keyBase}.before`, beforeEvent, {});
+
+        // Explicit veto: a listener set `allow = false`. Emit `reject` and
+        // answer 403 — unless the listener already wrote its own response, in
+        // which case we report that status rather than a misleading 403.
+        if (beforeEvent.allow === false) {
+            events.emit(
+                `${keyBase}.reject`,
+                {
+                    phase: 'reject',
+                    ...base,
+                    statusCode: res.headersSent ? res.statusCode : 403,
+                    durationMs: Date.now() - startedAt,
+                    rejectReason: beforeEvent.rejectReason,
+                },
+                {},
+            );
+            if (!res.headersSent) {
+                res.status(403).json({
+                    error: {
+                        code: 'forbidden',
+                        message:
+                            beforeEvent.rejectReason ?? 'Blocked by policy',
+                    },
+                });
+            }
+            return;
+        }
+
+        // A listener answered the request itself without vetoing (it wrote a
+        // response in the `before` hook). That's a normal terminal outcome,
+        // not a reject — emit `after`/`error` keyed off the real status.
+        if (res.headersSent) {
+            const statusCode = res.statusCode;
+            const isError = statusCode >= 500;
+            events.emit(
+                `${keyBase}.${isError ? 'error' : 'after'}`,
+                {
+                    phase: isError ? 'error' : 'after',
+                    ...base,
+                    statusCode,
+                    durationMs: Date.now() - startedAt,
+                },
+                {},
+            );
+            return;
+        }
+
+        let settled = false;
+        const settle = (aborted: boolean) => {
+            if (settled) return;
+            settled = true;
+            const statusCode = res.statusCode;
+            const isError = aborted || statusCode >= 500;
+            events.emit(
+                `${keyBase}.${isError ? 'error' : 'after'}`,
+                {
+                    phase: isError ? 'error' : 'after',
+                    ...base,
+                    statusCode,
+                    durationMs: Date.now() - startedAt,
+                    ...(aborted ? { error: new Error('request aborted') } : {}),
+                },
+                {},
+            );
+        };
+        res.once('finish', () => settle(false));
+        res.once('close', () => settle(!res.writableFinished));
+        next();
+    };
+};

src/backend/core/http/types.ts

@@ -66,6 +66,22 @@ export type RoutePath = string | RegExp | Array<string | RegExp>;
  * in the chain. Commented-out slots are reserved for the next chunks
  * (body parsing, post-auth gates, timing).
  */
+/** One rate-limit window. See `RouteOptions.rateLimit` for semantics. */
+export interface RouteRateLimit {
+    limit: number;
+    window: number;
+    /**
+     * Per-subscription overrides for `limit` (`SubscriptionPolicy.id`
+     * → cap). Same mechanic as `concurrent.bySubscription`: resolved
+     * via the metering service per request; falls back to the base
+     * `limit` when there's no actor / no metering / no match.
+     */
+    bySubscription?: Record<string, number>;
+    key?: 'fingerprint' | 'ip' | 'user' | ((req: Request) => string);
+    scope?: string;
+    backend?: 'memory' | 'redis' | 'kv';
+}
+
 export interface RouteOptions {
     /** Extra per-route middleware. Applied after built-in gates, before the handler. */
     middleware?: RequestHandler[];
@@ -86,6 +102,17 @@ export interface RouteOptions {
     /** Reject app/access-token actors. Implies `requireAuth`. */
     requireUserActor?: boolean;
 
+    /**
+     * Only meaningful alongside `requireUserActor`. Relaxes the access-token
+     * half of that gate so a FULL-ACCESS personal access token (the user's own
+     * credential) is admitted — third-party apps and scoped tokens stay
+     * blocked. Use ONLY on user-resource / inference endpoints that gate with
+     * `requireUserActor` purely to keep apps out (e.g. the AI proxy). NEVER set
+     * on account/security/credential routes — those must stay closed to every
+     * access token. Default-deny: omitting this keeps PATs blocked.
+     */
+    allowFullAccessToken?: boolean;
+
     /** Allows access-tokens */
     allowAccessToken?: boolean;
 
@@ -198,21 +225,15 @@ export interface RouteOptions {
      * All registered backends stay co-resident at runtime, so different
      * routes can pick whichever fits their access pattern. Omitting
      * `backend` uses the server-wide default (`config.rate_limit.backend`).
+     *
+     * An array applies every limit independently (a request must pass all
+     * of them). Use this to pair a per-client budget with a coarser
+     * backstop on a different key — e.g. per-fingerprint for fairness on
+     * shared IPs, plus per-IP so rotating headers can't mint fresh
+     * fingerprint buckets indefinitely. Give each entry its own `scope`
+     * so the counters don't collide.
      */
-    rateLimit?: {
-        limit: number;
-        window: number;
-        /**
-         * Per-subscription overrides for `limit` (`SubscriptionPolicy.id`
-         * → cap). Same mechanic as `concurrent.bySubscription`: resolved
-         * via the metering service per request; falls back to the base
-         * `limit` when there's no actor / no metering / no match.
-         */
-        bySubscription?: Record<string, number>;
-        key?: 'fingerprint' | 'ip' | 'user' | ((req: Request) => string);
-        scope?: string;
-        backend?: 'memory' | 'redis' | 'kv';
-    };
+    rateLimit?: RouteRateLimit | RouteRateLimit[];
 
     /**
      * Concurrent in-flight limiting. Caps how many requests for this

src/backend/drivers/ai-chat/ChatCompletionDriver.test.ts

@@ -547,6 +547,46 @@ describe('ChatCompletionDriver.complete credit gate and max_tokens cap', () => {
         expect(passed.max_tokens!).toBeGreaterThan(0);
     });
 
+    it('throws 402 instead of leaving `max_tokens` unset when credits cannot afford one output token', async () => {
+        // Regression: previously a sub-1 cap set max_tokens to `undefined`,
+        // which let the provider run to the model's full output limit and
+        // overdraw the account. It must reject instead.
+        vi.spyOn(FakeChatProvider.prototype, 'models').mockResolvedValueOnce([
+            {
+                id: 'capme',
+                aliases: [],
+                costs_currency: 'usd-cents',
+                costs: { input_tokens: 1000, output_tokens: 2000 },
+                max_tokens: 8192,
+            },
+        ]);
+        const d = await makeDriver();
+
+        // Pass the cheap pre-flight gate but leave a balance too small to
+        // afford a single 2000-microcent output token.
+        vi.spyOn(server.services.metering, 'hasEnoughCredits').mockResolvedValue(
+            true,
+        );
+        vi.spyOn(server.services.metering, 'getRemainingUsage').mockResolvedValue(
+            100,
+        );
+
+        const completeSpy = vi.spyOn(FakeChatProvider.prototype, 'complete');
+
+        await expect(
+            withTestActor(() =>
+                d.complete({
+                    model: 'capme',
+                    messages: [{ role: 'user', content: 'hi' }],
+                }),
+            ),
+        ).rejects.toMatchObject({
+            statusCode: 402,
+            legacyCode: 'insufficient_funds',
+        });
+        expect(completeSpy).not.toHaveBeenCalled();
+    });
+
     it('rejects subscriber-only models for the default free subscription', async () => {
         vi.spyOn(FakeChatProvider.prototype, 'models').mockResolvedValueOnce([
             {

src/backend/drivers/ai-chat/ChatCompletionDriver.ts

@@ -17,8 +17,9 @@
  * along with this program.  If not, see <https://www.gnu.org/licenses/>.
  */
 
-import { PassThrough } from 'node:stream';
 import crypto from 'node:crypto';
+import { PassThrough } from 'node:stream';
+import { EventMap } from '../../clients/event/types.js';
 import { Context } from '../../core/context.js';
 import { HttpError } from '../../core/http/HttpError.js';
 import {
@@ -28,12 +29,17 @@ import {
 import type { DriverStreamResult } from '../meta.js';
 import { PuterDriver } from '../types.js';
 import { AI_CONCURRENT, AI_RATE_LIMIT } from '../util/aiLimits.js';
+import { AlibabaProvider } from './providers/alibaba/AlibabaProvider.js';
+import { AzureChatProvider } from './providers/azure/AzureChatProvider.js';
+import { AzureResponsesProvider } from './providers/azure/AzureResponsesProvider.js';
 import { ClaudeProvider } from './providers/claude/ClaudeProvider.js';
 import { DeepSeekProvider } from './providers/deepseek/DeepSeekProvider.js';
 import { FakeChatProvider } from './providers/FakeChatProvider.js';
 import { GeminiChatProvider } from './providers/gemini/GeminiChatProvider.js';
 import { GroqAIProvider } from './providers/groq/GroqAIProvider.js';
+import { MiniMaxProvider } from './providers/minimax/MiniMaxProvider.js';
 import { MistralAIProvider } from './providers/mistral/MistralAiProvider.js';
+import { MoonshotProvider } from './providers/moonshot/MoonshotProvider.js';
 import { OllamaChatProvider } from './providers/ollama/OllamaProvider.js';
 import { OpenAiChatProvider } from './providers/openai/OpenAiChatCompletionsProvider.js';
 import { OpenAiResponsesChatProvider } from './providers/openai/OpenAiChatResponsesProvider.js';
@@ -41,9 +47,6 @@ import { OpenRouterProvider } from './providers/openrouter/OpenRouterProvider.js
 import { TogetherAIProvider } from './providers/together/TogetherAIProvider.js';
 import { XAIProvider } from './providers/xai/XAIProvider.js';
 import { ZAIProvider } from './providers/zai/ZAIProvider.js';
-import { AlibabaProvider } from './providers/alibaba/AlibabaProvider.js';
-import { MoonshotProvider } from './providers/moonshot/MoonshotProvider.js';
-import { MiniMaxProvider } from './providers/minimax/MiniMaxProvider.js';
 import type {
     IChatCompleteResult,
     IChatModel,
@@ -57,7 +60,6 @@ import {
     normalize_single_message,
 } from './utils/Messages.js';
 import { AIChatStream } from './utils/Streaming.js';
-import { EventMap } from '../../clients/event/types.js';
 
 const MAX_FALLBACKS = 4; // includes first attempt
 
@@ -277,7 +279,7 @@ export class ChatCompletionDriver extends PuterDriver {
 
         let intendedProvider = args.provider || '';
         if (!args.model && !intendedProvider) {
-            intendedProvider = 'claude'; // default provider
+            intendedProvider = 'azure-openai'; // default provider
         }
         if (
             !args.model &&
@@ -398,16 +400,25 @@ export class ChatCompletionDriver extends PuterDriver {
                     remainingCredits - approximateInputCost;
                 const maxAllowedOutputTokens =
                     maxAllowedOutputUcents / outputTokenCost;
-                if (maxAllowedOutputTokens) {
-                    const cap = Math.floor(
-                        Math.min(
-                            args.max_tokens ?? Number.POSITIVE_INFINITY,
-                            maxAllowedOutputTokens,
-                            model.max_tokens - approximateTokenCount,
-                        ),
-                    );
-                    args.max_tokens = cap < 1 ? undefined : cap;
+                const cap = Math.floor(
+                    Math.min(
+                        args.max_tokens ?? Number.POSITIVE_INFINITY,
+                        maxAllowedOutputTokens,
+                        model.max_tokens - approximateTokenCount,
+                    ),
+                );
+                // `cap` is the credit-bounded ceiling on output tokens. When
+                // it drops below 1 the user can't afford even a single output
+                // token, so reject the request. Crucially we must NOT leave
+                // `max_tokens` unset here: an undefined max_tokens lets the
+                // provider run to the model's full output limit (e.g. 128k for
+                // Claude), billing far past the user's remaining balance.
+                if (cap < 1) {
+                    throw new HttpError(402, 'No usage left for request.', {
+                        legacyCode: 'insufficient_funds',
+                    });
                 }
+                args.max_tokens = cap;
             }
         }
 
@@ -796,6 +807,43 @@ export class ChatCompletionDriver extends PuterDriver {
             );
         }
 
+        // Azure AI Foundry (OpenAI + xAI Grok). Registered before the regular
+        // OpenAI/xAI providers so that since its costs mirror theirs but
+        // Azure is preferred for us, it takes precedence in the per-model
+        // bucket
+        const azureOpenai = providers['azure-openai'];
+        const azureOpenaiKey = readKey(azureOpenai);
+        const azureOpenaiURL = azureOpenai?.apiURL as string | undefined;
+        if (azureOpenaiKey && azureOpenaiURL) {
+            const azureStores = {
+                fsEntry: this.stores.fsEntry,
+                s3Object: this.stores.s3Object,
+            };
+            const azureConfig = {
+                apiKey: azureOpenaiKey,
+                apiURL: azureOpenaiURL,
+            };
+            const azureCompletions = new AzureChatProvider(
+                metering,
+                azureStores,
+                this.services.fs,
+                azureConfig,
+            );
+            // Codex / Responses-API-only models can't use Chat Completions, so
+            // they route through a sibling Responses provider pointed at the
+            // same Azure endpoint. web_search (also Responses-only) delegates
+            // here too.
+            const azureResponses = new AzureResponsesProvider(
+                metering,
+                azureStores,
+                this.services.fs,
+                azureConfig,
+            );
+            azureCompletions.setResponsesProvider(azureResponses);
+            this.#providers['azure-openai'] = azureCompletions;
+            this.#providers['azure-openai-responses'] = azureResponses;
+        }
+
         const openaiKey = readKey(providers['openai-completion']);
         if (openaiKey) {
             const openaiStores = {

src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.integration.test.ts

@@ -0,0 +1,101 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+/**
+ * Integration test for the Azure AI Foundry chat-completions provider.
+ *
+ * Hits the real Azure endpoint. Exercises both flavours of model the
+ * provider fronts:
+ *   - an OpenAI model (`gpt-4o`, non-reasoning so `max_tokens=16` returns
+ *     visible text), and
+ *   - an xAI Grok model (`grok-4-20-non-reasoning`), which regression-tests
+ *     the `safety_identifier` param being stripped for Grok (Azure's Grok
+ *     deployments 400 on that OpenAI-only argument).
+ *
+ * Skipped unless both `PUTER_TEST_AI_AZURE_OPENAI_API_KEY` and
+ * `PUTER_TEST_AI_AZURE_OPENAI_API_URL` are set.
+ */
+
+import { describe, expect, it } from 'vitest';
+import {
+    INTEGRATION_TEST_TIMEOUT_MS,
+    makeMeteringStub,
+    optionalEnv,
+    skipUnlessEnv,
+    withTestActor,
+} from '../../../integrationTestUtil.js';
+import { AzureChatProvider } from './AzureChatProvider.js';
+
+const KEY_ENV = 'PUTER_TEST_AI_AZURE_OPENAI_API_KEY';
+const URL_ENV = 'PUTER_TEST_AI_AZURE_OPENAI_API_URL';
+
+describe.skipIf(skipUnlessEnv(KEY_ENV) || skipUnlessEnv(URL_ENV))(
+    'AzureChatProvider (integration)',
+    () => {
+        const buildProvider = () =>
+            new AzureChatProvider(
+                makeMeteringStub(),
+                { fsEntry: undefined as never, s3Object: undefined as never },
+                undefined as never,
+                { apiKey: optionalEnv(KEY_ENV)!, apiURL: optionalEnv(URL_ENV)! },
+            );
+
+        const expectNonEmptyText = (result: unknown) => {
+            const text = (result as { message?: { content?: string } }).message
+                ?.content;
+            expect(typeof text === 'string' && text.length > 0).toBe(true);
+        };
+
+        it(
+            'returns a non-empty completion from gpt-4o',
+            { timeout: INTEGRATION_TEST_TIMEOUT_MS },
+            async () => {
+                const provider = buildProvider();
+                const result = await withTestActor(() =>
+                    provider.complete({
+                        model: 'gpt-4o',
+                        messages: [
+                            { role: 'user', content: 'Say hi in one word.' },
+                        ],
+                        max_tokens: 16,
+                    }),
+                );
+                expectNonEmptyText(result);
+            },
+        );
+
+        it(
+            'returns a non-empty completion from grok-4-20-non-reasoning (no safety_identifier 400)',
+            { timeout: INTEGRATION_TEST_TIMEOUT_MS },
+            async () => {
+                const provider = buildProvider();
+                const result = await withTestActor(() =>
+                    provider.complete({
+                        model: 'grok-4-20-non-reasoning',
+                        messages: [
+                            { role: 'user', content: 'Say hi in one word.' },
+                        ],
+                        max_tokens: 16,
+                    }),
+                );
+                expectNonEmptyText(result);
+            },
+        );
+    },
+);

src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.ts

@@ -0,0 +1,277 @@
+/*
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import { HttpError } from '@heyputer/backend/src/core/http/HttpError.js';
+import { OpenAI } from 'openai';
+import { ChatCompletionCreateParams } from 'openai/resources/index.js';
+import { Context } from '../../../../core/context.js';
+import type { FSService } from '../../../../services/fs/FSService.js';
+import type { MeteringService } from '../../../../services/metering/MeteringService.js';
+import type { FSEntryStore } from '../../../../stores/fs/FSEntryStore.js';
+import type { S3ObjectStore } from '../../../../stores/fs/S3ObjectStore.js';
+import type { IChatProvider, ICompleteArguments } from '../../types.js';
+import {
+    messagesHaveCompaction,
+    wantsCompaction,
+} from '../../utils/compaction.js';
+import * as OpenAiUtil from '../../utils/OpenAIUtil.js';
+import { processPuterPathUploads } from '../openai/fileUpload.js';
+import { AZURE_MODELS } from './models.js';
+
+/**
+ * AzureChatProvider exposes the models we serve through Azure AI Foundry.
+ * Despite the name, this is not OpenAI-only — Azure AI also fronts xAI's Grok
+ * models — so it carries its own {@link AZURE_MODELS} list instead of reusing
+ * the OpenAI one. It speaks the OpenAI-compatible Chat Completions API,
+ * pointing the client at a configurable Azure endpoint authenticated with an
+ * Azure-issued API key.
+ *
+ * Billing note: the model `costs` are the standard public OpenAI / xAI list
+ * prices, NOT Azure's. Azure is subsidised for us, so routing through it is
+ * cheaper while we still bill users at the normal model price.
+ *
+ * Implements the puter-chat-completion interface and handles usage tracking,
+ * spending records, and content moderation.
+ */
+export class AzureChatProvider implements IChatProvider {
+    /**
+     * @type {import('openai').OpenAI}
+     */
+    #openAi: OpenAI;
+
+    #defaultModel = 'gpt-5.4-nano';
+
+    #meteringService: MeteringService;
+
+    #stores: { fsEntry: FSEntryStore; s3Object: S3ObjectStore };
+
+    #fsService: FSService;
+
+    // Sibling Responses-API provider (Azure or OpenAI) used to handle
+    // Responses-only features like web_search. Typed loosely since we only
+    // ever forward `complete()` to it.
+    #responsesProvider: IChatProvider | null = null;
+
+    constructor(
+        meteringService: MeteringService,
+        stores: { fsEntry: FSEntryStore; s3Object: S3ObjectStore },
+        fsService: FSService,
+        config: { apiKey: string; apiURL: string },
+    ) {
+        this.#meteringService = meteringService;
+        this.#stores = stores;
+        this.#fsService = fsService;
+        this.#openAi = new OpenAI({
+            apiKey: config.apiKey,
+            baseURL: config.apiURL,
+        });
+    }
+    checkModeration(_text: string): { flagged: boolean; categories: string[] } {
+        throw new Error('Method not implemented.');
+    }
+
+    // Wired up by the driver after the OpenAI providers are built, so the
+    // Chat Completions path can delegate `web_search` tool calls (Responses-only)
+    // to the OpenAI Responses provider without a circular constructor dependency.
+    setResponsesProvider(provider: IChatProvider): void {
+        this.#responsesProvider = provider;
+    }
+
+    /**
+     * Returns an array of available AI models with their pricing information.
+     * Each model object includes an ID and cost details (currency, tokens, input/output rates).
+     */
+    models() {
+        return AZURE_MODELS.filter((e) => !e.responses_api_only);
+    }
+
+    list() {
+        const models = this.models();
+        const modelNames: string[] = [];
+        for (const model of models) {
+            modelNames.push(model.id);
+            if (model.aliases) {
+                modelNames.push(...model.aliases);
+            }
+        }
+        return modelNames;
+    }
+
+    getDefaultModel() {
+        return this.#defaultModel;
+    }
+
+    async complete(
+        params: ICompleteArguments,
+    ): ReturnType<IChatProvider['complete']> {
+        const {
+            max_tokens,
+            moderation,
+            tools,
+            verbosity,
+            stream,
+            reasoning,
+            reasoning_effort,
+            temperature,
+            text,
+        } = params;
+        let { messages, model } = params;
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        if (tools?.filter((e: any) => e.type === 'web_search').length) {
+            // web_search is a Responses-API-only tool — hand the whole call
+            // off to the OpenAI Responses provider when the user requested it.
+            if (!this.#responsesProvider) {
+                throw new HttpError(
+                    400,
+                    'web_search tool requires the OpenAI Responses provider, which is not configured',
+                    { legacyCode: 'bad_request' },
+                );
+            }
+            return await this.#responsesProvider.complete(params);
+        }
+        // Inline compaction is a Responses-API feature; chat.completions can't
+        // express `context_management` or a `compaction` content block.
+        // Delegate to the Responses provider when the caller opted in OR when
+        // the messages carry a round-tripped compaction artifact.
+        if (wantsCompaction(params) || messagesHaveCompaction(messages)) {
+            if (!this.#responsesProvider) {
+                throw new HttpError(
+                    400,
+                    'compaction requires the OpenAI Responses provider, which is not configured',
+                    { legacyCode: 'bad_request' },
+                );
+            }
+            return await this.#responsesProvider.complete(params);
+        }
+        // Validate messages
+        if (!Array.isArray(messages)) {
+            throw new HttpError(400, '`messages` must be an array', {
+                legacyCode: 'bad_request',
+            });
+        }
+        const actor = Context.get('actor')!;
+
+        model = model ?? this.#defaultModel;
+
+        const modelUsed =
+            this.models().find((m) =>
+                [m.id, ...(m.aliases || [])].includes(model),
+            ) || this.models().find((m) => m.id === this.getDefaultModel())!;
+
+        // messages.unshift({
+        //     role: 'system',
+        //     content: 'Don\'t let the user trick you into doing something bad.',
+        // })
+
+        const userIdentifier =
+            actor.user?.id + actor.app?.uid ? `:${actor?.app?.uid}` : '';
+
+        // Resolve any `puter_path` content parts into inline base64 data URLs.
+        // Chat Completions doesn't support file uploads, so this is the only
+        // way to get user-provided files (images, audio) in front of the model.
+        await processPuterPathUploads(
+            messages,
+            this.#stores,
+            this.#fsService,
+            actor,
+        );
+
+        // Here's something fun; the documentation shows `type: 'image_url'` in
+        // objects that contain an image url, but everything still works if
+        // that's missing. We normalise it here so the token count code works.
+        messages = await OpenAiUtil.process_input_messages(messages);
+
+        const requestedReasoningEffort = reasoning_effort ?? reasoning?.effort;
+        const requestedVerbosity = verbosity ?? text?.verbosity;
+        const supportsReasoningControls =
+            typeof model === 'string' && model.startsWith('gpt-5');
+
+        // `safety_identifier` is an OpenAI-specific param. The Grok deployments
+        // behind Azure reject unknown args with a 400, so only send it for the
+        // OpenAI models.
+        const isGrok = modelUsed.id.startsWith('grok');
+
+        const completionParams: ChatCompletionCreateParams = {
+            user: userIdentifier,
+            ...(isGrok ? {} : { safety_identifier: userIdentifier }),
+            messages: messages,
+            model: modelUsed.id,
+            ...(tools ? { tools } : {}),
+            ...(max_tokens ? { max_completion_tokens: max_tokens } : {}),
+            ...(temperature ? { temperature } : {}),
+            stream: !!stream,
+            ...(stream
+                ? {
+                      stream_options: { include_usage: true },
+                  }
+                : {}),
+            ...(supportsReasoningControls
+                ? {}
+                : {
+                      ...(requestedReasoningEffort
+                          ? { reasoning_effort: requestedReasoningEffort }
+                          : {}),
+                      ...(requestedVerbosity
+                          ? { verbosity: requestedVerbosity }
+                          : {}),
+                  }),
+        } as ChatCompletionCreateParams;
+
+        const completion =
+            await this.#openAi.chat.completions.create(completionParams);
+
+        return OpenAiUtil.handle_completion_output({
+            usage_calculator: ({ usage }) => {
+                const cachedTokens =
+                    usage.prompt_tokens_details?.cached_tokens ?? 0;
+                const trackedUsage = {
+                    // OpenAI includes cached tokens in `prompt_tokens`, so we
+                    // subtract them out to meter the non-cached remainder. Grok
+                    // reports `prompt_tokens` already excluding cached tokens
+                    // (they're additive, not a subset) — subtracting there
+                    // underflows to a negative count, which bills a negative
+                    // (crediting) cost. Match xAI's `extractMeteredUsage` and
+                    // take Grok's `prompt_tokens` as-is.
+                    prompt_tokens: isGrok
+                        ? (usage.prompt_tokens ?? 0)
+                        : (usage.prompt_tokens ?? 0) - cachedTokens,
+                    completion_tokens: usage.completion_tokens ?? 0,
+                    cached_tokens: cachedTokens,
+                };
+
+                const costsOverrideFromModel = Object.fromEntries(
+                    Object.entries(trackedUsage).map(([k, v]) => {
+                        return [k, v * modelUsed.costs[k]];
+                    }),
+                );
+
+                this.#meteringService.utilRecordUsageObject(
+                    trackedUsage,
+                    actor!,
+                    `azure-openai:${modelUsed?.id}`,
+                    costsOverrideFromModel,
+                );
+                return trackedUsage;
+            },
+            stream,
+            completion,
+            moderate: moderation ? this.checkModeration.bind(this) : undefined,
+        });
+    }
+}

src/backend/drivers/ai-chat/providers/azure/AzureResponsesProvider.integration.test.ts

@@ -0,0 +1,83 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+/**
+ * Integration test for the Azure AI Foundry Responses provider.
+ *
+ * Hits the real Azure endpoint with `gpt-5-codex` — a Responses-API-only
+ * (Codex) model that the Chat Completions endpoint rejects. This verifies
+ * the completions/responses split actually routes Codex correctly. Codex is
+ * a reasoning model, so we give it a generous `max_tokens` and low reasoning
+ * effort to leave room for visible output.
+ *
+ * Skipped unless both `PUTER_TEST_AI_AZURE_OPENAI_API_KEY` and
+ * `PUTER_TEST_AI_AZURE_OPENAI_API_URL` are set.
+ */
+
+import { describe, expect, it } from 'vitest';
+import {
+    INTEGRATION_TEST_TIMEOUT_MS,
+    makeMeteringStub,
+    optionalEnv,
+    skipUnlessEnv,
+    withTestActor,
+} from '../../../integrationTestUtil.js';
+import { AzureResponsesProvider } from './AzureResponsesProvider.js';
+
+const KEY_ENV = 'PUTER_TEST_AI_AZURE_OPENAI_API_KEY';
+const URL_ENV = 'PUTER_TEST_AI_AZURE_OPENAI_API_URL';
+
+describe.skipIf(skipUnlessEnv(KEY_ENV) || skipUnlessEnv(URL_ENV))(
+    'AzureResponsesProvider (integration)',
+    () => {
+        it(
+            'returns a non-empty completion from gpt-5-codex',
+            { timeout: INTEGRATION_TEST_TIMEOUT_MS },
+            async () => {
+                const provider = new AzureResponsesProvider(
+                    makeMeteringStub(),
+                    {
+                        fsEntry: undefined as never,
+                        s3Object: undefined as never,
+                    },
+                    undefined as never,
+                    {
+                        apiKey: optionalEnv(KEY_ENV)!,
+                        apiURL: optionalEnv(URL_ENV)!,
+                    },
+                );
+
+                const result = await withTestActor(() =>
+                    provider.complete({
+                        model: 'gpt-5-codex',
+                        messages: [
+                            { role: 'user', content: 'Say hi in one word.' },
+                        ],
+                        max_tokens: 2048,
+                        reasoning: { effort: 'low' },
+                    }),
+                );
+
+                const text = (result as { message?: { content?: string } })
+                    .message?.content;
+                expect(typeof text === 'string' && text.length > 0).toBe(true);
+            },
+        );
+    },
+);

src/backend/drivers/ai-chat/providers/azure/AzureResponsesProvider.ts

@@ -0,0 +1,307 @@
+/*
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import { OpenAI } from 'openai';
+import { ResponseCreateParams } from 'openai/resources/responses/responses.mjs';
+import { Context } from '../../../../core/context.js';
+import type { FSService } from '../../../../services/fs/FSService.js';
+import type { MeteringService } from '../../../../services/metering/MeteringService.js';
+import type { FSEntryStore } from '../../../../stores/fs/FSEntryStore.js';
+import type { S3ObjectStore } from '../../../../stores/fs/S3ObjectStore.js';
+import type { IChatProvider, ICompleteArguments } from '../../types.js';
+import { toOpenAiContextManagement } from '../../utils/compaction.js';
+import * as OpenAiUtil from '../../utils/OpenAIUtil.js';
+import { processPuterPathUploads } from '../openai/fileUpload.js';
+import { AZURE_MODELS } from './models.js';
+import { HttpError } from '@heyputer/backend/src/core/http/HttpError.js';
+
+/**
+ * AzureResponsesProvider serves the Responses-API-only models we expose through
+ * Azure AI Foundry (the Codex family and similar). It mirrors
+ * {@link OpenAiResponsesChatProvider}, but points the OpenAI client at the
+ * configurable Azure endpoint and draws from {@link AZURE_MODELS}.
+ *
+ * Codex / `responses_api_only` models reject the Chat Completions endpoint, so
+ * the sibling {@link AzureChatProvider} (Chat Completions) filters them out and
+ * the driver routes them here instead.
+ *
+ * Billing note: the model `costs` are the standard public OpenAI list prices,
+ * NOT Azure's — Azure is subsidised for us.
+ */
+export class AzureResponsesProvider implements IChatProvider {
+    /**
+     * @type {import('openai').OpenAI}
+     */
+    #openAi: OpenAI;
+
+    #defaultModel = 'gpt-5-codex';
+
+    #meteringService: MeteringService;
+
+    #stores: { fsEntry: FSEntryStore; s3Object: S3ObjectStore };
+
+    #fsService: FSService;
+
+    constructor(
+        meteringService: MeteringService,
+        stores: { fsEntry: FSEntryStore; s3Object: S3ObjectStore },
+        fsService: FSService,
+        config: { apiKey: string; apiURL: string },
+    ) {
+        this.#meteringService = meteringService;
+        this.#stores = stores;
+        this.#fsService = fsService;
+        this.#openAi = new OpenAI({
+            apiKey: config.apiKey,
+            baseURL: config.apiURL,
+        });
+    }
+
+    /**
+     * Returns an array of available AI models with their pricing information.
+     * Each model object includes an ID and cost details (currency, tokens, input/output rates).
+     */
+    models(extra_params?: { no_restrictions?: boolean }) {
+        if (extra_params?.no_restrictions) {
+            return AZURE_MODELS;
+        }
+        return AZURE_MODELS.filter((e) => e.responses_api_only === true);
+    }
+
+    list() {
+        const models = this.models({ no_restrictions: false });
+        const modelNames: string[] = [];
+        for (const model of models) {
+            modelNames.push(model.id);
+            if (model.aliases) {
+                modelNames.push(...model.aliases);
+            }
+        }
+        return modelNames;
+    }
+
+    getDefaultModel() {
+        return this.#defaultModel;
+    }
+
+    async complete({
+        messages,
+        model,
+        max_tokens,
+        moderation,
+        tools,
+        tool_choice,
+        parallel_tool_calls,
+        include,
+        conversation,
+        compaction,
+        context_management,
+        previous_response_id,
+        instructions,
+        metadata,
+        prompt,
+        prompt_cache_key,
+        prompt_cache_retention,
+        store,
+        top_p,
+        truncation,
+        background,
+        service_tier,
+        verbosity,
+        stream,
+        reasoning,
+        reasoning_effort,
+        temperature,
+        text,
+    }: ICompleteArguments): ReturnType<IChatProvider['complete']> {
+        // Validate messages
+        if (!Array.isArray(messages)) {
+            throw new HttpError(400, '`messages` must be an array', {
+                legacyCode: 'bad_request',
+            });
+        }
+        const actor = Context.get('actor');
+
+        model = model ?? this.#defaultModel;
+
+        const modelUsed =
+            this.models({ no_restrictions: true }).find((m) =>
+                [m.id, ...(m.aliases || [])].includes(model),
+            ) ||
+            this.models({ no_restrictions: true }).find(
+                (m) => m.id === this.getDefaultModel(),
+            )!;
+
+        const userIdentifier =
+            actor?.user.id + actor?.app?.uid ? `:${actor?.app?.uid}` : '';
+
+        // Resolve any `puter_path` content parts into inline base64 data URLs
+        // before the Responses API sees them.
+        await processPuterPathUploads(
+            messages,
+            this.#stores,
+            this.#fsService,
+            actor,
+        );
+
+        if (tools) {
+            // Unravel tools to OpenAI Responses API format
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            tools = (tools as any).map((e) => {
+                if (e.type === 'function') {
+                    const tool = e.function;
+                    tool.type = 'function';
+                    return tool;
+                } else {
+                    return e;
+                }
+            });
+        }
+
+        // Here's something fun; the documentation shows `type: 'image_url'` in
+        // objects that contain an image url, but everything still works if
+        // that's missing. We normalise it here so the token count code works.
+        messages =
+            await OpenAiUtil.process_input_messages_responses_api(messages);
+
+        const requestedReasoningEffort = reasoning_effort ?? reasoning?.effort;
+        const requestedVerbosity = verbosity ?? text?.verbosity;
+        const supportsReasoningControls =
+            typeof model === 'string' && model.startsWith('gpt-5');
+
+        // Translate the neutral compaction opt-in (or pass a raw
+        // `context_management` payload through) to OpenAI's Responses shape.
+        const contextManagement = toOpenAiContextManagement({
+            compaction,
+            context_management,
+        });
+
+        const completionParams: ResponseCreateParams = {
+            user: userIdentifier,
+            safety_identifier: userIdentifier,
+            input: messages,
+            model: modelUsed.id,
+            ...(tools ? { tools } : {}),
+            ...(tool_choice !== undefined ? { tool_choice } : {}),
+            ...(parallel_tool_calls !== undefined
+                ? { parallel_tool_calls }
+                : {}),
+            ...(include !== undefined ? { include } : {}),
+            ...(contextManagement !== undefined
+                ? { context_management: contextManagement }
+                : {}),
+            ...(conversation !== undefined ? { conversation } : {}),
+            ...(previous_response_id !== undefined
+                ? { previous_response_id }
+                : {}),
+            ...(instructions !== undefined ? { instructions } : {}),
+            ...(metadata !== undefined ? { metadata } : {}),
+            ...(prompt !== undefined ? { prompt } : {}),
+            ...(prompt_cache_key !== undefined ? { prompt_cache_key } : {}),
+            ...(prompt_cache_retention !== undefined
+                ? { prompt_cache_retention }
+                : {}),
+            ...(store !== undefined ? { store } : {}),
+            ...(max_tokens !== undefined
+                ? { max_output_tokens: max_tokens }
+                : {}),
+            ...(temperature !== undefined ? { temperature } : {}),
+            ...(top_p !== undefined ? { top_p } : {}),
+            ...(truncation !== undefined ? { truncation } : {}),
+            ...(background !== undefined ? { background } : {}),
+            ...(service_tier !== undefined ? { service_tier } : {}),
+            ...(stream !== undefined ? { stream: !!stream } : {}),
+            ...(text !== undefined ? { text } : {}),
+            ...(supportsReasoningControls
+                ? {}
+                : {
+                      ...(requestedReasoningEffort
+                          ? { reasoning_effort: requestedReasoningEffort }
+                          : {}),
+                      ...(requestedVerbosity
+                          ? { verbosity: requestedVerbosity }
+                          : {}),
+                  }),
+            ...(supportsReasoningControls && reasoning ? { reasoning } : {}),
+        } as ResponseCreateParams;
+
+        const completion =
+            await this.#openAi.responses.create(completionParams);
+        return OpenAiUtil.handle_completion_output_responses_api({
+            usage_calculator: ({ usage }) => {
+                const trackedUsage = {
+                    prompt_tokens:
+                        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                        ((usage as any).input_tokens ?? 0) -
+                        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                        ((usage as any).input_tokens_details?.cached_tokens ??
+                            0),
+                    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                    completion_tokens: (usage as any).output_tokens ?? 0,
+                    cached_tokens:
+                        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                        (usage as any).input_tokens_details?.cached_tokens ?? 0,
+                };
+
+                const costsOverrideFromModel = Object.fromEntries(
+                    Object.entries(trackedUsage).map(([k, v]) => {
+                        return [k, v * modelUsed.costs[k]];
+                    }),
+                );
+
+                this.#meteringService.utilRecordUsageObject(
+                    trackedUsage,
+                    actor,
+                    `azure-openai:${modelUsed?.id}`,
+                    costsOverrideFromModel,
+                );
+                return trackedUsage;
+            },
+            stream,
+            completion,
+            moderate: moderation ? this.checkModeration.bind(this) : undefined,
+        });
+    }
+
+    async checkModeration(text: string) {
+        // create moderation
+        const results = await this.#openAi.moderations.create({
+            model: 'omni-moderation-latest',
+            input: text,
+        });
+
+        let flagged = false;
+
+        for (const result of results?.results ?? []) {
+            // OpenAI does a crazy amount of false positives. We filter by their 80% interval
+            const veryFlaggedEntries = Object.entries(
+                result.category_scores,
+            ).filter((e) => e[1] > 0.8);
+            if (veryFlaggedEntries.length > 0) {
+                flagged = true;
+                break;
+            }
+        }
+
+        return {
+            flagged,
+            results,
+        };
+    }
+}

src/backend/drivers/ai-chat/providers/azure/models.ts

@@ -0,0 +1,464 @@
+/**
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import type { IChatModel } from '../../types.js';
+
+// Models served through our Azure AI Foundry deployment. This is NOT just
+// OpenAI — Azure AI also fronts xAI's Grok models — so the list lives in its
+// own provider folder rather than sharing the OpenAI list.
+//
+// IMPORTANT: the `costs` below intentionally mirror the public list prices of
+// the equivalent OpenAI / xAI models (see `../openai/models.ts` and
+// `../xai/models.ts`). Azure is subsidised for us, so our actual spend is
+// lower — but we bill users at the standard model price, which is the whole
+// reason we route through Azure. Do NOT replace these with Azure's own rates.
+//
+// `id` is the Azure deployment name and is what we send upstream.
+export const AZURE_MODELS: IChatModel[] = [
+    // -- xAI Grok (via Azure AI Foundry) -----------------------------------
+    {
+        // Costs mirror xai grok-4-1-fast-non-reasoning.
+        puterId: 'azure:x-ai/grok-4-1-fast-non-reasoning',
+        id: 'grok-4-1-fast-non-reasoning',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2025-11-19',
+        name: 'Grok 4.1 Fast (Non-Reasoning)',
+        aliases: ['x-ai/grok-4-1-fast-non-reasoning'],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 20,
+            completion_tokens: 50,
+            cached_tokens: 5,
+        },
+        max_tokens: 2_000_000,
+    },
+    {
+        // Costs mirror xai grok-4-1-fast (alias grok-4-1-fast-reasoning).
+        puterId: 'azure:x-ai/grok-4-1-fast-reasoning',
+        id: 'grok-4-1-fast-reasoning',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2025-11-19',
+        name: 'Grok 4.1 Fast (Reasoning)',
+        aliases: ['x-ai/grok-4-1-fast-reasoning'],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 20,
+            completion_tokens: 50,
+            cached_tokens: 5,
+        },
+        max_tokens: 2_000_000,
+    },
+    {
+        // Costs mirror xai grok-4.3.
+        puterId: 'azure:x-ai/grok-4.3',
+        id: 'grok-4.3',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        release_date: '2026-05-01',
+        name: 'Grok 4.3',
+        aliases: ['x-ai/grok-4.3'],
+        context: 1_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            completion_tokens: 250,
+            cached_tokens: 20,
+        },
+        max_tokens: 30_000,
+    },
+    {
+        // Costs mirror xai grok-4.20 (grok-4.20-0309-non-reasoning).
+        puterId: 'azure:x-ai/grok-4-20-non-reasoning',
+        id: 'grok-4-20-non-reasoning',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2026-03-09',
+        name: 'Grok 4.20 (Non-Reasoning)',
+        aliases: ['x-ai/grok-4-20-non-reasoning'],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            completion_tokens: 250,
+            cached_tokens: 20,
+        },
+        max_tokens: 30_000,
+    },
+    {
+        // Costs mirror xai grok-4.20 (grok-4.20-0309-reasoning).
+        puterId: 'azure:x-ai/grok-4-20-reasoning',
+        id: 'grok-4-20-reasoning',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2026-03-09',
+        name: 'Grok 4.20 (Reasoning)',
+        aliases: ['x-ai/grok-4-20-reasoning'],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            completion_tokens: 250,
+            cached_tokens: 20,
+        },
+        max_tokens: 30_000,
+    },
+
+    // -- OpenAI (via Azure AI Foundry) -------------------------------------
+    {
+        // Costs mirror openai gpt-5.
+        puterId: 'azure:openai/gpt-5',
+        id: 'gpt-5',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-08-07',
+        aliases: ['openai/gpt-5'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            cached_tokens: 13,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5-codex (same list price as gpt-5).
+        puterId: 'azure:openai/gpt-5-codex',
+        id: 'gpt-5-codex',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-08-07',
+        aliases: ['openai/gpt-5-codex'],
+        // Codex models are Responses-API only on Azure, same as OpenAI.
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            cached_tokens: 13,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5-nano.
+        puterId: 'azure:openai/gpt-5-nano',
+        id: 'gpt-5-nano',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-05-30',
+        release_date: '2025-08-07',
+        aliases: ['openai/gpt-5-nano'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 5,
+            cached_tokens: 1,
+            completion_tokens: 40,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5-mini.
+        puterId: 'azure:openai/gpt-5-mini',
+        id: 'gpt-5-mini',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-05-30',
+        release_date: '2025-08-07',
+        aliases: ['openai/gpt-5-mini'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 25,
+            cached_tokens: 3,
+            completion_tokens: 200,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-4o.
+        puterId: 'azure:openai/gpt-4o',
+        id: 'gpt-4o',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2023-09',
+        release_date: '2024-05-13',
+        aliases: ['openai/gpt-4o'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 250,
+            cached_tokens: 125,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 16384,
+    },
+    {
+        // Costs mirror openai gpt-5.1-codex-mini.
+        puterId: 'azure:openai/gpt-5.1-codex-mini',
+        id: 'gpt-5.1-codex-mini',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-11-13',
+        aliases: ['openai/gpt-5.1-codex-mini'],
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 25,
+            cached_tokens: 3,
+            completion_tokens: 200,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.1.
+        puterId: 'azure:openai/gpt-5.1',
+        id: 'gpt-5.1',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-11-13',
+        aliases: ['openai/gpt-5.1'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            cached_tokens: 13,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.1-codex.
+        puterId: 'azure:openai/gpt-5.1-codex',
+        id: 'gpt-5.1-codex',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-11-13',
+        aliases: ['openai/gpt-5.1-codex'],
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            cached_tokens: 13,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.2.
+        puterId: 'azure:openai/gpt-5.2',
+        id: 'gpt-5.2',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        release_date: '2025-12-11',
+        aliases: ['openai/gpt-5.2'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 175,
+            cached_tokens: 17.5,
+            completion_tokens: 1400,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.2-codex.
+        puterId: 'azure:openai/gpt-5.2-codex',
+        id: 'gpt-5.2-codex',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        release_date: '2025-12-11',
+        aliases: ['openai/gpt-5.2-codex'],
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 175,
+            cached_tokens: 18,
+            completion_tokens: 1400,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.4-nano.
+        puterId: 'azure:openai/gpt-5.4-nano',
+        id: 'gpt-5.4-nano',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        release_date: '2026-03-19',
+        aliases: ['openai/gpt-5.4-nano'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 20,
+            cached_tokens: 2,
+            completion_tokens: 125,
+        },
+        context: 400_000,
+        max_tokens: 128_000,
+    },
+    {
+        // Costs mirror openai gpt-5.4-mini.
+        puterId: 'azure:openai/gpt-5.4-mini',
+        id: 'gpt-5.4-mini',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        aliases: ['openai/gpt-5.4-mini'],
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 75,
+            cached_tokens: 7.5,
+            completion_tokens: 450,
+        },
+        context: 400_000,
+        max_tokens: 128_000,
+    },
+    {
+        // Costs mirror openai gpt-5.3-codex.
+        puterId: 'azure:openai/gpt-5.3-codex',
+        id: 'gpt-5.3-codex',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        aliases: ['openai/gpt-5.3-codex'],
+        responses_api_only: true,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 175,
+            cached_tokens: 17.5,
+            completion_tokens: 1400,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+    },
+    {
+        // Costs mirror openai gpt-5.4.
+        puterId: 'azure:openai/gpt-5.4',
+        id: 'gpt-5.4',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-08-31',
+        release_date: '2026-03-05',
+        aliases: ['openai/gpt-5.4'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 250,
+            cached_tokens: 25,
+            completion_tokens: 1500,
+        },
+        context: 1_050_000,
+        max_tokens: 1_050_000,
+    },
+];

src/backend/drivers/ai-chat/providers/claude/ClaudeProvider.test.ts

@@ -486,6 +486,52 @@ describe('ClaudeProvider.complete non-stream output', () => {
             10 * Number(haiku.costs.cache_read_input_tokens),
         );
     });
+
+    it('bills the compaction pass by summing usage.iterations', async () => {
+        const { provider } = makeProvider();
+        // Per Anthropic: top-level input/output reflect only the message pass;
+        // the compaction pass lives in `iterations` and must be summed to bill.
+        const msg = {
+            content: [{ type: 'text', text: 'done' }],
+            usage: {
+                input_tokens: 23000, // message pass only (NOT the total)
+                output_tokens: 1000,
+                iterations: [
+                    {
+                        type: 'compaction',
+                        input_tokens: 180000,
+                        output_tokens: 3500,
+                    },
+                    { type: 'message', input_tokens: 23000, output_tokens: 1000 },
+                ],
+            },
+        };
+        messagesCreateMock.mockResolvedValueOnce(msg);
+
+        const result = (await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [{ role: 'user', content: 'hi' }],
+                compaction: true,
+            }),
+        )) as { usage: Record<string, number> };
+
+        // Totals are the SUM across iterations, not the top-level fields.
+        expect(result.usage.input_tokens).toBe(203000); // 180000 + 23000
+        expect(result.usage.output_tokens).toBe(4500); // 3500 + 1000
+
+        const haiku = CLAUDE_MODELS.find(
+            (m) => m.id === 'claude-haiku-4-5-20251001',
+        )!;
+        const [usage, , , overrides] = recordSpy.mock.calls[0]!;
+        expect(usage.input_tokens).toBe(203000);
+        expect(overrides.input_tokens).toBe(
+            203000 * Number(haiku.costs.input_tokens),
+        );
+        expect(overrides.output_tokens).toBe(
+            4500 * Number(haiku.costs.output_tokens),
+        );
+    });
 });
 
 // ── Streaming deltas ────────────────────────────────────────────────
@@ -618,6 +664,220 @@ describe('ClaudeProvider.complete streaming', () => {
     });
 });
 
+// ── Inline compaction ───────────────────────────────────────────────
+
+describe('ClaudeProvider.complete compaction', () => {
+    it('translates the neutral opt-in to context_management + the compaction beta', async () => {
+        const { provider } = makeProvider();
+        messagesCreateMock.mockResolvedValueOnce({
+            content: [{ type: 'text', text: 'hi' }],
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [{ role: 'user', content: 'hi' }],
+                compaction: { trigger_tokens: 50000 },
+            }),
+        );
+
+        const [args] = messagesCreateMock.mock.calls[0]!;
+        expect(args.context_management).toEqual({
+            edits: [
+                {
+                    type: 'compact_20260112',
+                    trigger: { type: 'input_tokens', value: 50000 },
+                },
+            ],
+        });
+        expect(args.betas).toContain('compact-2026-01-12');
+    });
+
+    it('emits a canonical compaction event from a streamed compaction block', async () => {
+        const { provider } = makeProvider();
+        messagesStreamMock.mockReturnValueOnce(
+            makeStreamLike(
+                [
+                    { type: 'message_start' },
+                    {
+                        type: 'content_block_start',
+                        content_block: {
+                            type: 'compaction',
+                            id: 'cmpct_1',
+                            content: 'ENC', // Anthropic carries the summary here
+                        },
+                    },
+                    { type: 'content_block_stop' },
+                    {
+                        type: 'message_delta',
+                        usage: { input_tokens: 1, output_tokens: 1 },
+                    },
+                    { type: 'message_stop' },
+                ],
+                { input_tokens: 1, output_tokens: 1 },
+            ),
+        );
+
+        const result = await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [{ role: 'user', content: 'hi' }],
+                stream: true,
+                compaction: true,
+            }),
+        );
+        const harness = makeCapturingChatStream();
+        await (
+            result as {
+                init_chat_stream: (p: { chatStream: unknown }) => Promise<void>;
+            }
+        ).init_chat_stream({ chatStream: harness.chatStream });
+
+        const compaction = harness
+            .events()
+            .find((e) => e.type === 'compaction');
+        expect(compaction).toEqual({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+    });
+
+    it('enables the compaction beta when a round-tripped compaction block is resent (no opt-in)', async () => {
+        const { provider } = makeProvider();
+        messagesCreateMock.mockResolvedValueOnce({
+            content: [{ type: 'text', text: 'ok' }],
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [
+                    { role: 'user', content: 'continue' },
+                    {
+                        role: 'assistant',
+                        content: [
+                            {
+                                type: 'compaction',
+                                id: 'cmpct_1',
+                                encrypted_content: 'ENC',
+                            },
+                        ],
+                    },
+                ],
+                // note: no `compaction`/`context_management` opt-in
+            }),
+        );
+
+        const [args] = messagesCreateMock.mock.calls[0]!;
+        expect(args.betas).toContain('compact-2026-01-12');
+        // No new compaction was requested, so no context_management is sent.
+        expect(args.context_management).toBeUndefined();
+    });
+
+    it('surfaces a compaction block from a non-streaming response as result.compaction', async () => {
+        const { provider } = makeProvider();
+        messagesCreateMock.mockResolvedValueOnce({
+            content: [
+                { type: 'text', text: 'done' },
+                {
+                    type: 'compaction',
+                    id: 'cmpct_2',
+                    content: 'ENC2', // Anthropic carries the summary here
+                },
+            ],
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+
+        const result = (await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [{ role: 'user', content: 'hi' }],
+                compaction: true,
+            }),
+        )) as { compaction?: { id?: string; encrypted_content: string } };
+
+        // Anthropic's `content` is surfaced under the unified `encrypted_content`.
+        expect(result.compaction).toEqual({
+            type: 'compaction',
+            id: 'cmpct_2',
+            encrypted_content: 'ENC2',
+        });
+    });
+
+    it('maps a round-tripped compaction block back to Anthropic `content` on input', async () => {
+        const { provider } = makeProvider();
+        messagesCreateMock.mockResolvedValueOnce({
+            content: [{ type: 'text', text: 'ok' }],
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [
+                    { role: 'user', content: 'continue' },
+                    {
+                        role: 'assistant',
+                        content: [
+                            { type: 'compaction', encrypted_content: 'SUMMARY' },
+                        ],
+                    },
+                ],
+            }),
+        );
+
+        const [args] = messagesCreateMock.mock.calls[0]!;
+        const compactionBlock = args.messages
+            .flatMap((m: { content?: unknown[] }) =>
+                Array.isArray(m.content) ? m.content : [],
+            )
+            .find((c: { type?: string }) => c?.type === 'compaction');
+        // Internal `encrypted_content` carrier → Anthropic native `content`.
+        expect(compactionBlock).toEqual({
+            type: 'compaction',
+            content: 'SUMMARY',
+        });
+    });
+
+    it('preserves the reply text when an assistant turn carries compaction + text', async () => {
+        const { provider } = makeProvider();
+        messagesCreateMock.mockResolvedValueOnce({
+            content: [{ type: 'text', text: 'ok' }],
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'claude-haiku-4-5-20251001',
+                messages: [
+                    {
+                        role: 'assistant',
+                        content: [
+                            { type: 'compaction', encrypted_content: 'SUMMARY' },
+                            { type: 'text', text: 'earlier reply' },
+                        ],
+                    },
+                    { role: 'user', content: 'continue' },
+                ],
+            }),
+        );
+
+        const [args] = messagesCreateMock.mock.calls[0]!;
+        const blocks = args.messages.flatMap((m: { content?: unknown[] }) =>
+            Array.isArray(m.content) ? m.content : [],
+        );
+        // Compaction mapped to Anthropic `content`, AND the reply text kept.
+        expect(blocks).toContainEqual({
+            type: 'compaction',
+            content: 'SUMMARY',
+        });
+        expect(blocks).toContainEqual({ type: 'text', text: 'earlier reply' });
+    });
+});
+
 // ── Moderation ──────────────────────────────────────────────────────
 
 describe('ClaudeProvider.checkModeration', () => {

src/backend/drivers/ai-chat/providers/claude/ClaudeProvider.ts

@@ -34,6 +34,10 @@ import type {
     ICompleteArguments,
     IChatCompleteResult,
 } from '../../types.js';
+import {
+    messagesHaveCompaction,
+    toAnthropicContextManagement,
+} from '../../utils/compaction.js';
 import { make_claude_tools } from '../../utils/FunctionCalling.js';
 import { extract_and_remove_system_messages } from '../../utils/Messages.js';
 import type {
@@ -44,6 +48,11 @@ import type {
 import { FILES_API_BETA, processPuterPathUploads } from './fileUpload.js';
 import { CLAUDE_MODELS } from './models.js';
 
+// Anthropic inline-compaction beta. The vendored SDK (0.68.0) doesn't type the
+// `compact_20260112` edit or the `compaction` content block, so the request
+// params and streamed/returned blocks are handled with `as any` casts.
+const COMPACTION_BETA = 'compact-2026-01-12';
+
 export class ClaudeProvider implements IChatProvider {
     anthropic: Anthropic;
 
@@ -97,9 +106,18 @@ export class ClaudeProvider implements IChatProvider {
         temperature,
         reasoning,
         reasoning_effort,
+        compaction,
+        context_management,
     }: ICompleteArguments): Promise<IChatCompleteResult> {
         tools = make_claude_tools(tools);
 
+        // Translate the neutral compaction opt-in (or pass a raw
+        // `context_management` payload through) to Anthropic's beta shape.
+        const contextManagement = toAnthropicContextManagement({
+            compaction,
+            context_management,
+        });
+
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
         let system_prompts: string | any[];
         [system_prompts, messages] =
@@ -218,6 +236,23 @@ export class ClaudeProvider implements IChatProvider {
             return message;
         });
 
+        // Map round-tripped compaction blocks back to Anthropic's native shape.
+        // The internal/unified carrier field is `encrypted_content`; Anthropic's
+        // compaction block uses `content` (a plaintext summary).
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        messages = messages.map((message: any) => {
+            if (!Array.isArray(message.content)) return message;
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            message.content = message.content.map((part: any) => {
+                if (part?.type !== 'compaction') return part;
+                return {
+                    type: 'compaction',
+                    content: part.content ?? part.encrypted_content ?? '',
+                };
+            });
+            return message;
+        });
+
         const modelUsed =
             this.models().find((m) =>
                 [m.id, ...(m.aliases || [])].includes(model),
@@ -229,9 +264,10 @@ export class ClaudeProvider implements IChatProvider {
             reasoningEffort: requestedReasoningEffort,
             maxTokens: max_tokens,
         });
-        // Opus 4.7 and 4.8 error on non-default sampling params; omit temperature entirely.
+        // Fable 5 and Opus 4.7/4.8 error on non-default sampling params; omit temperature entirely.
         // Other models require temperature=1 when thinking is enabled.
         const omitsTemperature = [
+            'claude-fable-5',
             'claude-opus-4-7',
             'claude-opus-4-8',
         ].includes(modelUsed.id);
@@ -241,6 +277,8 @@ export class ClaudeProvider implements IChatProvider {
               ? 1
               : (temperature ?? 0);
         const supportsEffort = [
+            'claude-fable-5',
+            'claude-opus-4-8',
             'claude-opus-4-7',
             'claude-opus-4-6',
             'claude-sonnet-4-6',
@@ -259,6 +297,18 @@ export class ClaudeProvider implements IChatProvider {
             actor,
         );
         const usesBetaFiles = uploadedFileIds.length > 0;
+        // The compaction beta is needed both to *request* compaction
+        // (contextManagement) and to *accept a round-tripped* compaction block
+        // back as input (messagesHaveCompaction).
+        const usesCompaction =
+            !!contextManagement || messagesHaveCompaction(messages);
+        // Compaction and Files API both require the beta endpoint; combine their
+        // beta headers and route through `beta.messages.*` if either is active.
+        const betas = [
+            ...(usesBetaFiles ? [FILES_API_BETA] : []),
+            ...(usesCompaction ? [COMPACTION_BETA] : []),
+        ];
+        const usesBeta = betas.length > 0;
 
         const sdkParams: MessageCreateParams & {
             betas?: string[];
@@ -288,7 +338,11 @@ export class ClaudeProvider implements IChatProvider {
             ...(supportsEffort && requestedReasoningEffort
                 ? { output_config: { effort: requestedReasoningEffort } }
                 : {}),
-            ...(usesBetaFiles ? { betas: [FILES_API_BETA] } : {}),
+            // Cast: `context_management` compaction edits aren't typed in SDK 0.68.0.
+            ...(contextManagement
+                ? { context_management: contextManagement as any }
+                : {}),
+            ...(usesBeta ? { betas } : {}),
         } as MessageCreateParams & { betas?: string[] };
 
         const cleanupUploads = async () => {
@@ -312,13 +366,23 @@ export class ClaudeProvider implements IChatProvider {
             }: {
                 chatStream: AIChatStream;
             }) => {
-                const completion = usesBetaFiles
+                const completion = usesBeta
                     ? this.anthropic.beta.messages.stream(sdkParams)
                     : this.anthropic.messages.stream(sdkParams);
                 const usageSum: Record<string, number> = {};
 
                 let message, contentBlock;
                 let currentContentBlockType: string | null = null;
+                // Inline-compaction block is an untyped beta block; capture its
+                // artifact across start/delta and emit on stop. Anthropic carries
+                // the summary in `content` (plaintext) — unlike OpenAI's
+                // `encrypted_content` — so read `content` first.
+                let compactionData: {
+                    id?: string;
+                    payload: string;
+                    buffer: string;
+                } | null = null;
+                let emittedCompaction = false;
                 for await (const event of completion) {
                     if (event.type === 'message_delta') {
                         const meteredData = this.#usageFormatterUtil(
@@ -342,6 +406,24 @@ export class ClaudeProvider implements IChatProvider {
                     }
                     if (event.type === 'content_block_start') {
                         currentContentBlockType = event.content_block.type;
+                        if (
+                            (event.content_block.type as string) ===
+                            'compaction'
+                        ) {
+                            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                            const block = event.content_block as any;
+                            compactionData = {
+                                id: block.id,
+                                // Anthropic uses `content`; fall back to
+                                // `encrypted_content` in case the field varies.
+                                payload:
+                                    block.content ??
+                                    block.encrypted_content ??
+                                    '',
+                                buffer: '',
+                            };
+                            continue;
+                        }
                         if (event.content_block.type === 'tool_use') {
                             contentBlock = message!.contentBlock({
                                 type: event.content_block.type,
@@ -360,12 +442,42 @@ export class ClaudeProvider implements IChatProvider {
                         continue;
                     }
                     if (event.type === 'content_block_stop') {
+                        if (currentContentBlockType === 'compaction') {
+                            const encrypted_content =
+                                compactionData?.payload ||
+                                compactionData?.buffer ||
+                                '';
+                            // Only emit (and mark done) if we actually captured
+                            // the summary; otherwise let the finalMessage
+                            // fallback recover it from the complete block.
+                            if (encrypted_content) {
+                                chatStream.compaction({
+                                    id: compactionData?.id,
+                                    encrypted_content,
+                                });
+                                emittedCompaction = true;
+                            }
+                            compactionData = null;
+                            currentContentBlockType = null;
+                            continue;
+                        }
                         contentBlock!.end();
                         contentBlock = null;
                         currentContentBlockType = null;
                         continue;
                     }
                     if (event.type === 'content_block_delta') {
+                        if (currentContentBlockType === 'compaction') {
+                            // Capture any streamed payload for the artifact.
+                            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                            const d = event.delta as any;
+                            const chunk =
+                                d.partial_json ?? d.text ?? d.data ?? '';
+                            if (typeof chunk === 'string' && compactionData) {
+                                compactionData.buffer += chunk;
+                            }
+                            continue;
+                        }
                         if (event.delta.type === 'input_json_delta') {
                             (contentBlock as AIChatToolUseStream)!.addPartialJSON(
                                 event.delta.partial_json,
@@ -388,18 +500,37 @@ export class ClaudeProvider implements IChatProvider {
                         // signature_delta — ignored
                     }
                 }
-                const finalUsage = await completion
+                const finalMessage = await completion
                     .finalMessage()
-                    .then((msg) =>
-                        this.#usageFormatterUtil(
-                            msg.usage as Usage | BetaUsage,
-                        ),
-                    )
                     .catch(() => null);
-                if (finalUsage) {
+                if (finalMessage) {
+                    const finalUsage = this.#usageFormatterUtil(
+                        finalMessage.usage as Usage | BetaUsage,
+                    );
                     for (const [key, value] of Object.entries(finalUsage)) {
                         usageSum[key] = value;
                     }
+                    // Fallback: some SDK versions surface the compaction block
+                    // only in the final message, not as a streamed block.
+                    if (!emittedCompaction) {
+                        const block = (
+                            (finalMessage.content as unknown[]) ?? []
+                        ).find(
+                            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                            (c: any) => c?.type === 'compaction',
+                            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                        ) as any;
+                        if (block) {
+                            chatStream.compaction({
+                                id: block.id,
+                                encrypted_content:
+                                    block.content ??
+                                    block.encrypted_content ??
+                                    '',
+                            });
+                            emittedCompaction = true;
+                        }
+                    }
                 }
                 chatStream.end(usageSum);
                 const costsOverrideFromModel =
@@ -420,7 +551,7 @@ export class ClaudeProvider implements IChatProvider {
         }
 
         try {
-            const msg = await (usesBetaFiles
+            const msg = await (usesBeta
                 ? this.anthropic.beta.messages.create(sdkParams)
                 : this.anthropic.messages.create(sdkParams));
             const usage = this.#usageFormatterUtil(
@@ -437,7 +568,37 @@ export class ClaudeProvider implements IChatProvider {
                 costsOverrideFromModel,
             );
 
-            return { message: msg, usage, finish_reason: 'stop' };
+            // Surface any inline-compaction artifact for stateless round-trip.
+            const compactionBlock = (
+                ((msg as Message).content as unknown[]) ?? []
+            ).find(
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                (c: any) => c?.type === 'compaction',
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            ) as any;
+
+            return {
+                message: msg,
+                usage,
+                finish_reason: 'stop',
+                ...(compactionBlock
+                    ? {
+                          compaction: {
+                              // `type` makes the artifact a drop-in `messages`
+                              // item for the round-trip (symmetric with the
+                              // streaming compaction chunk).
+                              type: 'compaction' as const,
+                              ...(compactionBlock.id !== undefined
+                                  ? { id: compactionBlock.id }
+                                  : {}),
+                              encrypted_content:
+                                  compactionBlock.content ??
+                                  compactionBlock.encrypted_content ??
+                                  '',
+                          },
+                      }
+                    : {}),
+            };
         } finally {
             await cleanupUploads();
         }
@@ -445,8 +606,31 @@ export class ClaudeProvider implements IChatProvider {
 
     // eslint-disable-next-line @typescript-eslint/no-explicit-any
     #usageFormatterUtil(usage: any) {
+        // Compaction responses report per-pass usage in `usage.iterations` (the
+        // compaction pass + the message pass). Per Anthropic's docs the
+        // top-level `input_tokens`/`output_tokens` reflect ONLY the
+        // non-compaction passes, so to bill the (often large) compaction pass we
+        // must sum across all iterations. Compaction is billed at normal token
+        // rates, and iterations don't break out cache fields — those stay
+        // top-level (the message pass's), which bills the compaction input at
+        // the full input rate.
+        const iterations = Array.isArray(usage?.iterations)
+            ? usage.iterations
+            : null;
+        const inputTokens = iterations
+            ? iterations.reduce(
+                  (sum: number, it: any) => sum + (it?.input_tokens || 0),
+                  0,
+              )
+            : usage?.input_tokens || 0;
+        const outputTokens = iterations
+            ? iterations.reduce(
+                  (sum: number, it: any) => sum + (it?.output_tokens || 0),
+                  0,
+              )
+            : usage?.output_tokens || 0;
         return {
-            input_tokens: usage?.input_tokens || 0,
+            input_tokens: inputTokens,
             ephemeral_5m_input_tokens:
                 usage?.cache_creation?.ephemeral_5m_input_tokens ||
                 usage?.cache_creation_input_tokens ||
@@ -454,7 +638,7 @@ export class ClaudeProvider implements IChatProvider {
             ephemeral_1h_input_tokens:
                 usage?.cache_creation?.ephemeral_1h_input_tokens || 0,
             cache_read_input_tokens: usage?.cache_read_input_tokens || 0,
-            output_tokens: usage?.output_tokens || 0,
+            output_tokens: outputTokens,
             thinking_tokens:
                 usage?.thinking_tokens ||
                 usage?.output_tokens_details?.thinking_tokens ||
@@ -489,11 +673,16 @@ export class ClaudeProvider implements IChatProvider {
     }) {
         if (!reasoningEffort) return undefined;
 
-        // Opus 4.7, 4.6, and Sonnet 4.6 use adaptive thinking
-        // (`budget_tokens` is deprecated on 4.6/Sonnet 4.6, removed on 4.7).
-        // Opus 4.7 omits thinking content by default; `display: 'summarized'`
-        // restores visible reasoning in the stream.
-        if (modelId === 'claude-opus-4-7') {
+        // Fable 5, Opus 4.7/4.8, 4.6, and Sonnet 4.6 use adaptive thinking
+        // (`budget_tokens` is deprecated on 4.6/Sonnet 4.6, removed on
+        // Fable 5 and 4.7+). Fable 5 and Opus 4.7/4.8 omit thinking content
+        // by default; `display: 'summarized'` restores visible reasoning in
+        // the stream.
+        if (
+            modelId === 'claude-fable-5' ||
+            modelId === 'claude-opus-4-8' ||
+            modelId === 'claude-opus-4-7'
+        ) {
             return {
                 type: 'adaptive' as const,
                 display: 'summarized' as const,

src/backend/drivers/ai-chat/providers/claude/models.ts

@@ -21,6 +21,35 @@ import type { IChatModel } from '../../types.js';
 
 // Hardcoded from https://models.dev/api.json
 export const CLAUDE_MODELS: IChatModel[] = [
+    {
+        puterId: 'anthropic:anthropic/claude-fable-5',
+        id: 'claude-fable-5',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        release_date: '2026-06-09',
+        aliases: [
+            'claude-fable',
+            'claude-fable-latest',
+            'claude-fable-5-latest',
+            'claude-fable-5',
+            'anthropic/claude-fable-5',
+        ],
+        name: 'Claude Fable 5',
+        costs_currency: 'usd-cents',
+        input_cost_key: 'input_tokens',
+        output_cost_key: 'output_tokens',
+        costs: {
+            tokens: 1_000_000,
+            input_tokens: 1000,
+            ephemeral_5m_input_tokens: 1000 * 1.25,
+            ephemeral_1h_input_tokens: 1000 * 2,
+            cache_read_input_tokens: 1000 * 0.1,
+            output_tokens: 5000,
+        },
+        context: 1000000,
+        max_tokens: 128000,
+    },
     {
         puterId: 'anthropic:anthropic/claude-opus-4-8',
         id: 'claude-opus-4-8',

src/backend/drivers/ai-chat/providers/gemini/GeminiChatProvider.test.ts

@@ -355,6 +355,8 @@ describe('GeminiChatProvider.complete non-stream output', () => {
             prompt_tokens: 90,
             completion_tokens: 50,
             cached_tokens: 10,
+            thinking_tokens: 0,
+            grounding_requests: 0,
         });
 
         // gemini-2.5-flash costs: prompt=30, completion=250, cached=3.
@@ -367,11 +369,15 @@ describe('GeminiChatProvider.complete non-stream output', () => {
             prompt_tokens: 90,
             completion_tokens: 50,
             cached_tokens: 10,
+            thinking_tokens: 0,
+            grounding_requests: 0,
         });
         expect(overrides).toEqual({
             prompt_tokens: 90 * Number(flash.costs.prompt_tokens),
             completion_tokens: 50 * Number(flash.costs.completion_tokens),
             cached_tokens: 10 * Number(flash.costs.cached_tokens ?? 0),
+            thinking_tokens: 0,
+            grounding_requests: 0,
         });
     });
 
@@ -446,6 +452,8 @@ describe('GeminiChatProvider.complete streaming', () => {
             prompt_tokens: 3, // 4 - 1 cached
             completion_tokens: 2,
             cached_tokens: 1,
+            thinking_tokens: 0,
+            grounding_requests: 0,
         });
 
         const flash = GEMINI_MODELS.find((m) => m.id === 'gemini-2.5-flash')!;
@@ -456,10 +464,192 @@ describe('GeminiChatProvider.complete streaming', () => {
             prompt_tokens: 3 * Number(flash.costs.prompt_tokens),
             completion_tokens: 2 * Number(flash.costs.completion_tokens),
             cached_tokens: 1 * Number(flash.costs.cached_tokens ?? 0),
+            thinking_tokens: 0,
+            grounding_requests: 0,
         });
     });
 });
 
+// ── Thinking-token metering ─────────────────────────────────────────
+
+describe('GeminiChatProvider.complete thinking token metering', () => {
+    it('splits thinking tokens out of completion_tokens and bills each at its own rate', async () => {
+        const { provider } = makeProvider();
+        createMock.mockResolvedValueOnce({
+            choices: [
+                {
+                    message: { content: 'result', role: 'assistant' },
+                    finish_reason: 'stop',
+                },
+            ],
+            usage: {
+                prompt_tokens: 100,
+                completion_tokens: 60,
+                completion_tokens_details: { reasoning_tokens: 20 },
+                prompt_tokens_details: { cached_tokens: 10 },
+            },
+        });
+
+        const result = await withTestActor(() =>
+            provider.complete({
+                model: 'gemini-2.5-flash',
+                messages: [{ role: 'user', content: 'think hard' }],
+            }),
+        );
+
+        expect((result as { usage: unknown }).usage).toEqual({
+            prompt_tokens: 90,       // 100 - 10 cached
+            completion_tokens: 40,   // 60 - 20 thinking
+            cached_tokens: 10,
+            thinking_tokens: 20,
+            grounding_requests: 0,
+        });
+
+        const flash = GEMINI_MODELS.find((m) => m.id === 'gemini-2.5-flash')!;
+        expect(recordSpy).toHaveBeenCalledTimes(1);
+        const [usage, , , overrides] = recordSpy.mock.calls[0]!;
+        expect(usage.thinking_tokens).toBe(20);
+        expect(usage.completion_tokens).toBe(40);
+        expect(overrides).toMatchObject({
+            thinking_tokens: 20 * Number(flash.costs.thinking_tokens),
+            completion_tokens: 40 * Number(flash.costs.completion_tokens),
+        });
+    });
+
+    it('sets thinking_tokens to 0 when completion_tokens_details is absent', async () => {
+        const { provider } = makeProvider();
+        createMock.mockResolvedValueOnce({
+            choices: [
+                {
+                    message: { content: 'ok', role: 'assistant' },
+                    finish_reason: 'stop',
+                },
+            ],
+            usage: { prompt_tokens: 5, completion_tokens: 3 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'gemini-2.5-flash',
+                messages: [{ role: 'user', content: 'hi' }],
+            }),
+        );
+
+        const [usage] = recordSpy.mock.calls[0]!;
+        expect(usage.thinking_tokens).toBe(0);
+        expect(usage.completion_tokens).toBe(3);
+    });
+});
+
+// ── Grounding-request metering ───────────────────────────────────────
+
+describe('GeminiChatProvider.complete grounding request metering', () => {
+    it('charges one grounding request when non-stream response contains grounding_metadata', async () => {
+        const { provider } = makeProvider();
+        createMock.mockResolvedValueOnce({
+            choices: [
+                {
+                    message: {
+                        content: 'result',
+                        role: 'assistant',
+                        extra_content: {
+                            grounding_metadata: { web_search_queries: ['foo'] },
+                        },
+                    },
+                    finish_reason: 'stop',
+                },
+            ],
+            usage: { prompt_tokens: 10, completion_tokens: 5 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'gemini-2.5-flash',
+                messages: [{ role: 'user', content: 'search for foo' }],
+            }),
+        );
+
+        const flash = GEMINI_MODELS.find((m) => m.id === 'gemini-2.5-flash')!;
+        const [usage, , , overrides] = recordSpy.mock.calls[0]!;
+        expect(usage.grounding_requests).toBe(1);
+        expect(overrides!.grounding_requests).toBe(
+            1 * Number(flash.costs.grounding_requests),
+        );
+    });
+
+    it('does not charge a grounding request when no grounding_metadata is present', async () => {
+        const { provider } = makeProvider();
+        createMock.mockResolvedValueOnce({
+            choices: [
+                {
+                    message: { content: 'ok', role: 'assistant' },
+                    finish_reason: 'stop',
+                },
+            ],
+            usage: { prompt_tokens: 5, completion_tokens: 3 },
+        });
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'gemini-2.5-flash',
+                messages: [{ role: 'user', content: 'hi' }],
+            }),
+        );
+
+        const [usage] = recordSpy.mock.calls[0]!;
+        expect(usage.grounding_requests).toBe(0);
+    });
+
+    it('charges one grounding request when streaming response contains grounding extra_content', async () => {
+        const { provider } = makeProvider();
+        createMock.mockReturnValueOnce(
+            asAsyncIterable([
+                { choices: [{ delta: { content: 'ans' } }] },
+                {
+                    choices: [
+                        {
+                            delta: {
+                                extra_content: {
+                                    grounding_metadata: {
+                                        web_search_queries: ['bar'],
+                                    },
+                                },
+                            },
+                        },
+                    ],
+                },
+                {
+                    choices: [{ delta: {} }],
+                    usage: { prompt_tokens: 8, completion_tokens: 4 },
+                },
+            ]),
+        );
+
+        const result = await withTestActor(() =>
+            provider.complete({
+                model: 'gemini-2.5-flash',
+                messages: [{ role: 'user', content: 'search bar' }],
+                stream: true,
+            }),
+        );
+
+        const harness = makeCapturingChatStream();
+        await (
+            result as {
+                init_chat_stream: (p: { chatStream: unknown }) => Promise<void>;
+            }
+        ).init_chat_stream({ chatStream: harness.chatStream });
+
+        const flash = GEMINI_MODELS.find((m) => m.id === 'gemini-2.5-flash')!;
+        expect(recordSpy).toHaveBeenCalledTimes(1);
+        const [usage, , , overrides] = recordSpy.mock.calls[0]!;
+        expect(usage.grounding_requests).toBe(1);
+        expect(overrides!.grounding_requests).toBe(
+            1 * Number(flash.costs.grounding_requests),
+        );
+    });
+});
+
 // ── Error mapping ───────────────────────────────────────────────────
 
 describe('GeminiChatProvider.complete error mapping', () => {

src/backend/drivers/ai-chat/providers/gemini/GeminiChatProvider.ts

@@ -102,24 +102,52 @@ export class GeminiChatProvider implements IChatProvider {
         }
 
         return handle_completion_output({
-            usage_calculator: ({ usage }) => {
+            usage_calculator: (args) => {
+                // Cast to access Gemini-specific extras passed alongside usage:
+                // - choices: non-stream grounding metadata lives in choices[0].message.extra_content
+                // - extra_content: streaming grounding metadata accumulated by the stream handler
+                const { usage, choices, extra_content } = args as {
+                    usage: typeof args.usage;
+                    choices?: Array<{
+                        message?: {
+                            extra_content?: { grounding_metadata?: unknown };
+                        };
+                    }>;
+                    extra_content?: { grounding_metadata?: unknown };
+                };
+
+                const cached_tokens =
+                    usage?.prompt_tokens_details?.cached_tokens ?? 0;
+
+                // Thinking tokens are a subset of completion_tokens billed at a different rate
+                const thinking_tokens =
+                    usage?.completion_tokens_details?.reasoning_tokens ?? 0;
+
                 const trackedUsage = {
-                    prompt_tokens:
-                        (usage.prompt_tokens ?? 0) -
-                        (usage.prompt_tokens_details?.cached_tokens ?? 0),
-                    completion_tokens: usage.completion_tokens ?? 0,
-                    cached_tokens:
-                        usage.prompt_tokens_details?.cached_tokens ?? 0,
+                    prompt_tokens: (usage?.prompt_tokens ?? 0) - cached_tokens,
+                    completion_tokens: Math.max(
+                        0,
+                        (usage?.completion_tokens ?? 0) - thinking_tokens,
+                    ),
+                    cached_tokens,
+                    thinking_tokens,
+                    // Grounding search is a per-request fee not reflected in token counts
+                    grounding_requests:
+                        (choices?.[0]?.message?.extra_content
+                            ?.grounding_metadata ??
+                        extra_content?.grounding_metadata)
+                            ? 1
+                            : 0,
                 };
 
                 const costsOverrideFromModel = Object.fromEntries(
                     Object.entries(trackedUsage).map(([k, v]) => {
-                        return [k, v * modelUsed.costs[k]];
+                        return [k, v * (modelUsed.costs[k] ?? 0)];
                     }),
                 );
                 this.meteringService.utilRecordUsageObject(
                     trackedUsage,
-                    actor,
+                    actor!,
                     `gemini:${modelUsed?.id}`,
                     costsOverrideFromModel,
                 );

src/backend/drivers/ai-chat/providers/gemini/models.ts

@@ -43,7 +43,10 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 150,
             completion_tokens: 900,
+            thinking_tokens: 900,
             cached_tokens: 15,
+            // Gemini 3.x grounding is $14 / 1,000 requests
+            grounding_requests: 1_400_000,
         },
     },
     {
@@ -68,6 +71,8 @@ export const GEMINI_MODELS: IChatModel[] = [
             prompt_tokens: 10,
             completion_tokens: 40,
             cached_tokens: 3,
+            // Gemini 2.x grounding is $35 / 1,000 requests
+            grounding_requests: 3_500_000,
         },
         max_tokens: 8192,
     },
@@ -115,8 +120,13 @@ export const GEMINI_MODELS: IChatModel[] = [
         costs: {
             tokens: 1_000_000,
             prompt_tokens: 30,
+            // Output is $2.50/M; thinking tokens bill at the same output rate
             completion_tokens: 250,
+            thinking_tokens: 250,
+            // Cache read is $0.03/M (10% of input)
             cached_tokens: 3,
+            // Gemini 2.x grounding is $35 / 1,000 requests
+            grounding_requests: 3_500_000,
         },
         max_tokens: 65536,
     },
@@ -141,7 +151,10 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 10,
             completion_tokens: 40,
+            thinking_tokens: 40,
             cached_tokens: 1,
+            // Gemini 2.x grounding is $35 / 1,000 requests
+            grounding_requests: 3_500_000,
         },
         max_tokens: 65536,
     },
@@ -166,7 +179,10 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 125,
             completion_tokens: 1000,
-            cached_tokens: 13,
+            thinking_tokens: 1000,
+            cached_tokens: 31,
+            // Gemini 2.x grounding is $35 / 1,000 requests
+            grounding_requests: 3_500_000,
         },
         max_tokens: 200_000,
     },
@@ -191,7 +207,9 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 200,
             completion_tokens: 1200,
+            thinking_tokens: 1200,
             cached_tokens: 20,
+            grounding_requests: 1_400_000,
         },
         max_tokens: 65536,
     },
@@ -216,7 +234,9 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 50,
             completion_tokens: 300,
+            thinking_tokens: 300,
             cached_tokens: 5,
+            grounding_requests: 1_400_000,
         },
         max_tokens: 65536,
     },
@@ -245,7 +265,9 @@ export const GEMINI_MODELS: IChatModel[] = [
             tokens: 1_000_000,
             prompt_tokens: 25,
             completion_tokens: 150,
+            thinking_tokens: 150,
             cached_tokens: 2.5,
+            grounding_requests: 1_400_000,
         },
         max_tokens: 65536,
     },

src/backend/drivers/ai-chat/providers/minimax/MiniMaxProvider.test.ts

@@ -143,6 +143,11 @@ describe('MiniMaxProvider model catalog', () => {
 
     it('uses MiniMax completion-token limits instead of the context window', () => {
         for (const model of MINIMAX_MODELS) {
+            if (model.id === 'minimax-m3') {
+                expect(model.context).toBe(1_048_576);
+                expect(model.max_tokens).toBe(512_000);
+                continue;
+            }
             expect(model.context).toBe(204_800);
             expect(model.max_tokens).toBe(196_608);
         }

src/backend/drivers/ai-chat/providers/minimax/models.ts

@@ -18,60 +18,215 @@
  */
 
 import type { IChatModel } from '../../types.js';
-import { usdPerMToken } from '../../utils/pricing.js';
-
-const CONTEXT_WINDOW = 204_800;
-const MAX_OUTPUT_TOKENS = 196_608;
 
 type MiniMaxChatModel = IChatModel & {
     apiModel: string;
 };
 
-const minimaxModel = (
-    apiModel: string,
-    name: string,
-    costs: IChatModel['costs'],
-): MiniMaxChatModel => {
-    const id = apiModel.toLowerCase();
-    return {
-        puterId: `minimax:minimax/${id}`,
-        id,
-        apiModel,
-        name,
-        aliases: [`minimax/${id}`, apiModel, `minimax/${apiModel}`],
-        modalities: { input: ['text'], output: ['text'] },
-        open_weights: false,
-        tool_call: true,
-        context: CONTEXT_WINDOW,
-        max_tokens: MAX_OUTPUT_TOKENS,
-        costs_currency: 'usd-cents',
-        input_cost_key: 'prompt_tokens',
-        output_cost_key: 'completion_tokens',
-        costs,
-    };
-};
-
 // Hardcoded from MiniMax OpenAI-compatible API docs and pay-as-you-go pricing:
 // https://platform.minimax.io/docs/api-reference/text-openai-api
 // https://platform.minimax.io/docs/guides/pricing-paygo
 export const MINIMAX_MODELS: MiniMaxChatModel[] = [
-    minimaxModel('MiniMax-M2.7', 'MiniMax M2.7', usdPerMToken(0.3, 1.2, 0.06)),
-    minimaxModel(
-        'MiniMax-M2.7-highspeed',
-        'MiniMax M2.7 Highspeed',
-        usdPerMToken(0.6, 2.4, 0.06),
-    ),
-    minimaxModel('MiniMax-M2.5', 'MiniMax M2.5', usdPerMToken(0.3, 1.2, 0.03)),
-    minimaxModel(
-        'MiniMax-M2.5-highspeed',
-        'MiniMax M2.5 Highspeed',
-        usdPerMToken(0.6, 2.4, 0.03),
-    ),
-    minimaxModel('MiniMax-M2.1', 'MiniMax M2.1', usdPerMToken(0.3, 1.2, 0.03)),
-    minimaxModel(
-        'MiniMax-M2.1-highspeed',
-        'MiniMax M2.1 Highspeed',
-        usdPerMToken(0.6, 2.4, 0.03),
-    ),
-    minimaxModel('MiniMax-M2', 'MiniMax M2', usdPerMToken(0.3, 1.2, 0.03)),
+    // -- MiniMax M3 (Flagship) --------------------------------------
+    // MiniMax Sparse Attention, 1M context, native multimodal input.
+    {
+        puterId: 'minimax:minimax/minimax-m3',
+        id: 'minimax-m3',
+        apiModel: 'MiniMax-M3',
+        name: 'MiniMax M3',
+        aliases: ['minimax/minimax-m3', 'MiniMax-M3', 'minimax/MiniMax-M3'],
+        modalities: { input: ['text', 'image', 'video'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 1_048_576,
+        max_tokens: 512_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 30, // $0.30 per 1M (input <= 512k tokens)
+            completion_tokens: 120, // $1.20 per 1M
+            cached_tokens: 6, // $0.06 per 1M
+        },
+    },
+
+    // -- MiniMax M2.7 -----------------------------------------------
+    {
+        puterId: 'minimax:minimax/minimax-m2.7',
+        id: 'minimax-m2.7',
+        apiModel: 'MiniMax-M2.7',
+        name: 'MiniMax M2.7',
+        aliases: [
+            'minimax/minimax-m2.7',
+            'MiniMax-M2.7',
+            'minimax/MiniMax-M2.7',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 30, // $0.30 per 1M
+            completion_tokens: 120, // $1.20 per 1M
+            cached_tokens: 6, // $0.06 per 1M
+        },
+    },
+    {
+        puterId: 'minimax:minimax/minimax-m2.7-highspeed',
+        id: 'minimax-m2.7-highspeed',
+        apiModel: 'MiniMax-M2.7-highspeed',
+        name: 'MiniMax M2.7 Highspeed',
+        aliases: [
+            'minimax/minimax-m2.7-highspeed',
+            'MiniMax-M2.7-highspeed',
+            'minimax/MiniMax-M2.7-highspeed',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 60, // $0.60 per 1M
+            completion_tokens: 240, // $2.40 per 1M
+            cached_tokens: 6, // $0.06 per 1M
+        },
+    },
+
+    // -- MiniMax M2.5 -----------------------------------------------
+    {
+        puterId: 'minimax:minimax/minimax-m2.5',
+        id: 'minimax-m2.5',
+        apiModel: 'MiniMax-M2.5',
+        name: 'MiniMax M2.5',
+        aliases: [
+            'minimax/minimax-m2.5',
+            'MiniMax-M2.5',
+            'minimax/MiniMax-M2.5',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 30, // $0.30 per 1M
+            completion_tokens: 120, // $1.20 per 1M
+            cached_tokens: 3, // $0.03 per 1M
+        },
+    },
+    {
+        puterId: 'minimax:minimax/minimax-m2.5-highspeed',
+        id: 'minimax-m2.5-highspeed',
+        apiModel: 'MiniMax-M2.5-highspeed',
+        name: 'MiniMax M2.5 Highspeed',
+        aliases: [
+            'minimax/minimax-m2.5-highspeed',
+            'MiniMax-M2.5-highspeed',
+            'minimax/MiniMax-M2.5-highspeed',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 60, // $0.60 per 1M
+            completion_tokens: 240, // $2.40 per 1M
+            cached_tokens: 3, // $0.03 per 1M
+        },
+    },
+
+    // -- MiniMax M2.1 -----------------------------------------------
+    {
+        puterId: 'minimax:minimax/minimax-m2.1',
+        id: 'minimax-m2.1',
+        apiModel: 'MiniMax-M2.1',
+        name: 'MiniMax M2.1',
+        aliases: [
+            'minimax/minimax-m2.1',
+            'MiniMax-M2.1',
+            'minimax/MiniMax-M2.1',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 30, // $0.30 per 1M
+            completion_tokens: 120, // $1.20 per 1M
+            cached_tokens: 3, // $0.03 per 1M
+        },
+    },
+    {
+        puterId: 'minimax:minimax/minimax-m2.1-highspeed',
+        id: 'minimax-m2.1-highspeed',
+        apiModel: 'MiniMax-M2.1-highspeed',
+        name: 'MiniMax M2.1 Highspeed',
+        aliases: [
+            'minimax/minimax-m2.1-highspeed',
+            'MiniMax-M2.1-highspeed',
+            'minimax/MiniMax-M2.1-highspeed',
+        ],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 60, // $0.60 per 1M
+            completion_tokens: 240, // $2.40 per 1M
+            cached_tokens: 3, // $0.03 per 1M
+        },
+    },
+
+    // -- MiniMax M2 -------------------------------------------------
+    {
+        puterId: 'minimax:minimax/minimax-m2',
+        id: 'minimax-m2',
+        apiModel: 'MiniMax-M2',
+        name: 'MiniMax M2',
+        aliases: ['minimax/minimax-m2', 'MiniMax-M2', 'minimax/MiniMax-M2'],
+        modalities: { input: ['text'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        context: 204_800,
+        max_tokens: 196_608,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            prompt_tokens: 30, // $0.30 per 1M
+            completion_tokens: 120, // $1.20 per 1M
+            cached_tokens: 3, // $0.03 per 1M
+        },
+    },
 ];

src/backend/drivers/ai-chat/providers/mistral/MistralAiProvider.test.ts

@@ -655,6 +655,123 @@ describe('MistralAIProvider.complete streaming', () => {
     });
 });
 
+// -- Mistral image_url coercion --
+
+describe('MistralAIProvider image_url coercion', () => {
+    const baseCompletion = {
+        choices: [
+            {
+                message: { content: 'ok', role: 'assistant' },
+                finishReason: 'stop',
+            },
+        ],
+        usage: { promptTokens: 1, completionTokens: 1 },
+    };
+
+    it('flattens image_url objects to plain strings before sending to Mistral', async () => {
+        const { provider } = makeProvider();
+        completeMock.mockResolvedValueOnce(baseCompletion);
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'mistral-small-2603',
+                messages: [
+                    {
+                        role: 'user',
+                        content: [
+                            { type: 'text', text: 'describe this image' },
+                            {
+                                type: 'image_url',
+                                image_url: { url: 'https://example.com/img.png' },
+                            },
+                        ],
+                    },
+                ],
+            }),
+        );
+
+        const [args] = completeMock.mock.calls[0]!;
+        const imagePart = (args.messages[0].content as unknown[]).find(
+            (p: unknown) => (p as { type: string }).type === 'image_url',
+        ) as { type: string; image_url: unknown };
+        // Mistral expects a plain string, not { url: string }.
+        expect(imagePart.image_url).toBe('https://example.com/img.png');
+    });
+
+    it('leaves image_url parts that are already plain strings unchanged', async () => {
+        const { provider } = makeProvider();
+        completeMock.mockResolvedValueOnce(baseCompletion);
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'mistral-small-2603',
+                messages: [
+                    {
+                        role: 'user',
+                        content: [
+                            {
+                                type: 'image_url',
+                                image_url: 'https://example.com/already-flat.png',
+                            },
+                        ],
+                    },
+                ],
+            }),
+        );
+
+        const [args] = completeMock.mock.calls[0]!;
+        const imagePart = (args.messages[0].content as unknown[]).find(
+            (p: unknown) => (p as { type: string }).type === 'image_url',
+        ) as { type: string; image_url: unknown };
+        expect(imagePart.image_url).toBe('https://example.com/already-flat.png');
+    });
+
+    it('leaves messages with plain string content untouched', async () => {
+        const { provider } = makeProvider();
+        completeMock.mockResolvedValueOnce(baseCompletion);
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'mistral-small-2603',
+                messages: [{ role: 'user', content: 'plain text message' }],
+            }),
+        );
+
+        const [args] = completeMock.mock.calls[0]!;
+        expect(args.messages[0].content).toBe('plain text message');
+    });
+
+    it('coerces only image_url parts and leaves other content parts intact', async () => {
+        const { provider } = makeProvider();
+        completeMock.mockResolvedValueOnce(baseCompletion);
+
+        await withTestActor(() =>
+            provider.complete({
+                model: 'mistral-small-2603',
+                messages: [
+                    {
+                        role: 'user',
+                        content: [
+                            { type: 'text', text: 'what is in this image?' },
+                            {
+                                type: 'image_url',
+                                image_url: { url: 'https://example.com/photo.jpg' },
+                            },
+                        ],
+                    },
+                ],
+            }),
+        );
+
+        const [args] = completeMock.mock.calls[0]!;
+        const parts = args.messages[0].content as { type: string; text?: string; image_url?: unknown }[];
+        const textPart = parts.find((p) => p.type === 'text');
+        const imgPart = parts.find((p) => p.type === 'image_url');
+        expect(textPart?.text).toBe('what is in this image?');
+        expect(imgPart?.image_url).toBe('https://example.com/photo.jpg');
+    });
+});
+
 // ── Error mapping ───────────────────────────────────────────────────
 
 describe('MistralAIProvider.complete error mapping', () => {

src/backend/drivers/ai-chat/providers/mistral/MistralAiProvider.ts

@@ -61,6 +61,38 @@ export class MistralAIProvider implements IChatProvider {
         return ids;
     }
 
+    /**
+     * Mistral's API expects `image_url` content parts to carry a plain
+     * string URL, not the OpenAI-style `{ url: string }` object.
+     * This method normalises any `{ type: 'image_url', image_url: { url } }`
+     * parts to `{ type: 'image_url', image_url: url }` before the request
+     * is sent. Messages whose `content` is a plain string are left untouched.
+     */
+    #coerceImageUrls(
+        messages: { role: string; content: unknown }[],
+    ): { role: string; content: unknown }[] {
+        return messages.map((message) => {
+            if (!Array.isArray(message.content)) return message;
+            const content = message.content.map(
+                (part: { type?: string; image_url?: unknown }) => {
+                    if (
+                        part.type === 'image_url' &&
+                        part.image_url !== null &&
+                        typeof part.image_url === 'object' &&
+                        'url' in (part.image_url as object)
+                    ) {
+                        return {
+                            ...part,
+                            image_url: (part.image_url as { url: string }).url,
+                        };
+                    }
+                    return part;
+                },
+            );
+            return { ...message, content };
+        });
+    }
+
     async complete({
         messages,
         stream,
@@ -70,6 +102,7 @@ export class MistralAIProvider implements IChatProvider {
         temperature,
     }: ICompleteArguments): Promise<IChatCompleteResult> {
         messages = await OpenAIUtil.process_input_messages(messages);
+        messages = this.#coerceImageUrls(messages);
         for (const message of messages) {
             if (message.tool_calls) {
                 message.toolCalls = message.tool_calls;

src/backend/drivers/ai-chat/providers/openai/OpenAiChatCompletionsProvider.ts

@@ -26,6 +26,10 @@ import type { MeteringService } from '../../../../services/metering/MeteringServ
 import type { FSEntryStore } from '../../../../stores/fs/FSEntryStore.js';
 import type { S3ObjectStore } from '../../../../stores/fs/S3ObjectStore.js';
 import type { IChatProvider, ICompleteArguments } from '../../types.js';
+import {
+    messagesHaveCompaction,
+    wantsCompaction,
+} from '../../utils/compaction.js';
 import * as OpenAiUtil from '../../utils/OpenAIUtil.js';
 import { processPuterPathUploads } from './fileUpload.js';
 import { OPEN_AI_MODELS } from './models.js';
@@ -125,6 +129,20 @@ export class OpenAiChatProvider implements IChatProvider {
             }
             return await this.#responsesProvider.complete(params);
         }
+        // Inline compaction is a Responses-API feature; chat.completions can't
+        // express `context_management` or a `compaction` content block.
+        // Delegate to the sibling Responses provider when the caller opted in
+        // OR when the messages carry a round-tripped compaction artifact.
+        if (wantsCompaction(params) || messagesHaveCompaction(messages)) {
+            if (!this.#responsesProvider) {
+                throw new HttpError(
+                    400,
+                    'compaction requires the OpenAI Responses provider, which is not configured',
+                    { legacyCode: 'bad_request' },
+                );
+            }
+            return await this.#responsesProvider.complete(params);
+        }
         // Validate messages
         if (!Array.isArray(messages)) {
             throw new HttpError(400, '`messages` must be an array', {

src/backend/drivers/ai-chat/providers/openai/OpenAiChatResponsesProvider.ts

@@ -25,6 +25,7 @@ import type { MeteringService } from '../../../../services/metering/MeteringServ
 import type { FSEntryStore } from '../../../../stores/fs/FSEntryStore.js';
 import type { S3ObjectStore } from '../../../../stores/fs/S3ObjectStore.js';
 import type { IChatProvider, ICompleteArguments } from '../../types.js';
+import { toOpenAiContextManagement } from '../../utils/compaction.js';
 import * as OpenAiUtil from '../../utils/OpenAIUtil.js';
 import { processPuterPathUploads } from './fileUpload.js';
 import { OPEN_AI_MODELS } from './models.js';
@@ -100,6 +101,8 @@ export class OpenAiResponsesChatProvider implements IChatProvider {
         parallel_tool_calls,
         include,
         conversation,
+        compaction,
+        context_management,
         previous_response_id,
         instructions,
         metadata,
@@ -177,6 +180,13 @@ export class OpenAiResponsesChatProvider implements IChatProvider {
         const supportsReasoningControls =
             typeof model === 'string' && model.startsWith('gpt-5');
 
+        // Translate the neutral compaction opt-in (or pass a raw
+        // `context_management` payload through) to OpenAI's Responses shape.
+        const contextManagement = toOpenAiContextManagement({
+            compaction,
+            context_management,
+        });
+
         const completionParams: ResponseCreateParams = {
             user: userIdentifier,
             safety_identifier: userIdentifier,
@@ -188,6 +198,9 @@ export class OpenAiResponsesChatProvider implements IChatProvider {
                 ? { parallel_tool_calls }
                 : {}),
             ...(include !== undefined ? { include } : {}),
+            ...(contextManagement !== undefined
+                ? { context_management: contextManagement }
+                : {}),
             ...(conversation !== undefined ? { conversation } : {}),
             ...(previous_response_id !== undefined
                 ? { previous_response_id }

src/backend/drivers/ai-chat/providers/openai/models.ts

@@ -363,6 +363,30 @@ export const OPEN_AI_MODELS: IChatModel[] = [
         context: 128_000,
         max_tokens: 128000,
     },
+    {
+        puterId: 'openai:openai/gpt-5-codex',
+        id: 'gpt-5-codex',
+        modalities: { input: ['text', 'image'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2024-09-30',
+        release_date: '2025-09-15',
+        aliases: ['openai/gpt-5-codex'],
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            // models.dev openai: input $1.25 / output $10 / cache_read $0.125
+            // per 1M — identical to gpt-5.
+            tokens: 1_000_000,
+            prompt_tokens: 125,
+            cached_tokens: 13,
+            completion_tokens: 1000,
+        },
+        context: 128_000,
+        max_tokens: 128000,
+        responses_api_only: true,
+    },
     {
         puterId: 'openai:openai/gpt-5-mini',
         id: 'gpt-5-mini-2025-08-07',

src/backend/drivers/ai-chat/providers/together/TogetherAIProvider.ts

@@ -150,21 +150,7 @@ export class TogetherAIProvider implements IChatProvider {
             messages,
             stream,
             ...(tools ? { tools } : {}),
-            // TODO: make this better but togetherai doesn't handle max tokens properly at all
-            ...(max_tokens
-                ? {
-                      max_tokens:
-                          max_tokens -
-                          messages.reduce((acc, curr) => {
-                              return (
-                                  acc +
-                                  (curr.type === 'text'
-                                      ? curr.text.length / 2
-                                      : 200)
-                              );
-                          }, 0),
-                  }
-                : {}),
+            ...(max_tokens ? { max_tokens } : {}),
             ...(temperature ? { temperature } : {}),
             ...(stream ? { stream_options: { include_usage: true } } : {}),
         } as Together.Chat.Completions.CompletionCreateParamsNonStreaming);

src/backend/drivers/ai-chat/providers/xai/models.ts

@@ -198,6 +198,66 @@ export const XAI_MODELS: IChatModel[] = [
         },
         max_tokens: 30_000,
     },
+    {
+        puterId: 'x-ai:x-ai/grok-4-20-reasoning',
+        // xAI exposes this as the dated snapshot id; `grok-4-20-reasoning`
+        // (and dotted forms) are accepted as aliases by callers.
+        id: 'grok-4.20-0309-reasoning',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2026-03-09',
+        name: 'Grok 4.20 (Reasoning)',
+        aliases: [
+            'x-ai/grok-4-20-reasoning',
+            'grok-4-20-reasoning',
+            'grok-4.20-reasoning',
+            'x-ai/grok-4.20-reasoning',
+        ],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            // models.dev xai: input $1.25 / output $2.5 / cache_read $0.2 per 1M.
+            prompt_tokens: 125,
+            completion_tokens: 250,
+            cached_tokens: 20,
+        },
+        max_tokens: 30_000,
+    },
+    {
+        puterId: 'x-ai:x-ai/grok-4-20-non-reasoning',
+        // xAI exposes this as the dated snapshot id; `grok-4-20-non-reasoning`
+        // (and dotted forms) are accepted as aliases by callers.
+        id: 'grok-4.20-0309-non-reasoning',
+        modalities: { input: ['text', 'image', 'pdf'], output: ['text'] },
+        open_weights: false,
+        tool_call: true,
+        knowledge: '2025-07',
+        release_date: '2026-03-09',
+        name: 'Grok 4.20 (Non-Reasoning)',
+        aliases: [
+            'x-ai/grok-4-20-non-reasoning',
+            'grok-4-20-non-reasoning',
+            'grok-4.20-non-reasoning',
+            'x-ai/grok-4.20-non-reasoning',
+        ],
+        context: 2_000_000,
+        costs_currency: 'usd-cents',
+        input_cost_key: 'prompt_tokens',
+        output_cost_key: 'completion_tokens',
+        costs: {
+            tokens: 1_000_000,
+            // models.dev xai: input $1.25 / output $2.5 / cache_read $0.2 per 1M.
+            prompt_tokens: 125,
+            completion_tokens: 250,
+            cached_tokens: 20,
+        },
+        max_tokens: 30_000,
+    },
     {
         puterId: 'x-ai:x-ai/grok-4-1-fast',
         id: 'grok-4-1-fast',

src/backend/drivers/ai-chat/providers/zai/models.ts

@@ -69,6 +69,13 @@ const visionModel = (
 // Hardcoded from https://docs.z.ai/api-reference/llm/chat-completion and
 // https://docs.z.ai/guides/overview/pricing.
 export const ZAI_MODELS: IChatModel[] = [
+    textModel(
+        'glm-5.2',
+        'GLM-5.2',
+        1_000 * K,
+        128 * K,
+        usdPerMToken(1.4, 4.4, 0.26),
+    ),
     textModel(
         'glm-5.1',
         'GLM-5.1',

src/backend/drivers/ai-chat/types.ts

@@ -67,6 +67,19 @@ export interface ICompleteArguments {
     parallel_tool_calls?: boolean;
     include?: unknown[];
     conversation?: unknown;
+    /**
+     * Provider-neutral inline-compaction opt-in. `true` enables compaction with
+     * provider defaults; `{ trigger_tokens }` sets the token threshold at which
+     * the upstream summarizes earlier context. Each provider translates this to
+     * its own SDK shape (OpenAI `context_management:[{type:'compaction',...}]`,
+     * Anthropic `context_management:{edits:[{type:'compact_20260112'}]}`).
+     */
+    compaction?: boolean | { trigger_tokens?: number };
+    /**
+     * Escape hatch: provider-native `context_management` payload, passed through
+     * untouched (used by `/responses` callers sending the OpenAI-native array).
+     */
+    context_management?: unknown;
     previous_response_id?: string;
     instructions?: string | PuterMessage[];
     metadata?: Record<string, string>;
@@ -116,6 +129,13 @@ export interface IChatMessageResult {
     finally_fn?: never;
     normalized?: boolean;
     via_ai_chat_service?: boolean;
+    /**
+     * Inline-compaction artifact, present when the upstream compacted earlier
+     * context during this (non-streaming) response. Carries `type:'compaction'`
+     * so it's a drop-in `messages` item — the caller resends it on the next turn
+     * in place of the summarized history. See [[ICompleteArguments]].
+     */
+    compaction?: { type: 'compaction'; id?: string; encrypted_content: string };
 }
 
 export type IChatCompleteResult = IChatStreamResult | IChatMessageResult;

src/backend/drivers/ai-chat/utils/Messages.js

@@ -49,6 +49,23 @@ export const normalize_single_message = (message, params = {}) => {
             legacyCode: 'bad_request',
         });
     }
+    // A round-tripped inline-compaction artifact may be supplied as a bare
+    // top-level item (the shape the client received from the stream). Wrap it
+    // into an internal compaction content block so it survives normalization;
+    // each provider maps it back to its native input shape (OpenAI top-level
+    // input item, Anthropic content block).
+    if (!message.role && !message.content && message.type === 'compaction') {
+        return {
+            role: 'assistant',
+            content: [
+                {
+                    type: 'compaction',
+                    ...(message.id !== undefined ? { id: message.id } : {}),
+                    encrypted_content: message.encrypted_content,
+                },
+            ],
+        };
+    }
     if (!message.role) {
         message.role = params.role;
     }

src/backend/drivers/ai-chat/utils/Messages.test.ts

@@ -46,6 +46,22 @@ describe('normalize_single_message', () => {
         expect(result.role).toBe('system');
     });
 
+    it('wraps a bare round-tripped compaction item into a compaction content block', () => {
+        const result = normalize_single_message({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+        expect(result.role).toBe('assistant');
+        expect(result.content).toEqual([
+            {
+                type: 'compaction',
+                id: 'cmpct_1',
+                encrypted_content: 'ENC',
+            },
+        ]);
+    });
+
     it('throws 400 when message is null/undefined/array', () => {
         expect(() => normalize_single_message(null)).toThrow(
             expect.objectContaining({ statusCode: 400 }),

src/backend/drivers/ai-chat/utils/OpenAIUtil.js

@@ -84,6 +84,39 @@ export const process_input_messages = async (messages) => {
 };
 
 export const process_input_messages_responses_api = async (messages) => {
+    // Pre-split round-tripped compaction blocks into standalone Responses
+    // compaction input items, preserving any sibling content (e.g. the
+    // assistant's reply text) as its own message. A compaction item represents
+    // prior history, so it precedes the message it was attached to. This avoids
+    // collapsing the whole message into a single compaction item and dropping
+    // the rest of its content.
+    const expanded = [];
+    for (const msg of messages) {
+        if (msg && Array.isArray(msg.content)) {
+            const compactionBlocks = msg.content.filter(
+                (c) => c && c.type === 'compaction',
+            );
+            if (compactionBlocks.length > 0) {
+                for (const block of compactionBlocks) {
+                    expanded.push({
+                        type: 'compaction',
+                        ...(block.id !== undefined ? { id: block.id } : {}),
+                        encrypted_content: block.encrypted_content,
+                    });
+                }
+                const rest = msg.content.filter(
+                    (c) => !(c && c.type === 'compaction'),
+                );
+                if (rest.length > 0) {
+                    expanded.push({ ...msg, content: rest });
+                }
+                continue;
+            }
+        }
+        expanded.push(msg);
+    }
+    messages = expanded;
+
     for (const msg of messages) {
         const content_as_string = (content) => {
             if (content === undefined || content === null) return '';
@@ -253,6 +286,7 @@ export const create_chat_stream_handler =
         const tool_call_blocks = [];
 
         let last_usage = null;
+        let last_extra_content = null;
         for await (let chunk of completion) {
             chunk = deviations.chunk_but_like_actually(chunk);
             const chunk_usage = deviations.index_usage_from_stream_chunk(chunk);
@@ -285,6 +319,14 @@ export const create_chat_stream_handler =
                 // Apps have to choose to handle extra_content themselves, it doesn't seem like theres a way we can do it in a backwards
                 // compatible fashion since most streaming apps will handle chat history by continuously updating content themselves
                 // This doesn't present us a chance to add in an extra object for gemini's chat continuing features
+                // Don't let a later extra_content chunk without grounding clobber an
+                // earlier one that carried grounding_metadata (used for metering).
+                if (
+                    choice.delta.extra_content.grounding_metadata ||
+                    !last_extra_content?.grounding_metadata
+                ) {
+                    last_extra_content = choice.delta.extra_content;
+                }
                 textblock.addExtraContent(choice.delta.extra_content);
             }
 
@@ -315,7 +357,10 @@ export const create_chat_stream_handler =
         }
 
         // TODO DS: this is a bit too abstracted... this is basically just doing the metering now
-        const usage = usage_calculator({ usage: last_usage });
+        const usage = usage_calculator({
+            usage: last_usage,
+            extra_content: last_extra_content,
+        });
 
         if (mode === 'text') textblock.end();
         if (mode === 'tool') toolblock.end();
@@ -355,6 +400,19 @@ export const create_chat_stream_handler_responses_api =
                 last_usage = chunk.response.usage;
             }
 
+            if (
+                chunk.type === 'response.output_item.done' &&
+                chunk.item?.type === 'compaction'
+            ) {
+                // Inline compaction fired mid-response — normalize the artifact
+                // into the canonical internal compaction event.
+                chatStream.compaction({
+                    id: chunk.item.id,
+                    encrypted_content: chunk.item.encrypted_content,
+                });
+                continue;
+            }
+
             if (
                 chunk.type === 'response.output_item.done' &&
                 chunk.item?.type === 'function_call'
@@ -488,10 +546,14 @@ export const handle_completion_output_responses_api = async ({
             ...(item.id ? { canonical_id: item.id } : {}),
         }));
 
+    // Inline-compaction artifact, if the upstream compacted this turn.
+    const compactionItem = output.find((item) => item?.type === 'compaction');
+
     const is_empty = completion.output_text.trim() === '';
-    if (is_empty && responseToolCalls.length < 1) {
+    if (is_empty && responseToolCalls.length < 1 && !compactionItem) {
         // GPT refuses to generate an empty response if you ask it to,
         // so this will probably only happen on an error condition.
+        // A compaction-only output is legitimate, so don't reject it.
         throw new HttpError(400, 'an empty response was generated', {
             legacyCode: 'bad_response',
         });
@@ -523,6 +585,18 @@ export const handle_completion_output_responses_api = async ({
     };
     ret.role = output.find((item) => item?.role)?.role ?? 'assistant';
 
+    if (compactionItem) {
+        // Include `type` so the artifact is a drop-in `messages` item for the
+        // stateless round-trip — symmetric with the streaming compaction chunk.
+        ret.compaction = {
+            type: 'compaction',
+            ...(compactionItem.id !== undefined
+                ? { id: compactionItem.id }
+                : {}),
+            encrypted_content: compactionItem.encrypted_content,
+        };
+    }
+
     delete ret.type;
 
     ret.usage = usage_calculator

src/backend/drivers/ai-chat/utils/OpenAIUtil.test.ts

@@ -232,6 +232,66 @@ describe('process_input_messages_responses_api', () => {
         expect(out!.output).toBe('part-apart-bpart-c');
     });
 
+    it('rewrites a round-tripped compaction content block into a top-level compaction item', async () => {
+        const messages: Array<Record<string, unknown>> = [
+            {
+                role: 'assistant',
+                content: [
+                    {
+                        type: 'compaction',
+                        id: 'cmpct_1',
+                        encrypted_content: 'ENC',
+                    },
+                ],
+            },
+        ];
+
+        const [out] = (await process_input_messages_responses_api(
+            messages,
+        )) as Array<Record<string, unknown>>;
+        expect(out!.type).toBe('compaction');
+        expect(out!.encrypted_content).toBe('ENC');
+        expect(out!.id).toBe('cmpct_1');
+        expect(out!.role).toBeUndefined();
+        expect(out!.content).toBeUndefined();
+    });
+
+    it('splits a compaction block + reply text into a compaction item plus a preserved message', async () => {
+        // Round-tripping an assistant turn that carries BOTH the compaction
+        // artifact and the reply text must keep the reply, not drop it.
+        const messages: Array<Record<string, unknown>> = [
+            {
+                role: 'assistant',
+                content: [
+                    {
+                        type: 'compaction',
+                        id: 'cmpct_1',
+                        encrypted_content: 'ENC',
+                    },
+                    { type: 'text', text: 'earlier reply' },
+                ],
+            },
+        ];
+
+        const out = (await process_input_messages_responses_api(
+            messages,
+        )) as Array<Record<string, unknown>>;
+
+        // Compaction item comes first (it represents prior history)...
+        expect(out).toHaveLength(2);
+        expect(out[0]).toEqual({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+        // ...followed by the assistant message with the reply preserved
+        // (assistant text upgraded to output_text).
+        expect(out[1]!.role).toBe('assistant');
+        expect(out[1]!.content).toEqual([
+            { type: 'output_text', text: 'earlier reply' },
+        ]);
+    });
+
     it('upgrades user/system text blocks to input_text', async () => {
         const messages: Array<Record<string, unknown>> = [
             {
@@ -539,6 +599,39 @@ describe('create_chat_stream_handler_responses_api', () => {
         });
     });
 
+    it('emits a compaction event when a compaction output_item completes', async () => {
+        const completion = asAsyncIterable([
+            {
+                type: 'response.output_item.done',
+                item: {
+                    type: 'compaction',
+                    id: 'cmpct_1',
+                    encrypted_content: 'ENC',
+                },
+            },
+            {
+                type: 'response.completed',
+                response: { usage: { input_tokens: 1, output_tokens: 2 } },
+            },
+        ]);
+        const init = create_chat_stream_handler_responses_api({
+            deviations: undefined,
+            completion,
+            usage_calculator: () => ({}),
+        });
+        const harness = makeCapturingChatStream();
+        await init({ chatStream: harness.chatStream });
+
+        const compaction = harness
+            .events()
+            .find((e: { type: string }) => e.type === 'compaction');
+        expect(compaction).toEqual({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+    });
+
     it('emits a tool_use block when a function_call output_item completes', async () => {
         const completion = asAsyncIterable([
             {
@@ -765,6 +858,31 @@ describe('handle_completion_output_responses_api non-stream', () => {
         ).rejects.toMatchObject({ statusCode: 400 });
     });
 
+    it('attaches a compaction artifact and allows a compaction-only output', async () => {
+        const completion = {
+            output: [
+                {
+                    type: 'compaction',
+                    id: 'cmpct_1',
+                    encrypted_content: 'ENC',
+                },
+            ],
+            output_text: '   ',
+            usage: { input_tokens: 1, output_tokens: 0 },
+        };
+        const result = await handle_completion_output_responses_api({
+            deviations: undefined,
+            stream: false,
+            completion,
+        });
+        // Compaction-only output is not rejected as "empty".
+        expect(result.compaction).toEqual({
+            type: 'compaction',
+            id: 'cmpct_1',
+            encrypted_content: 'ENC',
+        });
+    });
+
     it('runs moderation against output_text when a moderate fn is supplied', async () => {
         const completion = {
             output: [{ role: 'assistant' }],

src/backend/drivers/ai-chat/utils/Streaming.js

@@ -103,6 +103,25 @@ export class AIChatStream {
         this.stream.end();
     }
 
+    /**
+     * Emit a canonical compaction event into the NDJSON stream. Both the OpenAI
+     * and Anthropic providers normalize their native inline-compaction artifact
+     * to this single shape, so downstream consumers (controllers, puter.js) see
+     * an identical `{ type: 'compaction', id, encrypted_content }` chunk
+     * regardless of which upstream served the request.
+     *
+     * @param {{ id?: string, encrypted_content: string }} compaction
+     */
+    compaction({ id, encrypted_content }) {
+        this.stream.write(
+            `${JSON.stringify({
+                type: 'compaction',
+                ...(id !== undefined ? { id } : {}),
+                encrypted_content,
+            })}\n`,
+        );
+    }
+
     message() {
         return new AIChatMessageStream(this);
     }

src/backend/drivers/ai-chat/utils/compaction.js

@@ -0,0 +1,124 @@
+/*
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+/**
+ * Inline-compaction translation helpers.
+ *
+ * The driver-facing surface exposes a single provider-neutral opt-in
+ * (`compaction: boolean | { trigger_tokens }` on `ICompleteArguments`), plus a
+ * raw `context_management` escape hatch for callers hitting `/responses` with
+ * the OpenAI-native shape. These helpers map that neutral opt-in to each
+ * provider's SDK shape so the providers stay free of opt-in-parsing logic.
+ */
+
+/**
+ * @param {boolean | { trigger_tokens?: number } | undefined} compaction
+ * @returns {{ enabled: boolean, trigger_tokens?: number }}
+ */
+const readCompaction = (compaction) => {
+    if (compaction === true) return { enabled: true };
+    if (compaction && typeof compaction === 'object') {
+        return {
+            enabled: true,
+            ...(typeof compaction.trigger_tokens === 'number'
+                ? { trigger_tokens: compaction.trigger_tokens }
+                : {}),
+        };
+    }
+    return { enabled: false };
+};
+
+/**
+ * Build OpenAI Responses `context_management` from the neutral opt-in. A raw
+ * `context_management` passthrough (already in OpenAI shape) wins.
+ *
+ * @param {{ compaction?: boolean | { trigger_tokens?: number }, context_management?: unknown }} args
+ * @returns {Array<{ type: 'compaction', compact_threshold?: number }> | undefined}
+ */
+export const toOpenAiContextManagement = (args) => {
+    if (args.context_management !== undefined) {
+        return /** @type {any} */ (args.context_management);
+    }
+    const { enabled, trigger_tokens } = readCompaction(args.compaction);
+    if (!enabled) return undefined;
+    return [
+        {
+            type: 'compaction',
+            ...(trigger_tokens !== undefined
+                ? { compact_threshold: trigger_tokens }
+                : {}),
+        },
+    ];
+};
+
+/**
+ * Build Anthropic `context_management` (beta `compact-2026-01-12`) from the
+ * neutral opt-in. A raw `context_management` passthrough wins.
+ *
+ * @param {{ compaction?: boolean | { trigger_tokens?: number }, context_management?: unknown }} args
+ * @returns {{ edits: Array<Record<string, unknown>> } | undefined}
+ */
+export const toAnthropicContextManagement = (args) => {
+    if (args.context_management !== undefined) {
+        return /** @type {any} */ (args.context_management);
+    }
+    const { enabled, trigger_tokens } = readCompaction(args.compaction);
+    if (!enabled) return undefined;
+    return {
+        edits: [
+            {
+                type: 'compact_20260112',
+                ...(trigger_tokens !== undefined
+                    ? {
+                          trigger: {
+                              type: 'input_tokens',
+                              value: trigger_tokens,
+                          },
+                      }
+                    : {}),
+            },
+        ],
+    };
+};
+
+/**
+ * Whether the request opted into inline compaction by any route.
+ *
+ * @param {{ compaction?: boolean | { trigger_tokens?: number }, context_management?: unknown }} args
+ */
+export const wantsCompaction = (args) =>
+    args.context_management !== undefined ||
+    readCompaction(args.compaction).enabled;
+
+/**
+ * Whether the (normalized) message list carries a round-tripped compaction
+ * artifact. Such a request must route through a compaction-capable surface even
+ * if it didn't request *new* compaction — chat.completions can't represent a
+ * compaction content block, and Anthropic needs its compaction beta to accept
+ * one as input.
+ *
+ * @param {unknown} messages
+ */
+export const messagesHaveCompaction = (messages) =>
+    Array.isArray(messages) &&
+    messages.some(
+        (m) =>
+            Array.isArray(m?.content) &&
+            m.content.some((c) => c && c.type === 'compaction'),
+    );

src/backend/drivers/ai-chat/utils/compaction.test.ts

@@ -0,0 +1,132 @@
+/*
+ * Copyright (C) 2024-present Puter Technologies Inc.
+ *
+ * This file is part of Puter.
+ *
+ * Puter is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published
+ * by the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Affero General Public License for more details.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+import { describe, expect, it } from 'vitest';
+import {
+    messagesHaveCompaction,
+    toAnthropicContextManagement,
+    toOpenAiContextManagement,
+    wantsCompaction,
+} from './compaction.js';
+
+describe('toOpenAiContextManagement', () => {
+    it('returns undefined when compaction is off', () => {
+        expect(toOpenAiContextManagement({})).toBeUndefined();
+        expect(
+            toOpenAiContextManagement({ compaction: false }),
+        ).toBeUndefined();
+    });
+
+    it('builds a compaction entry from `true`', () => {
+        expect(toOpenAiContextManagement({ compaction: true })).toEqual([
+            { type: 'compaction' },
+        ]);
+    });
+
+    it('maps trigger_tokens to compact_threshold', () => {
+        expect(
+            toOpenAiContextManagement({ compaction: { trigger_tokens: 1000 } }),
+        ).toEqual([{ type: 'compaction', compact_threshold: 1000 }]);
+    });
+
+    it('passes a raw context_management payload through verbatim', () => {
+        const raw = [{ type: 'compaction', compact_threshold: 5 }];
+        expect(
+            toOpenAiContextManagement({
+                compaction: true,
+                context_management: raw,
+            }),
+        ).toBe(raw);
+    });
+});
+
+describe('toAnthropicContextManagement', () => {
+    it('returns undefined when compaction is off', () => {
+        expect(toAnthropicContextManagement({})).toBeUndefined();
+    });
+
+    it('builds a compact_20260112 edit from `true`', () => {
+        expect(toAnthropicContextManagement({ compaction: true })).toEqual({
+            edits: [{ type: 'compact_20260112' }],
+        });
+    });
+
+    it('maps trigger_tokens to an input_tokens trigger', () => {
+        expect(
+            toAnthropicContextManagement({
+                compaction: { trigger_tokens: 2000 },
+            }),
+        ).toEqual({
+            edits: [
+                {
+                    type: 'compact_20260112',
+                    trigger: { type: 'input_tokens', value: 2000 },
+                },
+            ],
+        });
+    });
+
+    it('passes a raw context_management payload through verbatim', () => {
+        const raw = { edits: [{ type: 'compact_20260112' }] };
+        expect(
+            toAnthropicContextManagement({ context_management: raw }),
+        ).toBe(raw);
+    });
+});
+
+describe('wantsCompaction', () => {
+    it('is false without opt-in', () => {
+        expect(wantsCompaction({})).toBe(false);
+        expect(wantsCompaction({ compaction: false })).toBe(false);
+    });
+
+    it('is true for the neutral opt-in or a raw passthrough', () => {
+        expect(wantsCompaction({ compaction: true })).toBe(true);
+        expect(wantsCompaction({ compaction: { trigger_tokens: 1 } })).toBe(
+            true,
+        );
+        expect(wantsCompaction({ context_management: [] })).toBe(true);
+    });
+});
+
+describe('messagesHaveCompaction', () => {
+    it('detects a round-tripped compaction content block', () => {
+        expect(
+            messagesHaveCompaction([
+                { role: 'user', content: [{ type: 'text', text: 'hi' }] },
+                {
+                    role: 'assistant',
+                    content: [
+                        { type: 'compaction', encrypted_content: 'ENC' },
+                    ],
+                },
+            ]),
+        ).toBe(true);
+    });
+
+    it('is false for ordinary messages or non-arrays', () => {
+        expect(
+            messagesHaveCompaction([
+                { role: 'user', content: [{ type: 'text', text: 'hi' }] },
+            ]),
+        ).toBe(false);
+        expect(messagesHaveCompaction(undefined)).toBe(false);
+        expect(messagesHaveCompaction('nope')).toBe(false);
+    });
+});

src/backend/drivers/apps/AppDriver.js

@@ -52,6 +52,12 @@ const INDEX_URL_UNIQUENESS_EXEMPTION_CANDIDATES = [
     'https://dev-center.puter.com/coming-soon',
 ];
 
+// Sentinel host for builtin apps. The GUI rewrites index_urls on this
+// host to `<gui origin>/builtin/<name>` (see launch_app.js), so rows
+// carrying it are reserved for migration-seeded builtins — a user app
+// claiming it would load its code same-origin with the desktop.
+const BUILTIN_APPS_HOST = 'builtins.namespaces.puter.com';
+
 // Canonical-uid alias namespace. When a user-created app is merged into
 // an existing origin-bootstrap row, the source uid is mapped to the
 // canonical (kept) uid so any client that still holds the old uid keeps
@@ -465,7 +471,7 @@ export class AppDriver extends PuterDriver {
 
     // -- Validation ---------------------------------------------------
 
-    async #validateInput(object, { isCreate }) {
+    async #validateInput(object, { isCreate, existing }) {
         const out = {};
 
         if (isCreate || object.name !== undefined) {
@@ -497,6 +503,15 @@ export class AppDriver extends PuterDriver {
                 maxLen: 3000,
                 required: isCreate,
             });
+            // Only enforce on new/changed values so rows that already
+            // carry a reserved host (migration-seeded builtins) can still
+            // have their other fields updated.
+            if (
+                out.index_url !== undefined &&
+                out.index_url !== existing?.index_url
+            ) {
+                this.#assertIndexUrlHostAllowed(out.index_url);
+            }
         }
         if (object.icon !== undefined) {
             validateString(object.icon, {
@@ -581,6 +596,54 @@ export class AppDriver extends PuterDriver {
         return out;
     }
 
+    /**
+     * App iframes run with `allow-same-origin allow-scripts`, so an
+     * index_url loading from the GUI host would execute third-party code
+     * same-origin with the desktop — a full sandbox escape. The API host
+     * is reserved for the same reason, and the builtin sentinel host is
+     * rewritten by the GUI to `<gui origin>/builtin/…` (see
+     * BUILTIN_APPS_HOST above). Host comparison (rather than full origin)
+     * deliberately also catches scheme/port variants of these hosts.
+     */
+    #assertIndexUrlHostAllowed(indexUrl) {
+        let hostname;
+        try {
+            hostname = new URL(indexUrl).hostname;
+        } catch {
+            // Unparseable values are rejected by `validateUrl` upstream.
+            return;
+        }
+
+        const config = this.config ?? {};
+        const reserved = new Set([BUILTIN_APPS_HOST]);
+        // `origin`/`api_base_url` are computed at boot from `domain`; the
+        // domain-based fallbacks cover callers (tests, embedders) that
+        // construct a server without that normalization step.
+        const candidates = [config.origin, config.api_base_url];
+        if (config.domain) {
+            candidates.push(
+                `http://${config.domain}`,
+                `http://api.${config.domain}`,
+            );
+        }
+        for (const candidate of candidates) {
+            if (!candidate) continue;
+            try {
+                reserved.add(new URL(candidate).hostname);
+            } catch {
+                // Malformed config value — nothing to reserve from it.
+            }
+        }
+
+        if (reserved.has(hostname)) {
+            throw new HttpError(
+                400,
+                '`index_url` cannot point at a Puter system host',
+                { legacyCode: 'bad_request' },
+            );
+        }
+    }
+
     // -- Permission checks --------------------------------------------
 
     #requireActor() {
@@ -849,7 +912,9 @@ export class AppDriver extends PuterDriver {
     // `#ensure_puter_site_subdomain_is_owned`, `#ensureIndexUrlNotAlreadyInUse`,
     // and the `app:canonicalUidAlias:*` kvstore pair). When a user creates
     // or repoints an app at a puter-hosted subdomain (`*.puter.site`,
-    // `*.puter.app`, …) we want exactly one app row to back that URL:
+    // `*.puter.app`, …) — or at a custom host claimed by an
+    // `app_origin_aliases` group — we want exactly one app row to back
+    // that URL:
     //   • If a no-owner row exists (origin-bootstrap stub auto-created
     //     when an unknown origin first hit Puter) → claim it for the
     //     user, merge the new fields into it.
@@ -1156,7 +1221,8 @@ export class AppDriver extends PuterDriver {
 
     /**
      * Merge an incoming create/update into an existing app row that
-     * already owns the same puter-hosted index_url. Returns the joined
+     * already owns the same puter-hosted or alias-group index_url.
+     * Returns the joined
      * (client-shaped) app on success, or `null` when no merge applied.
      * Throws `app_index_url_already_in_use` when a conflict exists but
      * cannot be merged (different owner, or same-owner non-bootstrap).
@@ -1173,7 +1239,18 @@ export class AppDriver extends PuterDriver {
         excludeAppId,
     } = {}) {
         const indexUrl = object?.index_url;
-        if (!this.#isPuterHostedIndexUrl(indexUrl)) return null;
+        // Alias-group hosts (`app_origin_aliases`) get the same merge
+        // treatment as puter-hosted subdomains. Without this, a bootstrap
+        // stub on a custom domain could never be absorbed — and since
+        // `#findIndexUrlConflictRow` *does* honor alias groups, the
+        // uniqueness check below would hard-reject the owner's own
+        // create/update instead of merging into the stub.
+        if (
+            !this.#isPuterHostedIndexUrl(indexUrl) &&
+            !this.#findOriginAliasGroupForIndexUrl(indexUrl)
+        ) {
+            return null;
+        }
 
         const conflictRow = await this.#findIndexUrlConflictRow({
             indexUrl,

src/backend/drivers/apps/AppDriver.test.ts

@@ -143,6 +143,57 @@ describe('AppDriver.create', () => {
         ).rejects.toMatchObject({ statusCode: 400 });
     });
 
+    // App iframes get `allow-same-origin allow-scripts`, so an index_url
+    // on a Puter system host would run third-party code same-origin with
+    // the GUI. The test server's domain is `puter.localhost` (from
+    // config.default.json).
+    it.each([
+        ['the GUI host', 'https://puter.localhost/evil.html'],
+        ['the GUI host on another port/scheme', 'http://puter.localhost:4100/evil.html'],
+        ['the API host', 'https://api.puter.localhost/evil.html'],
+        ['the builtin sentinel host', 'https://builtins.namespaces.puter.com/emulator'],
+    ])('rejects an index_url on %s with 400', async (_label, index_url) => {
+        const { actor } = await makeUser();
+        await expect(
+            withActor(actor, () =>
+                driver.create({
+                    object: {
+                        name: uniqueName('sys-host'),
+                        title: 't',
+                        index_url,
+                    },
+                }),
+            ),
+        ).rejects.toMatchObject({
+            statusCode: 400,
+            message: /system host/,
+        });
+    });
+
+    it('rejects updating an index_url to a Puter system host with 400', async () => {
+        const { actor } = await makeUser();
+        const created = await withActor(actor, () =>
+            driver.create({
+                object: {
+                    name: uniqueName('sys-host-upd'),
+                    title: 't',
+                    index_url: uniqueIndexUrl(),
+                },
+            }),
+        );
+        await expect(
+            withActor(actor, () =>
+                driver.update({
+                    uid: created.uid,
+                    object: { index_url: 'https://puter.localhost/evil.html' },
+                }),
+            ),
+        ).rejects.toMatchObject({
+            statusCode: 400,
+            message: /system host/,
+        });
+    });
+
     it('rejects a duplicate app name with 400', async () => {
         const a = await makeUser();
         const b = await makeUser();
@@ -897,3 +948,136 @@ describe('AppDriver.isNameAvailable additional branches', () => {
         );
     });
 });
+
+// ── alias-group custom domains (`app_origin_aliases`) ──────────────
+//
+// Custom hosts claimed by an alias group get the same bootstrap-stub
+// merge treatment as puter-hosted subdomains: creating or repointing
+// an app at an aliased host absorbs the unowned origin-bootstrap row
+// instead of rejecting with `app_index_url_already_in_use`.
+
+describe('AppDriver alias-group index_url merge', () => {
+    const aliasHostA = `alias-a-${Math.random().toString(36).slice(2, 10)}.test`;
+    const aliasHostB = `alias-b-${Math.random().toString(36).slice(2, 10)}.test`;
+
+    // `config` is protected on PuterDriver; reach in to toggle the alias
+    // groups for this block only. `#getOriginAliasGroups` reads config at
+    // call time, so runtime mutation takes effect immediately.
+    const driverConfig = () =>
+        (driver as unknown as { config: Record<string, unknown> }).config;
+
+    beforeAll(() => {
+        driverConfig().app_origin_aliases = [[aliasHostA], [aliasHostB]];
+    });
+
+    afterAll(() => {
+        delete driverConfig().app_origin_aliases;
+    });
+
+    const makeBootstrapStub = async (host: string) => {
+        const stubUid = `app-${uuidv4()}`;
+        // Mirrors AuthController's get-user-app-token bootstrap path:
+        // origin persisted as index_url, no owner, name === uid.
+        await server.stores.app.createFromOrigin(stubUid, `https://${host}`);
+        return stubUid;
+    };
+
+    it('create at an aliased host absorbs the unowned bootstrap stub', async () => {
+        const { actor, userId } = await makeUser();
+        const stubUid = await makeBootstrapStub(aliasHostA);
+        const name = uniqueName('alias-create');
+
+        const result = await withActor(actor, () =>
+            driver.create({
+                object: {
+                    name,
+                    title: 'Aliased',
+                    index_url: `https://${aliasHostA}/`,
+                },
+            }),
+        );
+
+        // The stub row survives as the canonical app, claimed + merged.
+        expect(result.uid).toBe(stubUid);
+        expect(result.name).toBe(name);
+        const stored = await server.stores.app.getByUid(stubUid);
+        expect(stored?.owner_user_id).toBe(userId);
+    });
+
+    it('rejects another user registering an app under a reserved aliased host', async () => {
+        const other = await makeUser();
+        await expect(
+            withActor(other.actor, () =>
+                driver.create({
+                    object: {
+                        name: uniqueName('squatter'),
+                        title: 't',
+                        index_url: `https://${aliasHostA}/index.html`,
+                    },
+                }),
+            ),
+        ).rejects.toMatchObject({ statusCode: 400 });
+    });
+
+    it('update repointing to an aliased host merges and aliases the old uid', async () => {
+        const { actor, userId } = await makeUser();
+        const stubUid = await makeBootstrapStub(aliasHostB);
+
+        const created = await withActor(actor, () =>
+            driver.create({
+                object: {
+                    name: uniqueName('alias-upd'),
+                    title: 't',
+                    index_url: uniqueIndexUrl(),
+                },
+            }),
+        );
+
+        const updated = await withActor(actor, () =>
+            driver.update({
+                uid: created.uid,
+                object: { index_url: `https://${aliasHostB}/` },
+            }),
+        );
+
+        // Merged into the stub; source row deleted; old uid still resolves
+        // via the canonical-uid alias.
+        expect(updated.uid).toBe(stubUid);
+        expect(await server.stores.app.getByUid(created.uid as string)).toBeNull();
+        const stored = await server.stores.app.getByUid(stubUid);
+        expect(stored?.owner_user_id).toBe(userId);
+
+        const viaOldUid = await withActor(actor, () =>
+            driver.read({ uid: created.uid }),
+        );
+        expect(viaOldUid.uid).toBe(stubUid);
+    });
+
+    it('leaves unrelated custom domains untouched (no alias group, no conflict check)', async () => {
+        const a = await makeUser();
+        const b = await makeUser();
+        const sharedUrl = uniqueIndexUrl();
+
+        // Non-puter, non-aliased hosts keep their historical behavior:
+        // no uniqueness enforcement, both creates succeed.
+        const first = await withActor(a.actor, () =>
+            driver.create({
+                object: {
+                    name: uniqueName('plain-a'),
+                    title: 't',
+                    index_url: sharedUrl,
+                },
+            }),
+        );
+        const second = await withActor(b.actor, () =>
+            driver.create({
+                object: {
+                    name: uniqueName('plain-b'),
+                    title: 't',
+                    index_url: sharedUrl,
+                },
+            }),
+        );
+        expect(first.uid).not.toBe(second.uid);
+    });
+});

src/backend/drivers/kv/costs.ts

@@ -20,6 +20,6 @@
 // Microcents per underlying DynamoDB capacity unit, as reported by
 // SystemKVStore.KVUsage. Cost is `KV_COSTS[op] * usage.<op>`.
 export const KV_COSTS = {
-    'kv:read': 63,
-    'kv:write': 125,
+    'kv:read': 17,
+    'kv:write': 90,
 } as const;

src/backend/drivers/util/fileInput.ts

@@ -175,9 +175,7 @@ export async function loadFileInput(
     // `path`/`uid`/`uuid` (e.g. AI chat `puter_path` content parts) could
     // exfiltrate any user's file. Must run before the S3 read below.
     await fsService.checkFSAccess(entry, actor, 'read');
-    // S3 object key is recorded in entry.metadata.objectKey when written by
-    // fsv2; older rows fall back to the entry uuid.
-    const objectKey = deriveObjectKey(entry);
+    const objectKey = entry.uuid;
     const { body, contentType, contentLength } =
         await stores.s3Object.getObjectStream(
             {
@@ -244,29 +242,6 @@ function filenameFromMime(mime: string): string {
     return `input.${ext}`;
 }
 
-// Mirrors FSService's private deriveObjectKeyFromEntry helper. fsv2-era
-// rows persist an `objectKey` in metadata; older rows simply use the uuid.
-function deriveObjectKey(entry: {
-    uuid: string;
-    metadata: string | null;
-}): string {
-    if (entry.metadata) {
-        try {
-            const parsed = JSON.parse(entry.metadata);
-            if (
-                parsed &&
-                typeof parsed.objectKey === 'string' &&
-                parsed.objectKey.length > 0
-            ) {
-                return parsed.objectKey;
-            }
-        } catch {
-            // Not JSON — fall through.
-        }
-    }
-    return entry.uuid;
-}
-
 export function inferFilenameFromUrlOrPath(
     value: string,
     fallback = 'input',

src/backend/extensions.test.ts

@@ -78,6 +78,12 @@ describe('extension.import("client") optional client access', () => {
                 return null;
             }
         };
-        expect(probe()).toBe(fake);
+        // The import proxy method-binds, so the result is a binding proxy over
+        // `fake` rather than the raw reference (identity is intentionally not
+        // preserved). What the probe pattern locks is that a registered client
+        // surfaces a callable method.
+        const result = probe();
+        expect(result).not.toBeNull();
+        expect(typeof (result as { query: unknown }).query).toBe('function');
     });
 });