refactor(protocol): derive simulation wire contracts

Co-authored-by: Aiden Cline <rekram1-node@users.noreply.github.com>

.changeset/calm-moves-cache.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/core": patch
+---
+
+Preserve prompt cache prefixes when sessions move between locations with unchanged instructions.

.changeset/quiet-forms-replace-questions.md

@@ -0,0 +1,8 @@
+---
+"@opencode-ai/core": minor
+"@opencode-ai/schema": minor
+"@opencode-ai/protocol": minor
+"@opencode-ai/client": minor
+---
+
+Remove the unused question request API and use session forms for question tool interactions.

.changeset/shell-permission-scan.md

@@ -0,0 +1,9 @@
+---
+"@opencode-ai/core": minor
+"@opencode-ai/schema": minor
+"@opencode-ai/protocol": minor
+"@opencode-ai/client": minor
+---
+
+Add an opt-in portable shell permission scanner. Opaque commands use normal shell authorization without inferring
+external directories, while the default tree-sitter path remains unchanged.

.gitattributes

@@ -1,3 +1,4 @@
 packages/core/migration/**/snapshot.json linguist-generated
 packages/core/src/database/migration.gen.ts linguist-generated
+packages/core/src/models-dev/snapshot.txt linguist-generated
 packages/core/src/**/*.txt text eol=lf

.github/TEAM_MEMBERS

@@ -10,6 +10,7 @@ kitlangton
 kommander
 ludvigrask
 MrMushrooooom
+neriousy
 nexxeln
 R44VC0RP
 rekram1-node

.github/workflows/generate.yml

@@ -4,6 +4,7 @@ on:
   push:
     branches:
       - dev
+      - v2
 
 jobs:
   generate:

.github/workflows/publish.yml

@@ -6,7 +6,6 @@ on:
     branches:
       - ci
       - dev
-      - v2
       - beta
       - fix/npm-native-binary-install
       - snapshot-*
@@ -75,7 +74,7 @@ jobs:
   build-cli:
     needs: version
     runs-on: blacksmith-4vcpu-ubuntu-2404
-    if: github.repository == 'anomalyco/opencode'
+    if: github.repository == 'anomalyco/opencode' && github.ref_name != 'beta'
     steps:
       - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
         with:
@@ -91,7 +90,7 @@ jobs:
           opencode-app-secret: ${{ secrets.OPENCODE_APP_SECRET }}
 
       - name: Build legacy CLI
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         run: ./packages/opencode/script/build.ts ${{ (github.ref_name == 'beta' && '--sourcemaps') || '' }}
         env:
           OPENCODE_VERSION: ${{ needs.version.outputs.version }}
@@ -109,7 +108,7 @@ jobs:
           GH_TOKEN: ${{ steps.committer.outputs.token }}
 
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         with:
           name: opencode-cli
           path: |
@@ -117,22 +116,76 @@ jobs:
             packages/opencode/dist/opencode-linux*
 
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         with:
           name: opencode-cli-windows
           path: packages/opencode/dist/opencode-windows*
 
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
         with:
-          name: opencode-preview-cli
+          name: opencode-preview-cli-unsigned
           path: packages/cli/dist/cli-*
 
     outputs:
       version: ${{ needs.version.outputs.version }}
 
+  sign-cli-macos:
+    needs: build-cli
+    runs-on: macos-26
+    if: github.repository == 'anomalyco/opencode'
+    steps:
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+
+      - uses: apple-actions/import-codesign-certs@8f3fb608891dd2244cdab3d69cd68c0d37a7fe93 # v2.0.0
+        with:
+          keychain: build
+          p12-file-base64: ${{ secrets.APPLE_CERTIFICATE }}
+          p12-password: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}
+
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        with:
+          name: opencode-preview-cli-unsigned
+          path: packages/cli/dist
+
+      - name: Sign macOS CLI binaries
+        run: |
+          identity=$(security find-identity -v -p codesigning build.keychain | sed -n 's/.*"\(Developer ID Application:.*\)"/\1/p' | head -n 1)
+          if [ -z "$identity" ]; then
+            echo "Developer ID Application identity not found"
+            exit 1
+          fi
+
+          found=0
+          for file in packages/cli/dist/cli-darwin-*/bin/opencode2; do
+            if [ ! -f "$file" ]; then
+              continue
+            fi
+            found=1
+            codesign \
+              --force \
+              --timestamp \
+              --options runtime \
+              --entitlements packages/cli/script/entitlements.plist \
+              --sign "$identity" \
+              "$file"
+            codesign --verify --deep --strict --verbose=4 "$file"
+            codesign --display --requirements - "$file"
+          done
+
+          if [ "$found" -eq 0 ]; then
+            echo "No macOS CLI binaries found"
+            exit 1
+          fi
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: opencode-preview-cli
+          path: packages/cli/dist/cli-*
+          if-no-files-found: error
+
   build-node-cli:
     needs: version
-    if: github.repository == 'anomalyco/opencode'
+    if: github.repository == 'anomalyco/opencode' && github.ref_name != 'beta'
     strategy:
       fail-fast: false
       matrix:
@@ -184,7 +237,7 @@ jobs:
       - build-cli
       - version
     runs-on: blacksmith-4vcpu-windows-2025
-    if: github.repository == 'anomalyco/opencode' && github.ref_name != 'v2'
+    if: github.repository == 'anomalyco/opencode' && github.ref_name != 'v2' && github.ref_name != 'beta'
     env:
       AZURE_CLIENT_ID: ${{ secrets.AZURE_CLIENT_ID }}
       AZURE_TENANT_ID: ${{ secrets.AZURE_TENANT_ID }}
@@ -281,7 +334,6 @@ jobs:
 
   build-electron:
     needs:
-      - build-cli
       - version
     if: github.repository == 'anomalyco/opencode' && github.ref_name != 'v2'
     continue-on-error: false
@@ -378,16 +430,14 @@ jobs:
         env:
           OPENCODE_VERSION: ${{ needs.version.outputs.version }}
           OPENCODE_CHANNEL: ${{ (github.ref_name == 'beta' && 'beta') || 'prod' }}
-          OPENCODE_CLI_ARTIFACT: ${{ (runner.os == 'Windows' && 'opencode-cli-windows') || 'opencode-cli' }}
-          RUST_TARGET: ${{ matrix.settings.target }}
-          GH_TOKEN: ${{ github.token }}
-          GITHUB_RUN_ID: ${{ github.run_id }}
+          OPENCODE_CLI_TARGET: ${{ matrix.settings.target }}
 
       - name: Build
         run: bun run build
         working-directory: packages/desktop
         env:
           NODE_OPTIONS: --max-old-space-size=4096
+          OPENCODE_VERSION: ${{ needs.version.outputs.version }}
           OPENCODE_CHANNEL: ${{ (github.ref_name == 'beta' && 'beta') || 'prod' }}
           SENTRY_AUTH_TOKEN: ${{ secrets.SENTRY_AUTH_TOKEN }}
           SENTRY_ORG: ${{ vars.SENTRY_ORG }}
@@ -396,6 +446,7 @@ jobs:
           VITE_SENTRY_DSN: ${{ vars.WEB_SENTRY_DSN }}
           VITE_SENTRY_ENVIRONMENT: ${{ (github.ref_name == 'beta' && 'beta') || 'production' }}
           VITE_SENTRY_RELEASE: desktop@${{ needs.version.outputs.version }}
+          OPENCODE_CLI_TARGET: ${{ matrix.settings.target }}
 
       - name: Package
         if: needs.version.outputs.release
@@ -405,8 +456,7 @@ jobs:
         env:
           OPENCODE_CHANNEL: ${{ (github.ref_name == 'beta' && 'beta') || 'prod' }}
           GH_TOKEN: ${{ steps.committer.outputs.token }}
-          CSC_LINK: ${{ secrets.APPLE_CERTIFICATE }}
-          CSC_KEY_PASSWORD: ${{ secrets.APPLE_CERTIFICATE_PASSWORD }}
+          CSC_KEYCHAIN: build.keychain
           APPLE_API_KEY: ${{ runner.temp }}/apple-api-key.p8
           APPLE_API_KEY_ID: ${{ secrets.APPLE_API_KEY }}
           APPLE_API_ISSUER: ${{ secrets.APPLE_API_ISSUER }}
@@ -471,6 +521,7 @@ jobs:
     needs:
       - version
       - build-cli
+      - sign-cli-macos
       - build-node-cli
       - sign-cli-windows
       - build-electron
@@ -500,29 +551,31 @@ jobs:
           registry-url: "https://registry.npmjs.org"
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         with:
           name: opencode-cli
           path: packages/opencode/dist
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         with:
           name: opencode-cli-windows
           path: packages/opencode/dist
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        if: github.ref_name != 'v2'
+        if: github.ref_name != 'v2' && github.ref_name != 'beta'
         with:
           name: opencode-cli-signed-windows
           path: packages/opencode/dist
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        if: github.ref_name != 'beta'
         with:
           name: opencode-preview-cli
           path: packages/cli/dist
 
       - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        if: github.ref_name != 'beta'
         with:
           pattern: opencode-node-cli-*
           path: packages/cli/dist/node

.github/workflows/test.yml

@@ -55,6 +55,12 @@ jobs:
           git config --global user.email "bot@opencode.ai"
           git config --global user.name "opencode"
 
+      - name: Install ffmpeg
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install --yes ffmpeg
+
       - name: Cache Turbo
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:

.opencode/glossary/he.md

@@ -0,0 +1,48 @@
+# he Glossary
+
+## Sources
+
+- Hebrew Academy approved IT terminology: https://terms.hebrew-academy.org.il/Millonim/ShowMillon?KodMillon=192
+- Firefox Hebrew localization corpus: https://github.com/mozilla-l10n/firefox-l10n/tree/main/he
+- KDE Hebrew localization team and corpus: https://l10n.kde.org/team-infos.php?teamcode=he
+- Community-maintained VS Code Hebrew language pack: https://github.com/AMAARETS/vscode-language-pack-he
+- Microsoft Hebrew developer documentation for Git terminology: https://learn.microsoft.com/he-il/power-platform/alm/tutorials/github-actions-deploy
+- W3C guidance for bidirectional text: https://www.w3.org/International/articles/strings-and-bidi/
+
+## Do Not Translate (Locale Additions)
+
+- `OpenCode` (preserve casing in prose and UI copy)
+- `API`, `MCP`, `LSP`, `OAuth`, `Git`, model names, and provider names
+- Commands, flags, keyboard shortcuts, file paths, URLs, identifiers, hashes, and code literals
+- Keep `commit` and `diff` when they name the exact Git artifact or operation
+
+## Preferred Terms
+
+| English / Context | Preferred     | Notes                                                                     |
+| ----------------- | ------------- | ------------------------------------------------------------------------- |
+| session           | `הפעלה`       | Use `שיחה` only when the source specifically means a chat or conversation |
+| workspace         | `סביבת עבודה` |                                                                           |
+| terminal          | `מסוף`        | Prefer the established Hebrew term over transliteration                   |
+| command           | `פקודה`       |                                                                           |
+| provider          | `ספק`         | Use `ספק מודלים` where the bare noun is ambiguous                         |
+| model             | `מודל`        |                                                                           |
+| API key           | `מפתח API`    | Keep the acronym in Latin letters                                         |
+| plugin            | `תוסף`        |                                                                           |
+| repository        | `מאגר`        | Use `מאגר Git` where context is ambiguous                                 |
+| branch            | `ענף`         |                                                                           |
+| context           | `הקשר`        | Use `חלון הקשר` for context window                                        |
+| tokens            | `אסימונים`    |                                                                           |
+
+## Guidance
+
+- Prefer natural modern Israeli Hebrew over word-for-word translation or obscure coined terms.
+- Use short action verbs for controls and translate complete phrases in context.
+- Keep recognized developer acronyms and exact Git vocabulary in Latin script instead of phonetic transliteration.
+- Treat embedded code, paths, commands, shortcuts, hashes, model IDs, and other Latin technical artifacts as LTR content inside the RTL interface.
+- Keep recurring concepts consistent and do not collapse session, chat, run, and launch into one Hebrew term.
+
+## Avoid
+
+- Avoid transliterations such as `טרמינל`, `פלאגין`, and `קומנד` when `מסוף`, `תוסף`, and `פקודה` are clear.
+- Avoid translating `commit` as `התחייבות`.
+- Avoid inventing Hebrew expansions for `API`, `MCP`, or `LSP`.

.opencode/plugins/tui/discovery-smoke.ts

@@ -1,13 +0,0 @@
-import type { Context } from "../../../packages/plugin/src/tui/context"
-
-export default {
-  id: "test.tui-discovery-smoke",
-  setup(_context: Context) {
-    // context.ui.toast.show({
-    //   title: "TUI plugin discovery works",
-    //   message: "Loaded .opencode/plugins/tui/discovery-smoke.ts",
-    //   variant: "success",
-    //   duration: 30_000,
-    // })
-  },
-}

.opencode/skills/ideal-pseudocode/SKILL.md

@@ -1,68 +0,0 @@
----
-name: ideal-pseudocode
-description: Function-by-function refactoring loop driven by ideal pseudocode. Use when the user says "ideal pseudocode", asks to make a function read like its pseudocode, or wants a dense module cleaned up one function at a time.
----
-
-# Ideal Pseudocode
-
-Clean up one function at a time by writing the pseudocode it _should_ read as, naming every delta between that and the real code, and closing only the gaps the user approves.
-
-## Loop
-
-One function per round. Never touch code before the user picks a direction.
-
-1. **Pick the target** with the user — usually the next function up or down the call chain from the last round.
-2. **Read the current code** fresh from disk. It may have unsaved or parallel edits; ask before overwriting anything unexpected.
-3. **Distill.** Write the function's ideal pseudocode in a `ts`-fenced code block — TypeScript-flavored for syntax highlighting, but pseudocode: comments over mechanics, one line per idea, every arm of a loop visible as an arm. For a dense or unfamiliar function, first show the _current_ structure as pseudocode, then the ideal.
-4. **Name the deltas.** A numbered list; each delta is one concrete gap: control flow smuggled through combinators, a flag-typed parameter, colliding names, side-effect plumbing drowning a loop body, twin functions differing only by a filter. For each, give an honest cost/benefit and a decisive recommendation — including "leave it" when the density has reasons (state the reasons).
-5. **Verify before proposing.** Any API the proposal builds on (a combinator, schedule, matcher) gets checked against the installed version's source, not memory.
-6. **Get buy-in, then apply.** Minimal edits. Quarantine irreducible ugliness inside a small helper whose name is the pseudocode line it implements. Collapse shallow sibling functions into one deep function only when the distinguishing argument is a domain noun, not a behavior switch.
-7. **Validate.** Typecheck and the focused test suites from the package directory; rerun timing-sensitive tests repeatedly. Flag any semantic change the restructure introduced, even when tests pass.
-8. **Offer the next target** and stop.
-
-## Style
-
-- Keep each round tight: pseudocode block, deltas, recommendation, one question.
-- State mutation belongs in one visible place (usually the loop tail), never spread across error channels or callback closures.
-- Prefer early return and `if` chains for control flow; matchers are for producing values.
-- Compare names against the question each caller is actually asking; rename toward the domain noun.
-
-## Example round
-
-Every round should read like this (abridged from a real one, on an Effect step-runner):
-
-> `runStep` it is. The ideal pseudocode:
->
-> ```ts
-> // One logical step: one settled model call, however many attempts it takes.
-> // An attempt can end without settling in two ways:
-> //   transient provider failure -> retry the same call, same assistant message
-> //   compaction rewrote history -> rebuild the request and call again
-> function runStep(session, promotable, step) {
->   while (true) {
->     const result = callModel(session, promotable, step)
->
->     if (result.completed) return result
->
->     if (result.retryable) {
->       wait(retrySchedule) // exhausted -> publish Step.Failed, fail drain
->       promotable = none // never re-promote on a second attempt
->       continue
->     }
->
->     // compaction restarted the step: fresh request from rewritten history
->     promotable = none
->     step = result.step
->   }
-> }
-> ```
->
-> Comparing against the real thing, three deltas:
->
-> 1. **The retry arm doesn't live in the loop.** In the pseudocode, retry is just another loop arm. In the real code it's smuggled through the error channel: `Effect.tapError` mutating loop variables via `Effect.sync` closures, then `Effect.retryOrElse` re-running a suspended call. Four mutable variables mutated from three different places — the state flow is invisible unless you already know the trick.
-> 2. **`recoverOverflow: typeof compaction.compact | undefined`** — a function-typed parameter used as a one-shot armed/disarmed flag. The callee already closes over `compaction`; a boolean says what it is.
-> 3. **`assistantMessageID`** exists solely to resume the same durable message across retries — fine, but only legible once the retry arm is visible in the loop.
->
-> The unifying move for (1): a schedule driver, so a retryable failure becomes a caught value handled in the loop body exactly like the compaction arm. One honest trade: it's more manual than the blessed combinator — but the manual version is the loop the combinator was hiding, and the loop already exists for restarts. Two restart mechanisms, one control structure.
->
-> Want me to apply it — unified loop, simplified schedule input, boolean `recoverOverflow`?

.opencode/skills/opencode-drive/SKILL.md

@@ -1,254 +0,0 @@
----
-name: opencode-drive
-description: Use when an agent needs drive OpenCode via a script or interact with an isolated instance
----
-
-# OpenCode Drive
-
-Use `opencode-drive` to launch an isolated OpenCode instance and control it via commands or a script.
-
-There are two modes. Always default to using a script unless specifically directed to be interactive (connect
-to an existing running instance, or start a new one, and make a few changes to the UI and read it, and iterate
-on changes).
-
-Scripts allow you to run a full walkthrough in one run. When the script is done opencode-drive exits,
-stops all processes, and cleans up all artifacts.
-
-# Prepare The Environment
-
-Use `init` when files must be added to the isolated home or project before OpenCode starts. It prints the artifact directory without launching OpenCode. A later `start` with the same name reuses it.
-
-```bash
-artifacts=$(opencode-drive init --name demo)
-cp -R ./fixtures/home/. "$artifacts/"
-cp -R ./fixtures/project/. "$artifacts/files/"
-opencode-drive start --name demo --dev ~/projects/opencode
-```
-
-The simulated project is under `$artifacts/files`. Running `start` without a prior `init` initializes the artifacts automatically.
-
-# Scripted usage
-
-You can write scripts that walk through entire flows, and gives you full access to controlling
-the backend too. See examples of the script API at the bottom of this file.
-
-After creating or editing a script, always typecheck it before running. Never skip this step:
-
-```bash
-opencode-drive check ./reproduce-stale-exploring-empty.ts
-```
-
-Run it by passing `--script` to start:
-
-```bash
-opencode-drive start --name auto-stop-reproduction --script ./reproduce-stale-exploring-empty.ts
-```
-
-It will output information about the run, including paths to log files which you can read
-to inspect what happened. If you need to dig into failures that aren't clear, read those log
-files. If the script is unsuccessful, automatically fix the script and run it again.
-
-Scripts use one typed definition object. `setup` runs before OpenCode starts,
-and `fs.writeFile` always writes inside the simulated project.
-
-You can read the full typed API here: https://raw.githubusercontent.com/jlongster/opencode-drive/refs/heads/main/src/script/types.ts
-
-```ts
-import { defineScript } from "opencode-drive"
-
-export default defineScript({
-  async setup({ fs, config }) {
-    config.autoupdate = false
-    await fs.writeFile("src/example.ts", "export const value = 1\n")
-  },
-
-  async run({ ui, llm }) {
-    await ui.submit("Open src/example.ts")
-    await llm.send(llm.text("The file exports `value`."))
-    await ui.waitFor("The file exports `value`.")
-  },
-})
-```
-
-`setup` receives the current OpenCode config object, which starts from the
-default drive config unless the prepared instance already has one. When a script
-needs custom config, mutate this `config` parameter instead of generating and
-writing a new config object from scratch, so the script keeps the default
-provider/model settings unless it intentionally changes them.
-
-Note that the simulated model is a GPT model type, and opencode uses the `patch` tool for working with files Do not use a `edit` or `write` tool to edit files.
-
-Use `launch: "manual"` when the script needs to launch the server and every TUI
-itself (this is extremely rare, do not use this unless explicitly asked). In this
-mode `ui` is typed as `null`; call `server.launch()` exactly
-once before launching clients. Each `clients.launch(name)` result provides the
-same UI methods as the automatic client. You can see an example of this API
-here: https://raw.githubusercontent.com/jlongster/opencode-drive/refs/heads/main/examples/multiple-clients.ts
-
-Use the exported `wait(milliseconds)` utility for an unconditional delay.
-
-`await llm.send(...)` waits for the next request and resolves after OpenCode
-acknowledges its complete response. `llm.queue(...)` declares responses in
-advance. Chunks may be built with `text`, `reasoning`, `toolCall`, `raw`,
-`finish`, and `disconnect`. A normal response receives `finish("stop")`
-automatically unless it yields or queues an explicit terminal event.
-
-`llm.text(text, { delay, chunkSize })` defaults to a 2 ms delay and a
-15-character target varied by plus or minus 5 per chunk.
-
-`llm.reasoning` accepts the same options, and `llm.pause(milliseconds)` adds a
-delay between any two outputs.
-
-Use `llm.serve` for an ongoing typed response generator:
-
-```ts
-llm.serve(async function* (request, index) {
-  yield llm.reasoning(`Handling request ${index + 1}`)
-  yield llm.text(`Received ${request.id}`)
-  yield llm.finish("stop")
-})
-```
-
-The backend connection, response cleanup, cancellation, and recording
-completion are automatic.
-
-You can see some example scripts here:
-
-- https://raw.githubusercontent.com/jlongster/opencode-drive/refs/heads/main/examples/simple.ts
-- https://raw.githubusercontent.com/jlongster/opencode-drive/refs/heads/main/examples/serve.ts
-
-## Prune
-
-- `prune` removes artifact directories. These are always cleaned up after running a script
-  successfully, but leftover on failed runs. Always call this if a script fails.
-
-```bash
-opencode-drive prune --name demo
-
-// --force cleans up all artifcat directories
-opencode-dirve prune --force
-```
-
-# Live interaction usage
-
-- Always give headless instances a unique `--name`. Visible instances may omit it.
-- A normal headless `start` detaches automatically and returns after the instance is ready.
-- Do not add `&`; the long-running owner already runs in the background.
-- Configure simulated model responses after startup when needed.
-- Send ordered UI commands with `send`.
-- Always stop the instance when finished.
-
-```bash
-opencode-drive start --name demo
-
-opencode-drive send --name demo \
-  --command.ui.type '{"text":"Explain this project"}' \
-  --command.ui.enter
-
-opencode-drive stop --name demo
-```
-
-## Send UI Commands
-
-- Every `send` opens a connection to the named instance, runs its commands in order, and exits.
-- Combine typing and Enter in one command when submitting a prompt.
-- JSON-valued commands require one JSON argument.
-- Multiple command flags execute from left to right.
-
-Commands:
-
-- `--command.ui.type <json>` types into the focused editor. Arguments: `text` string.
-- `--command.ui.press <json>` presses a key. Arguments: `key` string; optional `modifiers` object with boolean `ctrl`, `shift`, `meta`, `super`, or `hyper`.
-- `--command.ui.enter` presses Enter. Arguments: none.
-- `--command.ui.arrow <json>` presses an arrow key. Arguments: `direction` is `up`, `down`, `left`, or `right`.
-- `--command.ui.focus <json>` focuses an element. Arguments: `target` is the numeric element `num` returned by `ui.state`.
-- `--command.ui.click <json>` clicks an element. Arguments: numeric `target`, `x`, and `y`; use the element `num` returned by `ui.state` as `target`.
-- `--command.ui.state` prints focus and interactive element metadata as JSON. Arguments: none.
-- `--command.ui.matches <json>` prints whether literal, case-sensitive text appears on screen. Arguments: `text` string.
-
-```bash
-opencode-drive send --name demo \
-  --command.ui.type '{"text":"Find the relevant code and explain it"}' \
-  --command.ui.enter
-
-opencode-drive send --name demo \
-  --command.ui.press '{"key":"p","modifiers":{"ctrl":true}}'
-
-opencode-drive send --name demo \
-  --command.ui.arrow '{"direction":"down"}'
-
-opencode-drive send --name demo \
-  --command.ui.focus '{"target":12}'
-
-opencode-drive send --name demo \
-  --command.ui.click '{"target":12,"x":4,"y":1}'
-
-opencode-drive send --name demo \
-  --command.ui.matches '{"text":"OpenCode"}'
-```
-
-To read the UI state and see information about interactable elements, use the `ui.state` command:
-
-```bash
-opencode-drive send --name demo --command.ui.state
-```
-
-## Configure LLM Responses
-
-- `responses` controls what the LLM responds with
-- Only use this if you are wanting to reproduce an exact type of response
-- Defaults are `text,reasoning,diff,tool` with `write,apply_patch`.
-- Supported types are `text`, `reasoning`, `diff`, and `tool`.
-- `--tools` limits generated tool calls to names offered by OpenCode.
-
-```bash
-opencode-drive responses --name demo \
-  --types text,reasoning,diff,tool \
-  --tools write,apply_patch
-
-opencode-drive responses --name demo \
-  --types tool \
-  --tools read,glob,grep
-```
-
-## Inspect The UI
-
-- `ui.state` prints focus and interactive element metadata as JSON.
-- `ui.matches` checks for literal, case-sensitive screen text.
-- `screenshot` prints the generated image path.
-
-```bash
-opencode-drive screenshot --name demo
-```
-
-## Lifecycle
-
-- `stop` waits for recording export and owner cleanup before returning.
-
-```bash
-opencode-drive stop --name demo
-```
-
-# Record The UI
-
-- Start with `--record` to capture a headless instance from its first rendered frame.
-- `stop` finishes the recording, exports an MP4, and prints its path.
-
-```bash
-opencode-drive start --name demo --record
-
-opencode-drive send --name demo \
-  --command.ui.type '{"text":"Show me the current architecture"}' \
-  --command.ui.enter
-
-opencode-drive stop --name demo
-```
-
-# Artifacts dir
-
-- `dir` prints the artifact directory for the instance.
-
-```bash
-opencode-drive dir --name demo
-```
-

.opencode/skills/rtl-aware-development/SKILL.md

@@ -0,0 +1,63 @@
+---
+name: rtl-aware-development
+description: OpenCode Desktop should be RTL-aware. Use when implementing or reviewing RTL/LTR behavior in the web app, desktop app, CSS, menus, scrolling, resizing, icons, mixed-direction text, or Electron title bars.
+---
+
+# RTL-Aware Development
+
+Treat direction as independent from language. Test English in both directions as well as real RTL and mixed-script content.
+
+## Guidelines
+
+- Set `lang` and `dir` on the document, and propagate direction through component providers used by portaled menus and popovers. Do not change the selected locale merely to force RTL.
+- Keep DOM and focus order semantic. Flexbox and Grid already follow `dir`; do not add `row-reverse`, CSS `order`, or reversed markup just to mirror a layout.
+- Prefer logical CSS for semantic layout. Reserve physical coordinates for pointer positions, canvas geometry, native window controls, and other genuinely physical placement.
+
+```css
+/* Avoid */
+padding-left: 12px;
+right: 0;
+border-right: 1px solid;
+text-align: left;
+
+/* Prefer */
+padding-inline-start: 12px;
+inset-inline-end: 0;
+border-inline-end: 1px solid;
+text-align: start;
+```
+
+- Isolate mixed-direction text. Use `dir="auto"` or `<bdi>` for unknown text; keep code, URLs, IDs, and filesystem paths LTR without forcing the surrounding component LTR.
+
+```html
+<span class="file-row"><bdi dir="auto">README.md</bdi></span> <bdi dir="ltr"><code>C:\src\app.ts</code></bdi>
+```
+
+- Mirror directional meaning, not every image. Back/forward, previous/next, disclosure, indentation, and directional progress may need mirroring. Do not mirror brands, clocks, media controls, charts, or text. Reverse physical gradients, `translateX`, SVG transforms, and animation deltas explicitly.
+- Map interactions through direction. `clientX` remains physical; resizing a logical edge needs an RTL-aware delta. Logical previous/next keyboard controls may swap ArrowLeft/ArrowRight. Follow the relevant WAI-ARIA widget pattern.
+- Do not assume LTR scrolling. RTL `scrollLeft` can start at `0` and become negative. Prefer `scrollIntoView({ inline: "nearest" })` or a tested direction-normalizing helper.
+- For Electron title bars, prefer native caption controls and use `titleBarOverlay` plus `env(titlebar-area-*)` for the safe content rectangle. Keep Windows/macOS native-control avoidance and `trafficLightPosition` physical; keep app navigation inside that rectangle logical. Mark interactive titlebar children `app-region: no-drag`.
+- Verify behavior, not screenshots alone. Check computed styles, pseudo-element geometry, hit zones, focus order, keyboard behavior, submenu direction, zoom/scaling, and both LTR and RTL scroll endpoints.
+
+## Test Matrix
+
+- English + LTR
+- English + forced RTL
+- A real RTL locale + RTL
+- Mixed RTL/LTR content, long labels, numbers, code, and paths
+- Keyboard, pointer resize, scrolling, menus/submenus, and Electron titlebar controls in both directions
+
+## References
+
+- [RTL Styling 101, Ahmad Shadeed](https://rtlstyling.com/posts/rtl-styling/)
+- [CSS-Tricks: RTL Styling 101](https://css-tricks.com/rtl-styling-101/)
+- [CSS-Tricks: CSS Logical Properties and Values](https://css-tricks.com/css-logical-properties-and-values/)
+- [W3C: Structural markup and right-to-left text](https://www.w3.org/International/questions/qa-html-dir)
+- [W3C: Inline bidirectional markup](https://www.w3.org/International/articles/inline-bidi-markup/)
+- [MDN: CSS logical properties and values](https://developer.mozilla.org/en-US/docs/Web/CSS/Guides/Logical_properties_and_values)
+- [MDN: `dir`](https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Global_attributes/dir)
+- [MDN: `scrollLeft`](https://developer.mozilla.org/en-US/docs/Web/API/Element/scrollLeft)
+- [web.dev: Logical properties](https://web.dev/learn/css/logical-properties/)
+- [Electron: Custom title bar](https://www.electronjs.org/docs/latest/tutorial/custom-title-bar)
+- [WAI-ARIA: Window splitter pattern](https://www.w3.org/WAI/ARIA/apg/patterns/windowsplitter/)
+- [Kobalte: I18n Provider](https://kobalte.dev/docs/core/components/i18n-provider/)

.opencode/tool/github-triage.ts

@@ -4,7 +4,7 @@ import { tool } from "@opencode-ai/plugin"
 const TEAM = {
   tui: ["kommander", "simonklee"],
   desktop_web: ["Hona", "Brendonovich"],
-  core: ["jlongster", "rekram1-node", "nexxeln", "kitlangton"],
+  core: ["jlongster", "rekram1-node", "neriousy", "nexxeln", "kitlangton"],
   inference: ["fwang", "MrMushrooooom", "starptech"],
   windows: ["Hona"],
 } as const

AGENTS.md

@@ -1,8 +1,9 @@
 - After changing the public Protocol or Server `HttpApi`, run `bun run generate` from `packages/client`. Do not edit `src/generated` or `src/generated-effect` directly.
 - Keep runtime dependencies directed from Schema to Core and Protocol, then from Core and Protocol to Server. Client runtime code may depend on Schema and Protocol but never Core or Server; `sdk-next` composes Client, Core, and Server.
 - Do not modify `packages/opencode` unless the user explicitly asks for V1 work. `packages/opencode` is the V1 implementation and is present for reference only. New implementation changes should land in the V2 package set: `packages/core`, `packages/cli`, `packages/server`, `packages/protocol`, `packages/schema`, and related generated client surfaces when required.
-- The default branch in this repo is `dev`.
-- Local `main` ref may not exist; use `dev` or `origin/dev` for diffs.
+- The default branch in this repo is `v2`.
+- Base all new branches and worktrees on `v2`, or `origin/v2` when the local `v2` ref is unavailable. Do not base them on `dev`.
+- Local `main` ref may not exist; use `v2` or `origin/v2` for diffs.
 
 ## Live V2 TUI Testing
 
@@ -11,6 +12,20 @@
 - The script discovers the server with `opencode2 service status`, injects its private local credential from `opencode2 service get password`, and uses the `next` TUI storage channel so tabs and other client-local state match the installed client.
 - Prefer `dev:live` over plain `bun run dev` for this workflow. An implicit managed-service connection may replace the live server when the worktree client version differs; explicit `--server` warns and continues without replacing it.
 
+## V2 TUI Stories
+
+- When a user asks for a TUI story, add a fixture-driven story under `packages/tui/src/feature-plugins/system/storybook` and register it in `index.tsx`.
+- Render the real production component rather than a visual copy. Keep submissions and other side effects local to the story so it is safe to explore repeatedly.
+- Expose the meaningful state dimensions through story keybindings and list them in `StoryFooter`; include a reset command when combinations can leave the fixture in a confusing state.
+- Run a specific story with `OPENCODE_STORY=<story-id> bun run dev:live` from the development worktree, and exercise narrow and wide terminal sizes when layout is relevant.
+
+## TUI Theme Tokens
+
+- Choose theme tokens by semantic role, not by their current color. Do not use raw `theme.hue` values or borrow an unrelated semantic token to achieve a preferred appearance.
+- Use `text.feedback` and `background.feedback` only for outcome or status feedback such as errors, warnings, success messages, and informational messages. Use `formfield` states for form-control text, ordinals, and selection markers, and `action` states for actions.
+- If the theme does not expose a token for the required semantic role, extend the theme schema, defaults, resolution, and types with that role before using it in a component. Do not repurpose the nearest-looking existing token.
+- When changing the public theme token surface, verify the built-in light and dark defaults and the custom-theme fallback path in addition to the affected TUI component.
+
 ## Branch Names
 
 Use a short branch name of at most three words, separated by hyphens. Do not use slashes or type prefixes such as `feat/` or `fix/`.
@@ -170,4 +185,4 @@ const table = sqliteTable("session", {
 - One step is one logical LLM call; its durable record covers only the model-visible span. Do not write "provider turn", and do not use bare "turn" for a single call: "turn" is reserved for the future assistant-turn unit containing all steps from prompt promotion until the session would go idle.
 - Keep EventV2 replay owner claims separate from clustered Session execution ownership.
 - Keep the Instructions algebra and built-ins in `src/instructions`; keep instruction producers with their observed domains, and keep Session History selection plus `InstructionState` and `InstructionEntry` persistence Session-owned. `InstructionDiscovery` observes ambient global and upward-project instructions. The runner composes built-ins, discovery, guidance, and entries explicitly in `loadInstructions`; there is no instruction registry.
-- `session.instructions.updated` stores only changed source keys and content hashes. Blob values live once in `instruction_blob`; `instruction_state` is a rebuildable fold cache, never primary state. Render initial instructions and chronological updates from values during request assembly. Completed compaction moves the instruction epoch; Session movement and committed revert clear it. Unavailable sources retain the last value and block only the initial complete delta.
+- `session.instructions.updated` stores only changed source keys and content hashes. Blob values live once in `instruction_blob`; `instruction_state` is a rebuildable fold cache, never primary state. Render initial instructions and chronological updates from values during request assembly. Completed compaction moves the instruction epoch; Session movement retains it so destination instruction changes are chronological, while committed revert clears it. Unavailable sources retain the last value and block only the initial complete delta.

CONTRIBUTING.md

@@ -252,33 +252,6 @@ These are not strictly enforced, they are just general guidelines:
 
 For net-new functionality, start with a design conversation. Open an issue describing the problem, your proposed approach (optional), and why it belongs in OpenCode. The core team will help decide whether it should move forward; please wait for that approval instead of opening a feature PR directly.
 
-## Trust & Vouch System
-
-This project uses [vouch](https://github.com/mitchellh/vouch) to manage contributor trust. The vouch list is maintained in [`.github/VOUCHED.td`](.github/VOUCHED.td).
-
-### How it works
-
-- **Vouched users** are explicitly trusted contributors.
-- **Denounced users** are explicitly blocked. Issues and pull requests from denounced users are automatically closed. If you have been denounced, you can request to be unvouched by reaching out to a maintainer on [Discord](https://opencode.ai/discord)
-- **Everyone else** can participate normally — you don't need to be vouched to open issues or PRs.
-
-### For maintainers
-
-Collaborators with write access can manage the vouch list by commenting on any issue:
-
-- `vouch` — vouch for the issue author
-- `vouch @username` — vouch for a specific user
-- `denounce` — denounce the issue author
-- `denounce @username` — denounce a specific user
-- `denounce @username <reason>` — denounce with a reason
-- `unvouch` / `unvouch @username` — remove someone from the list
-
-Changes are committed automatically to `.github/VOUCHED.td`.
-
-### Denouncement policy
-
-Denouncement is reserved for users who repeatedly submit low-quality AI-generated contributions, spam, or otherwise act in bad faith. It is not used for disagreements or honest mistakes.
-
 ## Issue Requirements
 
 All issues **must** use one of our issue templates:

infra/console.ts

@@ -288,8 +288,12 @@ new sst.cloudflare.x.SolidStart("Console", {
     server: {
       placement: { region: "aws:us-east-2" },
       transform: {
-        worker: {
-          tailConsumers: [{ service: logProcessor.nodes.worker.scriptName }],
+        worker: (args) => {
+          args.compatibilityFlags = $resolve(args.compatibilityFlags).apply((flags) => [
+            ...(flags ?? []),
+            "global_fetch_strictly_public",
+          ])
+          args.tailConsumers = [{ service: logProcessor.nodes.worker.scriptName }]
         },
       },
     },

install

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 set -euo pipefail
-APP=opencode
+APP=opencode2
 
 MUTED='\033[0;2m'
 RED='\033[0;31m'
@@ -15,14 +15,14 @@ Usage: install.sh [options]
 
 Options:
     -h, --help              Display this help message
-    -v, --version <version> Install a specific version (e.g., 1.0.180)
+    -v, --version <version> Install a specific version (e.g., 0.0.0-next-17236)
     -b, --binary <path>     Install from a local binary instead of downloading
         --no-modify-path    Don't modify shell config files (.zshrc, .bashrc, etc.)
 
 Examples:
-    curl -fsSL https://opencode.ai/install | bash
-    curl -fsSL https://opencode.ai/install | bash -s -- --version 1.0.180
-    ./install --binary /path/to/opencode
+    curl -fsSL https://raw.githubusercontent.com/anomalyco/opencode/v2/install | bash
+    curl -fsSL https://raw.githubusercontent.com/anomalyco/opencode/v2/install | bash -s -- --version 0.0.0-next-17236
+    ./install --binary /path/to/opencode2
 EOF
 }
 
@@ -109,11 +109,6 @@ else
         ;;
     esac
 
-    archive_ext=".zip"
-    if [ "$os" = "linux" ]; then
-      archive_ext=".tar.gz"
-    fi
-
     is_musl=false
     if [ "$os" = "linux" ]; then
       if [ -f /etc/alpine-release ]; then
@@ -165,42 +160,42 @@ else
       target="$target-musl"
     fi
 
-    filename="$APP-$target$archive_ext"
-
-
-    if [ "$os" = "linux" ]; then
-        if ! command -v tar >/dev/null 2>&1; then
-             echo -e "${RED}Error: 'tar' is required but not installed.${NC}"
-             exit 1
-        fi
-    else
-        if ! command -v unzip >/dev/null 2>&1; then
-            echo -e "${RED}Error: 'unzip' is required but not installed.${NC}"
-            exit 1
-        fi
+    if ! command -v tar >/dev/null 2>&1; then
+        echo -e "${RED}Error: 'tar' is required but not installed.${NC}"
+        exit 1
     fi
 
     if [ -z "$requested_version" ]; then
-        url="https://github.com/anomalyco/opencode/releases/latest/download/$filename"
-        specific_version=$(curl -s https://api.github.com/repos/anomalyco/opencode/releases/latest | sed -n 's/.*"tag_name": *"v\([^"]*\)".*/\1/p')
+        metadata=$(curl -fsSL https://registry.npmjs.org/@opencode-ai%2fcli/next || true)
+        specific_version=$(echo "$metadata" | sed -n 's/.*"version":"\([^"]*\)".*/\1/p')
 
-        if [[ $? -ne 0 || -z "$specific_version" ]]; then
+        if [ -z "$specific_version" ]; then
             echo -e "${RED}Failed to fetch version information${NC}"
             exit 1
         fi
     else
         # Strip leading 'v' if present
         requested_version="${requested_version#v}"
-        url="https://github.com/anomalyco/opencode/releases/download/v${requested_version}/$filename"
         specific_version=$requested_version
+    fi
 
-        # Verify the release exists before downloading
-        http_status=$(curl -sI -o /dev/null -w "%{http_code}" "https://github.com/anomalyco/opencode/releases/tag/v${requested_version}")
-        if [ "$http_status" = "404" ]; then
-            echo -e "${RED}Error: Release v${requested_version} not found${NC}"
-            echo -e "${MUTED}Available releases: https://github.com/anomalyco/opencode/releases${NC}"
-            exit 1
-        fi
+    package_name="@opencode-ai/cli-$target"
+    http_status=$(curl -s -o /dev/null -w "%{http_code}" "https://registry.npmjs.org/@opencode-ai%2fcli-$target/$specific_version" || true)
+    if [ "$http_status" = "404" ]; then
+        echo -e "${RED}Error: Version ${specific_version} is not available for $target${NC}"
+        echo -e "${MUTED}Available versions: https://www.npmjs.com/package/$package_name?activeTab=versions${NC}"
+        exit 1
+    fi
+    if [ "$http_status" != "200" ]; then
+        echo -e "${RED}Failed to fetch package information${NC}"
+        exit 1
+    fi
+
+    filename="cli-$target-$specific_version.tgz"
+    url="https://registry.npmjs.org/$package_name/-/$filename"
+    binary_name="$APP"
+    if [ "$os" = "windows" ]; then
+        binary_name="$APP.exe"
     fi
 fi
 
@@ -219,11 +214,13 @@ print_message() {
 }
 
 check_version() {
-    if command -v opencode >/dev/null 2>&1; then
-        opencode_path=$(which opencode)
+    if command -v "$APP" >/dev/null 2>&1; then
+        opencode_path=$(which "$APP")
 
         ## Check the installed version
-        installed_version=$(opencode --version 2>/dev/null || echo "")
+        installed_version=$("$APP" --version 2>/dev/null || echo "")
+        installed_version="${installed_version##* }"
+        installed_version="${installed_version#v}"
 
         if [[ "$installed_version" != "$specific_version" ]]; then
             print_message info "${MUTED}Installed version: ${NC}$installed_version."
@@ -287,7 +284,7 @@ download_with_progress() {
     trap "trap - RETURN; rm -f \"$tracefile\"; printf '\033[?25h' >&4; exec 4>&-" RETURN
 
     (
-        curl --trace-ascii "$tracefile" -s -L -o "$output" "$url"
+        curl --trace-ascii "$tracefile" -fsL -o "$output" "$url"
     ) &
     local curl_pid=$!
 
@@ -325,30 +322,25 @@ download_with_progress() {
 }
 
 download_and_install() {
-    print_message info "\n${MUTED}Installing ${NC}opencode ${MUTED}version: ${NC}$specific_version"
+    print_message info "\n${MUTED}Installing ${NC}$APP ${MUTED}version: ${NC}$specific_version"
     local tmp_dir="${TMPDIR:-/tmp}/opencode_install_$$"
     mkdir -p "$tmp_dir"
 
     if [[ "$os" == "windows" ]] || ! [ -t 2 ] || ! download_with_progress "$url" "$tmp_dir/$filename"; then
         # Fallback to standard curl on Windows, non-TTY environments, or if custom progress fails
-        curl -# -L -o "$tmp_dir/$filename" "$url"
+        curl -f -# -L -o "$tmp_dir/$filename" "$url"
     fi
 
-    if [ "$os" = "linux" ]; then
-        tar -xzf "$tmp_dir/$filename" -C "$tmp_dir"
-    else
-        unzip -q "$tmp_dir/$filename" -d "$tmp_dir"
-    fi
-
-    mv "$tmp_dir/opencode" "$INSTALL_DIR"
-    chmod 755 "${INSTALL_DIR}/opencode"
+    tar -xzf "$tmp_dir/$filename" -C "$tmp_dir"
+    mv "$tmp_dir/package/bin/$binary_name" "$INSTALL_DIR"
+    chmod 755 "${INSTALL_DIR}/$binary_name"
     rm -rf "$tmp_dir"
 }
 
 install_from_binary() {
-    print_message info "\n${MUTED}Installing ${NC}opencode ${MUTED}from: ${NC}$binary_path"
-    cp "$binary_path" "${INSTALL_DIR}/opencode"
-    chmod 755 "${INSTALL_DIR}/opencode"
+    print_message info "\n${MUTED}Installing ${NC}$APP ${MUTED}from: ${NC}$binary_path"
+    cp "$binary_path" "${INSTALL_DIR}/$APP"
+    chmod 755 "${INSTALL_DIR}/$APP"
 }
 
 if [ -n "$binary_path" ]; then
@@ -453,8 +445,8 @@ echo -e ""
 echo -e "${MUTED}OpenCode includes free models, to start:${NC}"
 echo -e ""
 echo -e "cd <project>  ${MUTED}# Open directory${NC}"
-echo -e "opencode      ${MUTED}# Run command${NC}"
+echo -e "opencode2     ${MUTED}# Run command${NC}"
 echo -e ""
-echo -e "${MUTED}For more information visit ${NC}https://opencode.ai/docs"
+echo -e "${MUTED}For more information visit ${NC}https://opencode.ai/v2/docs"
 echo -e ""
 echo -e ""

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-RFek0QoEEjsgbqmTE/SxQAmPtYyzs0IPR2ugFn5Okrs=",
-    "aarch64-linux": "sha256-BmAxapY1YrAFn7mVq3/6A9+6Au5UIvSqBboHMkyJH3I=",
-    "aarch64-darwin": "sha256-Sx3bGWQqLlgoa/RudJxanjSzhFRNklckT2ffnO2I5F4=",
-    "x86_64-darwin": "sha256-CMOhiisHNowg06qadvgg4K+60zrynglwiT0qKYQ4NiA="
+    "x86_64-linux": "sha256-uduwrM143NDSc+tXsi4lVVfoMll2a3BDHRUjuO7GB68=",
+    "aarch64-linux": "sha256-6DUda78XdXY6DP86lIUkweSjys3iG4Y4mo1PiaNuXbg=",
+    "aarch64-darwin": "sha256-AkJwfLULLZVwwz+XU1QcFUZoIS7oVPCn+n/MXEaxrqE=",
+    "x86_64-darwin": "sha256-hAxKGdiITTxQ2uujQt6prNjo3NxGAMMeo+9HlMWK6GU="
   }
 }

package.json

@@ -46,9 +46,9 @@
       "@octokit/rest": "22.0.0",
       "@hono/standard-validator": "0.2.0",
       "@hono/zod-validator": "0.4.2",
-      "@opentui/core": "0.4.5",
-      "@opentui/keymap": "0.4.5",
-      "@opentui/solid": "0.4.5",
+      "@opentui/core": "0.5.3",
+      "@opentui/keymap": "0.5.3",
+      "@opentui/solid": "0.5.3",
       "@tanstack/solid-virtual": "3.13.32",
       "@shikijs/stream": "4.2.0",
       "@standard-schema/spec": "1.1.0",
@@ -78,7 +78,7 @@
       "fuzzysort": "3.1.0",
       "get-east-asian-width": "1.6.0",
       "luxon": "3.6.1",
-      "marked": "17.0.6",
+      "marked": "18.0.7",
       "marked-shiki": "1.2.1",
       "remend": "1.3.0",
       "@playwright/test": "1.59.1",
@@ -100,6 +100,7 @@
       "@sentry/solid": "10.36.0",
       "@sentry/vite-plugin": "4.6.0",
       "solid-js": "1.9.10",
+      "solid-sonner": "0.3.1",
       "vite-plugin-solid": "2.11.10",
       "@lydell/node-pty": "1.2.0-beta.12"
     }
@@ -119,7 +120,8 @@
     "prettier": "3.6.2",
     "semver": "^7.6.0",
     "sst": "catalog:",
-    "turbo": "2.10.2"
+    "turbo": "2.10.2",
+    "vitest": "4.1.10"
   },
   "dependencies": {
     "@aws-sdk/client-s3": "3.933.0",
@@ -157,6 +159,8 @@
     "effect": "catalog:"
   },
   "patchedDependencies": {
+    "@ai-sdk/openai-compatible@2.0.41": "patches/@ai-sdk%2Fopenai-compatible@2.0.41.patch",
+    "@dnd-kit/dom@0.5.0": "patches/@dnd-kit%2Fdom@0.5.0.patch",
     "@npmcli/agent@4.0.2": "patches/@npmcli%2Fagent@4.0.2.patch",
     "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
     "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
@@ -168,6 +172,7 @@
     "@pierre/trees@1.0.0-beta.4": "patches/@pierre%2Ftrees@1.0.0-beta.4.patch",
     "@modelcontextprotocol/sdk@1.29.0": "patches/@modelcontextprotocol%2Fsdk@1.29.0.patch",
     "effect@4.0.0-beta.101": "patches/effect@4.0.0-beta.101.patch",
-    "@tanstack/virtual-core@3.17.3": "patches/@tanstack%2Fvirtual-core@3.17.3.patch"
+    "@tanstack/virtual-core@3.17.3": "patches/@tanstack%2Fvirtual-core@3.17.3.patch",
+    "@ff-labs/fff-bun@0.10.1": "patches/@ff-labs%2Ffff-bun@0.10.1.patch"
   }
 }

packages/ai/AGENTS.md

@@ -80,7 +80,7 @@ Route defaults are request-shaping defaults such as `headers`, `limits`, `genera
 
 The four-axis decomposition is the reason DeepSeek, TogetherAI, Cerebras, Baseten, Fireworks, and DeepInfra all reuse `OpenAIChat.protocol` verbatim — each provider deployment is a 5-15 line `Route.make(...)` call instead of a 300-400 line route clone. Bug fixes in one protocol propagate to every consumer of that protocol in a single commit.
 
-When a provider ships a non-HTTP transport (OpenAI's WebSocket Responses backend, hypothetical bidirectional streaming APIs), the seam is `Transport` — `WebSocketTransport.jsonTransport.with(...)` constructs an IO template whose `prepare` receives the route endpoint/auth at compile time, builds a WebSocket URL and message, and whose `frames` yields decoded text from the socket. Same protocol and endpoint source, different transport.
+When a provider supports multiple physical transports, selection remains execution policy below its semantic route. `OpenResponsesChannel.transport(...)` owns the provider-neutral Responses WebSocket concept: it prepares one final request, executes HTTP by default, strips WebSocket-disallowed fields, and passes a generic channel exchange to a per-call `WebSocketChannelExecutor` when supplied. Provider-specific Responses routes opt in with handshake and connection-age policy. `Route.streamPrepared` owns decoding and acknowledges channel completion only after successful full consumption.
 
 ### URL Construction
 
@@ -106,7 +106,7 @@ const proxied = gateway.model("openai/gpt-4o-mini")
 Keep provider facades small and explicit:
 
 - Use branded `ProviderID.make(...)` and `ModelID.make(...)` where ids are constructed directly.
-- Use `model` for the default API path and named methods for provider-native alternatives such as OpenAI `responses`, `responsesWebSocket`, and `chat`.
+- Use `model` for the default API path and named methods for provider-native alternatives such as OpenAI `responses` and `chat`.
 - Put provider-specific setup on `.configure(...)`; do not add `model(id, overrides)` as a duplicate construction path.
 - Export lower-level `routes` arrays separately only when advanced internal wiring needs them.
 - Prefer `apiKey` as provider-specific sugar and `auth` as the explicit override; keep them mutually exclusive in provider option types with `ProviderAuthOption`.
@@ -124,11 +124,10 @@ import { model } from "@opencode-ai/ai/providers/openai/responses"
 
 const selected = model("gpt-5", {
   apiKey,
-  transport: "websocket",
 })
 ```
 
-Keep semantic APIs as separate entrypoints, such as OpenAI `chat` and `responses`. Keep transport choices inside the semantic entrypoint settings, so OpenAI Responses HTTP and WebSocket share one entrypoint. Provider facades may still expose named selectors such as `responsesWebSocket` for direct typed call sites; the package-like contract maps its settings to those selectors before returning an executable `LanguageModel`.
+Keep semantic APIs as separate entrypoints, such as OpenAI `chat` and `responses`. Transport is execution policy: OpenAI Responses uses HTTP by default and may receive a per-call WebSocket channel executor through `StreamOptions` without changing model or route identity.
 
 Do not expose `Route` in provider package settings. Route composition stays an implementation detail behind `model(...)`.
 
@@ -154,14 +153,16 @@ packages/ai/src/
     auth-options.ts         ProviderAuthOption shape, AuthOptions.bearer, AtLeastOne helper
     framing.ts              Framing type + Framing.sse
     transport/              transport implementations
-      index.ts              Transport type + HttpTransport / WebSocketTransport namespaces
+      index.ts              Transport execution types + HttpTransport / WebSocketTransport namespaces
+      websocket-channel.ts  generic sequential channel executor/driver contract
       http.ts               HttpTransport.httpJson — POST + framing
-      websocket.ts          WebSocketTransport.json + WebSocketExecutor service
+      websocket.ts          direct one-request channel executor + raw socket adapter
   protocols/
     shared.ts               ProviderShared toolkit used inside protocol impls
     openai-chat.ts          protocol + route (compose OpenAIChat.protocol)
     open-responses.ts         provider-neutral Responses protocol baseline
-    openai-responses.ts       OpenAI tools/events/transports composed over OpenResponses
+    open-responses-channel.ts provider-neutral Responses WebSocket transport factory
+    openai-responses.ts       OpenAI tools/events and channel policy composed over OpenResponses
     anthropic-messages.ts
     gemini.ts
     bedrock-converse.ts

packages/ai/README.md

@@ -315,7 +315,6 @@ import { model } from "@opencode-ai/ai/providers/openai/responses"
 
 const selected = model("gpt-5", {
   apiKey: process.env.OPENAI_API_KEY,
-  transport: "websocket",
   headers: { "x-application": "opencode" },
   limits: { context: 200_000, output: 64_000 },
 })
@@ -332,7 +331,7 @@ OpenAI Chat and OpenAI Responses are separate semantic entrypoints:
 - `@opencode-ai/ai/providers/google-vertex/responses`
 - `@opencode-ai/ai/providers/google-vertex/messages`
 
-Responses HTTP versus WebSocket is a scoped `transport` setting on the OpenAI Responses entrypoint, not another entrypoint. Azure follows the same Chat/Responses split at `providers/azure/chat` and `providers/azure/responses`. Generic OpenAI-compatible Chat remains at `providers/openai-compatible`; the Responses adapter at `providers/openai-compatible/responses` uses the provider-neutral Open Responses protocol. OpenAI Responses extends that baseline with OpenAI tools, event variants, metadata, defaults, and transports. Generic Anthropic Messages-compatible providers use `providers/anthropic-compatible`, which the named Anthropic provider composes. Google Gemini and Amazon Bedrock expose their single native API through their existing provider paths.
+OpenAI Responses has one semantic route and uses HTTP by default. Advanced callers may supply a per-call WebSocket channel executor through `StreamOptions`; transport policy does not change provider settings, model identity, or route identity. The provider-neutral Open Responses implementation owns the reusable WebSocket request and event contract, while each provider opts in with its own handshake and connection policy. Azure follows the same Chat/Responses split at `providers/azure/chat` and `providers/azure/responses`. Generic OpenAI-compatible Chat remains at `providers/openai-compatible`; the Responses adapter at `providers/openai-compatible/responses` uses the provider-neutral Open Responses protocol. OpenAI Responses extends that baseline with OpenAI tools, event variants, metadata, and defaults. Generic Anthropic Messages-compatible providers use `providers/anthropic-compatible`, which the named Anthropic provider composes. Google Gemini and Amazon Bedrock expose their single native API through their existing provider paths.
 
 Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages are separate API entrypoints. All accept `project`, `location`, and an optional `accessToken`; when no explicit token or auth override is supplied they lazily use Google Application Default Credentials. Vertex Gemini instead selects express mode when `apiKey` or `GOOGLE_VERTEX_API_KEY` is present. Vertex Chat targets MaaS models through the OpenAI-compatible Chat Completions endpoint, while Vertex Responses targets Grok models and defaults `store` to `false` as required by Vertex. `providers/google-vertex` remains the default alias for `providers/google-vertex/gemini`.
 
@@ -368,11 +367,12 @@ Other provider exports listed above remain direct facades until they explicitly
 
 ## Provider options & HTTP overlays
 
-Three escape hatches in order of stability:
+Request options in order of stability:
 
 1. **`generation`** — portable knobs (`maxTokens`, `temperature`, `topP`, `topK`, penalties, seed, stop).
-2. **`providerOptions: { <provider>: {...} }`** — typed-at-the-facade provider-specific knobs (OpenAI `promptCacheKey`, Anthropic `thinking`, Gemini `thinkingConfig`, OpenRouter routing).
-3. **`http: { body, headers, query }`** — last-resort serializable overlays merged into the final HTTP request. Reach for this only when a stable typed path doesn't yet exist.
+2. **`promptCacheKey`** — stable cache affinity lowered by every protocol that supports it.
+3. **`providerOptions: { <provider>: {...} }`** — typed-at-the-facade provider-specific knobs (OpenAI `store`, Anthropic `thinking`, Gemini `thinkingConfig`, OpenRouter routing).
+4. **`http: { body, headers, query }`** — last-resort serializable overlays merged into the final HTTP request. Reach for this only when a stable typed path doesn't yet exist.
 
 Route/provider defaults are overridden by request-level values for each axis.
 

packages/ai/STATUS.md

@@ -1,6 +1,6 @@
 # LLM Provider Parity Status
 
-Last reviewed: 2026-07-24
+Last reviewed: 2026-08-07
 
 This file tracks the gap between the native `@opencode-ai/ai` package and the AI SDK provider packages that opencode still depends on for many catalog/runtime paths.
 
@@ -16,8 +16,7 @@ This file tracks the gap between the native `@opencode-ai/ai` package and the AI
 | Native slice                       | Source                                                                                                                            | Current state                                                                                                                                                            | Main gaps                                                                                                                                                                                                                                                 |
 | ---------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | OpenAI Chat                        | `src/protocols/openai-chat.ts`, `src/providers/openai.ts`                                                                         | Usable. Streams text, reasoning deltas, tool calls, usage, images, and common generation controls.                                                                       | No typed structured-output / `response_format` path. Limited typed OpenAI option surface compared with SDK escape hatches.                                                                                                                                |
-| OpenAI Responses HTTP              | `src/protocols/open-responses.ts`, `src/protocols/openai-responses.ts`, `src/providers/openai.ts`                                 | Usable. Extends the Open Responses baseline with hosted-tool event surfacing, reasoning replay metadata, GPT-5 defaults, and cache usage.                                | No explicit `previous_response_id` path. Typed options cover only a subset of Responses fields. Structured output is still mostly synthetic-tool based.                                                                                                   |
-| OpenAI Responses WebSocket         | `src/protocols/openai-responses.ts`, `src/route/transport/websocket.ts`                                                           | Present as `OpenAI.responsesWebSocket(...)`.                                                                                                                             | Runner/catalog support explicitly must not downgrade WebSocket routes; broader runtime selection is not complete.                                                                                                                                         |
+| OpenAI Responses                   | `src/protocols/open-responses.ts`, `src/protocols/openai-responses.ts`, `src/providers/openai.ts`                                 | Usable over HTTP by default, with optional per-call WebSocket channel execution on the same model and route identity.                                                    | No incremental `previous_response_id` path or persistent Session channel manager yet. Typed options cover only a subset of Responses fields. Structured output is still mostly synthetic-tool based.                                                      |
 | OpenAI-compatible Chat             | `src/protocols/openai-compatible-chat.ts`, `src/providers/openai-compatible.ts`                                                   | Usable for generic Chat and several profiles: Baseten, Cerebras, DeepInfra, DeepSeek, Fireworks, Groq, TogetherAI.                                                       | Family quirks are mostly endpoint defaults, not full typed behavior.                                                                                                                                                                                      |
 | Open Responses-compatible          | `src/protocols/open-responses.ts`, `src/protocols/openai-compatible-responses.ts`, `src/providers/openai-compatible-responses.ts` | Usable for deployments that implement the provider-neutral Open Responses protocol. The deployment adapter does not inherit OpenAI tools, events, metadata, or defaults. | No named family profiles or recorded deployment coverage yet.                                                                                                                                                                                             |
 | Anthropic-compatible Messages      | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic-compatible.ts`                                                    | Usable for deployments that implement the Anthropic Messages wire protocol. Named Anthropic composes this base; MiniMax M3 has recorded text and tool-loop coverage.     | No named compatible family profiles yet.                                                                                                                                                                                                                  |
@@ -48,19 +47,19 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 
 ## AI SDK Package Parity Matrix
 
-| AI SDK package                    | Intended native target                                         | Status           | Biggest gaps                                                                                                                                                                                                        |
-| --------------------------------- | -------------------------------------------------------------- | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `@ai-sdk/openai`                  | `OpenAI.chat`, `OpenAI.responses`, `OpenAI.responsesWebSocket` | Partial / usable | Add complete typed option coverage, structured output strategy, explicit Responses continuation support, and runner route selection between Chat/Responses/WebSocket.                                               |
-| `@ai-sdk/openai-compatible`       | Generic OpenAI-compatible Chat and Responses                   | Partial / usable | Decide per-family namespace/profile behavior and runner API selection for providers that support Responses versus Chat only.                                                                                        |
-| `@ai-sdk/anthropic`               | `AnthropicMessages`                                            | Partial / usable | Finish Messages API parity for headers/betas/metadata/newer fields and document hosted-tool continuation expectations.                                                                                              |
-| `@ai-sdk/google`                  | Gemini Developer API                                           | Partial / usable | Add typed options for safety, response schema/modalities, cached content, grounding/search/code execution, and non-text output modes where supported.                                                               |
-| `@ai-sdk/google-vertex`           | Vertex Gemini namespace/facade                                 | Partial / usable | Add runner/catalog mapping, recorded coverage, and broader provider-option parity.                                                                                                                                  |
-| `@ai-sdk/google-vertex/anthropic` | Anthropic Messages over Vertex namespace/facade                | Partial / usable | Add runner/catalog mapping, recorded coverage, and Vertex-specific hosted-tool parity.                                                                                                                              |
-| `@ai-sdk/google-vertex/maas`      | Vertex Chat                                                    | Partial / usable | Add runner/catalog mapping, recorded coverage, and MaaS family-specific request parity.                                                                                                                             |
-| `@ai-sdk/google-vertex/xai`       | Vertex Chat / Responses                                        | Partial / usable | Decide Chat/Responses selection for catalog models, add runner mapping and recorded coverage, and review xAI-specific request options.                                                                              |
-| `@ai-sdk/azure`                   | Azure OpenAI Chat/Responses facade                             | Partial          | Map runner/catalog metadata to native Azure, handle resourceName/baseURL/apiVersion variants, add AAD/token auth story, and verify Chat vs Responses deployment selection.                                          |
-| `@ai-sdk/amazon-bedrock`          | Bedrock Converse                                               | Partial          | Add default AWS credential chain/profile support, region/inference-profile model ID handling, provider option parity via `additionalModelRequestFields`, guardrails/performance config, and runner/catalog mapping. |
-| `@ai-sdk/amazon-bedrock/mantle`   | Bedrock Mantle OpenAI-compatible Chat/Responses namespace      | Partial / usable | Add default AWS credential chain/profile support; native catalog mapping currently requires bearer auth or explicit static credentials.                                                                             |
+| AI SDK package                    | Intended native target                                    | Status           | Biggest gaps                                                                                                                                                                                                        |
+| --------------------------------- | --------------------------------------------------------- | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `@ai-sdk/openai`                  | `OpenAI.chat`, `OpenAI.responses`                         | Partial / usable | Add complete typed option coverage, structured output strategy, explicit Responses continuation support, and runner execution policy for optional WebSocket channels.                                               |
+| `@ai-sdk/openai-compatible`       | Generic OpenAI-compatible Chat and Responses              | Partial / usable | Decide per-family namespace/profile behavior and runner API selection for providers that support Responses versus Chat only.                                                                                        |
+| `@ai-sdk/anthropic`               | `AnthropicMessages`                                       | Partial / usable | Finish Messages API parity for headers/betas/metadata/newer fields and document hosted-tool continuation expectations.                                                                                              |
+| `@ai-sdk/google`                  | Gemini Developer API                                      | Partial / usable | Add typed options for safety, response schema/modalities, cached content, grounding/search/code execution, and non-text output modes where supported.                                                               |
+| `@ai-sdk/google-vertex`           | Vertex Gemini namespace/facade                            | Partial / usable | Add runner/catalog mapping, recorded coverage, and broader provider-option parity.                                                                                                                                  |
+| `@ai-sdk/google-vertex/anthropic` | Anthropic Messages over Vertex namespace/facade           | Partial / usable | Add runner/catalog mapping, recorded coverage, and Vertex-specific hosted-tool parity.                                                                                                                              |
+| `@ai-sdk/google-vertex/maas`      | Vertex Chat                                               | Partial / usable | Add runner/catalog mapping, recorded coverage, and MaaS family-specific request parity.                                                                                                                             |
+| `@ai-sdk/google-vertex/xai`       | Vertex Chat / Responses                                   | Partial / usable | Decide Chat/Responses selection for catalog models, add runner mapping and recorded coverage, and review xAI-specific request options.                                                                              |
+| `@ai-sdk/azure`                   | Azure OpenAI Chat/Responses facade                        | Partial          | Map runner/catalog metadata to native Azure, handle resourceName/baseURL/apiVersion variants, add AAD/token auth story, and verify Chat vs Responses deployment selection.                                          |
+| `@ai-sdk/amazon-bedrock`          | Bedrock Converse                                          | Partial          | Add default AWS credential chain/profile support, region/inference-profile model ID handling, provider option parity via `additionalModelRequestFields`, guardrails/performance config, and runner/catalog mapping. |
+| `@ai-sdk/amazon-bedrock/mantle`   | Bedrock Mantle OpenAI-compatible Chat/Responses namespace | Partial / usable | Add default AWS credential chain/profile support; native catalog mapping currently requires bearer auth or explicit static credentials.                                                                             |
 
 ## Highest-Risk Gaps
 
@@ -78,24 +77,24 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 
 These are implementation/API slices, not separate npm packages.
 
-| API slice                     | Package-like entrypoint                                     | Purpose                                                                      |
-| ----------------------------- | ----------------------------------------------------------- | ---------------------------------------------------------------------------- |
-| OpenAI Chat                   | `@opencode-ai/ai/providers/openai/chat`                     | OpenAI `/chat/completions` semantics.                                        |
-| OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`                | OpenAI `/responses` semantics with HTTP/WebSocket selected through settings. |
-| OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`               | Generic OpenAI-compatible `/chat/completions`.                               |
-| Open Responses-compatible     | `@opencode-ai/ai/providers/openai-compatible/responses`     | Generic provider-neutral `/responses`.                                       |
-| Anthropic-compatible Messages | `@opencode-ai/ai/providers/anthropic-compatible`            | Generic Anthropic-compatible `/messages`.                                    |
-| Anthropic Messages            | `@opencode-ai/ai/providers/anthropic`                       | Anthropic Messages API.                                                      |
-| Gemini Developer API          | `@opencode-ai/ai/providers/google`                          | Google AI Studio Gemini API.                                                 |
-| Vertex Gemini                 | `@opencode-ai/ai/providers/google-vertex/gemini`            | Vertex Gemini API; `providers/google-vertex` is the default alias.           |
-| Vertex Chat                   | `@opencode-ai/ai/providers/google-vertex/chat`              | Vertex OpenAI-compatible Chat Completions for MaaS models.                   |
-| Vertex Responses              | `@opencode-ai/ai/providers/google-vertex/responses`         | Vertex Open Responses for Grok models.                                       |
-| Vertex Messages               | `@opencode-ai/ai/providers/google-vertex/messages`          | Vertex-hosted Anthropic Messages API.                                        |
-| Bedrock Converse              | `@opencode-ai/ai/providers/amazon-bedrock`                  | AWS Bedrock Converse API.                                                    |
-| Bedrock Mantle Chat           | `@opencode-ai/ai/providers/amazon-bedrock/mantle/chat`      | AWS Bedrock Mantle OpenAI-compatible Chat API.                               |
-| Bedrock Mantle Responses      | `@opencode-ai/ai/providers/amazon-bedrock/mantle/responses` | AWS Bedrock Mantle OpenAI-compatible Responses API.                          |
-| Azure OpenAI Chat             | `@opencode-ai/ai/providers/azure/chat`                      | Azure specialization of OpenAI Chat.                                         |
-| Azure OpenAI Responses        | `@opencode-ai/ai/providers/azure/responses`                 | Azure specialization of OpenAI Responses.                                    |
+| API slice                     | Package-like entrypoint                                     | Purpose                                                                                    |
+| ----------------------------- | ----------------------------------------------------------- | ------------------------------------------------------------------------------------------ |
+| OpenAI Chat                   | `@opencode-ai/ai/providers/openai/chat`                     | OpenAI `/chat/completions` semantics.                                                      |
+| OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`                | OpenAI `/responses` semantics with HTTP default and optional per-call WebSocket execution. |
+| OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`               | Generic OpenAI-compatible `/chat/completions`.                                             |
+| Open Responses-compatible     | `@opencode-ai/ai/providers/openai-compatible/responses`     | Generic provider-neutral `/responses`.                                                     |
+| Anthropic-compatible Messages | `@opencode-ai/ai/providers/anthropic-compatible`            | Generic Anthropic-compatible `/messages`.                                                  |
+| Anthropic Messages            | `@opencode-ai/ai/providers/anthropic`                       | Anthropic Messages API.                                                                    |
+| Gemini Developer API          | `@opencode-ai/ai/providers/google`                          | Google AI Studio Gemini API.                                                               |
+| Vertex Gemini                 | `@opencode-ai/ai/providers/google-vertex/gemini`            | Vertex Gemini API; `providers/google-vertex` is the default alias.                         |
+| Vertex Chat                   | `@opencode-ai/ai/providers/google-vertex/chat`              | Vertex OpenAI-compatible Chat Completions for MaaS models.                                 |
+| Vertex Responses              | `@opencode-ai/ai/providers/google-vertex/responses`         | Vertex Open Responses for Grok models.                                                     |
+| Vertex Messages               | `@opencode-ai/ai/providers/google-vertex/messages`          | Vertex-hosted Anthropic Messages API.                                                      |
+| Bedrock Converse              | `@opencode-ai/ai/providers/amazon-bedrock`                  | AWS Bedrock Converse API.                                                                  |
+| Bedrock Mantle Chat           | `@opencode-ai/ai/providers/amazon-bedrock/mantle/chat`      | AWS Bedrock Mantle OpenAI-compatible Chat API.                                             |
+| Bedrock Mantle Responses      | `@opencode-ai/ai/providers/amazon-bedrock/mantle/responses` | AWS Bedrock Mantle OpenAI-compatible Responses API.                                        |
+| Azure OpenAI Chat             | `@opencode-ai/ai/providers/azure/chat`                      | Azure specialization of OpenAI Chat.                                                       |
+| Azure OpenAI Responses        | `@opencode-ai/ai/providers/azure/responses`                 | Azure specialization of OpenAI Responses.                                                  |
 
 ## Suggested Next Work Slices
 

packages/ai/example/call-sites.md

@@ -67,7 +67,6 @@ Examples:
 ```ts
 OpenAI.responses("gpt-4o")
 OpenAI.chat("gpt-4o")
-OpenAI.responsesWebSocket("gpt-4o")
 
 Azure.configure({ resourceName, apiKey }).responses("my-deployment")
 AmazonBedrock.configure({ region, credentials }).model("anthropic.claude-3-5-sonnet-20241022-v2:0")
@@ -250,11 +249,6 @@ const openAIChat = Route.make({
   auth: Auth.envBearer("OPENAI_API_KEY"),
 })
 
-const openAIResponsesWebSocket = openAIResponses.with({
-  id: "openai-responses-websocket",
-  transport: WebSocketTransport.json,
-})
-
 const openAIConfig = (input: OpenAIConfig) => ({
   endpoint: input.endpoint,
   auth: input.auth ?? (input.apiKey ? Auth.bearer(input.apiKey) : undefined),
@@ -266,13 +260,11 @@ const openAIConfig = (input: OpenAIConfig) => ({
 
 const configureOpenAI = (input: OpenAIConfig = {}) => {
   const responses = openAIResponses.with(openAIConfig(input))
-  const responsesWebSocket = openAIResponsesWebSocket.with(openAIConfig(input))
   const chat = openAIChat.with(openAIConfig(input))
 
   return {
     id: openAIProvider,
     responses: responses.model,
-    responsesWebSocket: responsesWebSocket.model,
     chat: chat.model,
     model: responses.model,
     configure: configureOpenAI,
@@ -342,22 +334,19 @@ const response =
   )
 ```
 
-For direct provider-facade calls, HTTP versus WebSocket is represented as named
-route selectors, not as model or request overrides. Same protocol, different
-transport, different route:
+For direct provider-facade calls, Responses has one semantic model and route:
 
 ```ts
 OpenAI.responses("gpt-4o")
-OpenAI.responsesWebSocket("gpt-4o")
 ```
 
-The package-like OpenAI Responses entrypoint instead keeps transport scoped to
-Responses settings while preserving the same `model(...)` contract:
+The package-like OpenAI Responses entrypoint has the same transport-neutral
+`model(...)` contract:
 
 ```ts
 import { model } from "@opencode-ai/ai/providers/openai/responses"
 
-model("gpt-4o", { apiKey, transport: "websocket" })
+model("gpt-4o", { apiKey })
 ```
 
 Vertex keeps Gemini, Chat, Responses, and Messages as separate package-like entrypoints,
@@ -387,11 +376,9 @@ import { model } from "@opencode-ai/ai/providers/google-vertex/messages"
 model("claude-sonnet-4-6", { project, location: "global" })
 ```
 
-The client should not require a different public layer just because a selected
-route uses WebSocket. Use one `LLMClient.layer` with HTTP and WebSocket runtime
-capabilities available; routes that do not need WebSocket simply never touch it.
-If a WebSocket route is selected in an environment without WebSocket support,
-fail with a typed transport configuration error.
+The client does not require a different public layer for WebSocket execution.
+Responses routes use HTTP by default, and callers may pass a channel executor per
+call. Routes without channel support simply ignore that execution capability.
 
 Azure is a route specialization with auth/path/default changes plus input
 mapping. The public API configures the Azure resource once, then selects
@@ -497,18 +484,13 @@ generic dynamic resolver:
 
 ```ts
 const model =
-  providerID === "azure"
-    ? Azure.configure(resolvedAzureConfig).responses(apiModelID)
-    : endpoint.websocket
-      ? OpenAI.responsesWebSocket(apiModelID)
-      : OpenAI.responses(apiModelID)
+  providerID === "azure" ? Azure.configure(resolvedAzureConfig).responses(apiModelID) : OpenAI.responses(apiModelID)
 ```
 
 That boundary can branch on durable config/catalog metadata and call typed
-provider APIs directly. A direct provider-facade boundary maps metadata like
-`endpoint.websocket` to `OpenAI.responsesWebSocket(apiModelID)`. A package-loading
-boundary passes `transport: "websocket"` to the OpenAI Responses entrypoint.
-The client runtime only executes the route carried by the resulting model.
+provider APIs directly. Transport selection remains execution policy: a Session
+or other caller may pass a WebSocket channel executor per call without changing
+the model constructed by this boundary.
 
 ## Competitive Shape
 
@@ -544,9 +526,8 @@ App boundary = explicit durable-config -> typed-provider call
   id.
 - No `model(id, overrides)` escape hatch. Model selection takes the model id;
   endpoint/auth/deployment customization happens by configuring the route first.
-- No transport override on an executable model or request. Direct provider
-  facades use `responses` versus `responsesWebSocket`; the package-like Responses
-  entrypoint maps its scoped `transport` setting before constructing the model.
+- No transport setting on a provider or executable model. OpenAI Responses uses
+  HTTP by default and accepts an optional per-call channel executor as execution policy.
 - No separate public `LLMClient.layerWithWebSocket`. The runtime should expose one
   client layer with the available transport capabilities.
 - No executable `ModelRef`. The executable handle is `LanguageModel`; durable model
@@ -580,12 +561,10 @@ App boundary = explicit durable-config -> typed-provider call
 - [x] Make unconfigured transports reusable constants such as
       `HttpTransport.sseJson`; keep transport functions only for configured/fresh
       state construction.
-- [x] Collapse the public WebSocket runtime split so one `LLMClient.layer`
-      exposes available transport capabilities and selected routes fail with typed
-      transport config errors when a required capability is missing.
+- [x] Collapse the public WebSocket runtime split so one `LLMClient.layer` accepts
+      optional per-call channel execution without changing route identity.
 - [x] Convert OpenAI provider APIs to provider-facade shape:
-      `OpenAI.configure(config).responses(id)`, `.chat(id)`, and
-      `.responsesWebSocket(id)`.
+      `OpenAI.configure(config).responses(id)` and `.chat(id)`.
 - [x] Convert Azure to a configured facade where resource/base URL/api version
       setup happens before selecting deployment ids.
 - [x] Split Cloudflare products into separate facades such as
@@ -599,10 +578,8 @@ App boundary = explicit durable-config -> typed-provider call
 - [ ] Decide whether a tiny `Provider.define(...)` helper is warranted after two
       or three provider conversions; start with plain objects if duplication is not
       yet painful.
-- [x] Update `packages/opencode/src/session/llm/native-request.ts` to construct
-      executable models at the session boundary with explicit provider facade
-      calls, mapping catalog metadata such as `endpoint.websocket` to the correct
-      named route selector.
+- [x] Keep executable model construction transport-neutral at the Session boundary;
+      Session-scoped execution policy supplies channel capability separately.
 - [ ] Update tests so direct route/provider tests assert route values are carried
       by executable models, and opencode/native tests assert boundary-based route
       selection.

packages/ai/example/tutorial.ts

@@ -1,6 +1,6 @@
 import { Config, Effect, Formatter, Layer, Schema, Stream } from "effect"
 import { LLM, LLMClient, LLMRequest, Message, ProviderID, Tool, ToolRuntime } from "@opencode-ai/ai"
-import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor, WebSocketExecutor } from "@opencode-ai/ai/route"
+import { Route, Auth, Endpoint, Framing, Protocol, RequestExecutor } from "@opencode-ai/ai/route"
 import { OpenAI } from "@opencode-ai/ai/providers"
 
 /**
@@ -33,9 +33,10 @@ const model = OpenAI.configure({
 //
 //   - `generation`: common controls such as max tokens, temperature, topP/topK,
 //     penalties, seed, and stop sequences.
+//   - `promptCacheKey`: stable cache affinity for protocols that support it.
 //   - `providerOptions`: namespaced provider-native behavior. For example,
-//     OpenAI cache keys and store behavior, Anthropic thinking, Gemini thinking
-//     config, or OpenRouter routing/reasoning.
+//     OpenAI store behavior, Anthropic thinking, Gemini thinking config, or
+//     OpenRouter routing/reasoning.
 //   - `http`: last-resort serializable overlays for final request body, headers,
 //     and query params. Prefer typed `providerOptions` when a field is stable.
 //
@@ -45,9 +46,7 @@ const request = LLM.request({
   system: "You are concise and practical.",
   prompt: "Tell me a joke",
   generation: { maxTokens: 80, temperature: 0.7 },
-  providerOptions: {
-    openai: { promptCacheKey: "tutorial-joke" },
-  },
+  promptCacheKey: "tutorial-joke",
 })
 
 // 3. `generate` sends the request and collects the event stream into one
@@ -214,8 +213,7 @@ const FakeEcho = {
 // enabled at a time so the tutorial can demonstrate generate, stream, or
 // tool-loop behavior without spending tokens on every example.
 const requestExecutorLayer = RequestExecutor.fetchLayer
-const llmDeps = Layer.mergeAll(requestExecutorLayer, WebSocketExecutor.layer)
-const llmClientLayer = LLMClient.layer.pipe(Layer.provide(llmDeps))
+const llmClientLayer = LLMClient.layer.pipe(Layer.provide(requestExecutorLayer))
 
 const program = Effect.gen(function* () {
   // yield* generateOnce
@@ -223,6 +221,6 @@ const program = Effect.gen(function* () {
   // yield* generateStructuredObject
   // yield* generateDynamicObject.pipe(Effect.andThen((response) => Effect.sync(() => console.log(response.object))))
   yield* streamWithTools
-}).pipe(Effect.provide(Layer.mergeAll(llmDeps, llmClientLayer)))
+}).pipe(Effect.provide(Layer.mergeAll(requestExecutorLayer, llmClientLayer)))
 
 Effect.runPromise(program)

packages/ai/script/recording-cost-report.ts

@@ -2,7 +2,7 @@ import * as fs from "node:fs/promises"
 import * as path from "node:path"
 
 const RECORDINGS_DIR = path.resolve(import.meta.dir, "..", "test", "fixtures", "recordings")
-const MODELS_DEV_URL = "https://models.dev/api.json"
+const MODELS_DEV_URL = "https://models.opencode.ai/api.json"
 
 type JsonRecord = Record<string, unknown>
 

packages/ai/src/cache-policy.ts

@@ -10,8 +10,8 @@
 //
 // Manual `cache: CacheHint` placements on individual parts are preserved and
 // count against the four-breakpoint budget; auto only fills remaining slots.
-import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options"
-import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages"
+import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options.js"
+import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages.js"
 
 const AUTO: CachePolicyObject = {
   tools: true,
@@ -133,7 +133,8 @@ const countHints = (request: LLMRequest) =>
 
 export const applyCachePolicy = (request: LLMRequest): LLMRequest => {
   if (!RESPECTS_INLINE_HINTS.has(request.model.route.id)) return request
-  if (request.model.route.id === "openrouter" && (request.cache === undefined || request.cache === "auto")) return request
+  if (request.model.route.id === "openrouter" && (request.cache === undefined || request.cache === "auto"))
+    return request
   const policy = resolve(request.cache)
   if (!policy.tools && !policy.system && !policy.messages) return request
 

packages/ai/src/image-client.ts

@@ -1,7 +1,7 @@
 import { Context, Effect, Layer } from "effect"
-import { RequestExecutor } from "./route/executor"
-import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image"
-import type { AIError } from "./schema"
+import { RequestExecutor } from "./route/executor.js"
+import type { ImageOptions, ImageRequest, ImageRequestFor, ImageResponse } from "./image.js"
+import type { AIError } from "./schema/index.js"
 
 export type Execute = RequestExecutor.Interface["execute"]
 

packages/ai/src/image.ts

@@ -1,6 +1,14 @@
 import { Effect, Schema } from "effect"
-import { HttpOptions, InvalidRequestReason, AIError, ModelID, ProviderID, ProviderMetadata, Usage } from "./schema"
-import { ImageClient, Service, type Execute as ImageExecute } from "./image-client"
+import {
+  HttpOptions,
+  InvalidRequestReason,
+  AIError,
+  ModelID,
+  ProviderID,
+  ProviderMetadata,
+  Usage,
+} from "./schema/index.js"
+import { ImageClient, Service, type Execute as ImageExecute } from "./image-client.js"
 
 export interface ImageRoute<Options extends ImageOptions = ImageOptions> {
   readonly id: string

packages/ai/src/index.ts

@@ -1,22 +1,22 @@
-export { LLMClient } from "./route/client"
-export { ImageClient } from "./image-client"
-export { Auth } from "./route/auth"
-export { Provider } from "./provider"
-export { ProviderPackage } from "./provider-package"
-export { isContextOverflow, isContextOverflowFailure } from "./provider-error"
+export { LLMClient } from "./route/client.js"
+export { ImageClient } from "./image-client.js"
+export { Auth } from "./route/auth.js"
+export { Provider } from "./provider.js"
+export { ProviderPackage } from "./provider-package.js"
+export { isContextOverflow, isContextOverflowFailure } from "./provider-error.js"
 export type {
   RouteLanguageModelInput,
   RouteRoutedLanguageModelInput,
   Interface as LLMClientShape,
   Service as LLMClientService,
-} from "./route/client"
-export * from "./schema"
-export { GeneratedImage, ImageInput, ImageInputSchema, ImageModel, ImageRequest, ImageResponse } from "./image"
-export type { ImageModelOptions, ImageOptions, ImageRequestFor, ImageRequestInput, ImageRoute } from "./image"
-export { Image } from "./image"
-export { Tool, ToolFailure, toDefinitions } from "./tool"
-export { ToolRuntime } from "./tool-runtime"
-export type { DispatchResult as ToolDispatchResult, ToolSettlement } from "./tool-runtime"
+} from "./route/client.js"
+export * from "./schema/index.js"
+export { GeneratedImage, ImageInput, ImageInputSchema, ImageModel, ImageRequest, ImageResponse } from "./image.js"
+export type { ImageModelOptions, ImageOptions, ImageRequestFor, ImageRequestInput, ImageRoute } from "./image.js"
+export { Image } from "./image.js"
+export { Tool, ToolFailure, toDefinitions } from "./tool.js"
+export { ToolRuntime } from "./tool-runtime.js"
+export type { DispatchResult as ToolDispatchResult, ToolSettlement } from "./tool-runtime.js"
 export type {
   AnyExecutableTool,
   AnyTool,
@@ -29,11 +29,14 @@ export type {
   Tools,
   ToolSchema,
   ToolToModelOutput,
-} from "./tool"
-export * as LLM from "./llm"
+} from "./tool.js"
+export * as LLM from "./llm.js"
 export type {
   Definition as ProviderDefinition,
   LanguageModelFactory as ProviderLanguageModelFactory,
   LanguageModelOptions as ProviderLanguageModelOptions,
-} from "./provider"
-export type { Definition as ProviderPackageDefinition, Settings as ProviderPackageSettings } from "./provider-package"
+} from "./provider.js"
+export type {
+  Definition as ProviderPackageDefinition,
+  Settings as ProviderPackageSettings,
+} from "./provider-package.js"

packages/ai/src/llm.ts

@@ -1,5 +1,5 @@
 import { Effect, JsonSchema, Schema } from "effect"
-import { LLMClient, Service } from "./route/client"
+import { LLMClient, Service } from "./route/client.js"
 import {
   GenerationOptions,
   HttpOptions,
@@ -15,8 +15,8 @@ import {
   ToolDefinition,
   type ContentPart,
   type LanguageModelProviderOptions,
-} from "./schema"
-import { make as makeTool, toDefinitions, type ToolSchema } from "./tool"
+} from "./schema/index.js"
+import { make as makeTool, toDefinitions, type ToolSchema } from "./tool.js"
 
 /** Input accepted by `LLM.request`, normalized into the canonical `LLMRequest` class. */
 export type RequestInput<SelectedLanguageModel extends LanguageModel = LanguageModel> = Omit<

packages/ai/src/protocols.ts

@@ -1 +1 @@
-export * from "./protocols/index"
+export * from "./protocols/index.js"

packages/ai/src/protocols/anthropic-messages.ts

@@ -1,10 +1,10 @@
 import { Effect, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
-import { Route } from "../route/client"
-import { Auth } from "../route/auth"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { Protocol } from "../route/protocol"
+import { Route } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { Protocol } from "../route/protocol.js"
 import {
   AIError,
   LLMEvent,
@@ -21,13 +21,13 @@ import {
   type ToolCallPart,
   type ToolDefinition,
   type ToolResultPart,
-} from "../schema"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
-import { classifyProviderFailure } from "../provider-error"
-import * as Cache from "./utils/cache"
-import { Lifecycle } from "./utils/lifecycle"
-import { ToolSchemaProjection } from "./utils/tool-schema"
-import { ToolStream } from "./utils/tool-stream"
+} from "../schema/index.js"
+import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
+import { classifyProviderFailure } from "../provider-error.js"
+import * as Cache from "./utils/cache.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
+import { ToolStream } from "./utils/tool-stream.js"
 
 const ADAPTER = "anthropic-messages"
 const MEDIA_MIMES = new Set<string>([...ProviderShared.IMAGE_MIMES, ...ProviderShared.PDF_MIMES])
@@ -422,14 +422,12 @@ const lowerMedia = Effect.fn("AnthropicMessages.lowerMedia")(function* (part: Me
 
 // Tool results may carry structured text, images, and documents. Keep media as provider-native
 // content instead of JSON-stringifying base64 into a prompt string.
-const lowerToolResultContentItem = Effect.fn("AnthropicMessages.lowerToolResultContentItem")(function* (
-  item: Tool.Content,
-) {
+const lowerToolResultContentItem = Effect.fnUntraced(function* (item: Tool.Content) {
   if (item.type === "text") return { type: "text" as const, text: item.text } satisfies AnthropicTextBlock
   return yield* lowerMedia({ type: "media", mediaType: item.mime, data: item.uri, filename: item.name })
 })
 
-const lowerToolResultContent = Effect.fn("AnthropicMessages.lowerToolResultContent")(function* (part: ToolResultPart) {
+const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart) {
   // Text / json / error results stay as a string for backward compatibility
   // with existing cassettes and provider expectations.
   if (part.result.type !== "content") return ProviderShared.toolResultText(part)
@@ -575,7 +573,10 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
         cache_control: cacheControl(breakpoints, part.cache),
       })
     }
-    messages.push({ role: "user", content })
+    const previous = messages.at(-1)
+    if (previous?.role === "user" && previous.content.every((block) => block.type === "tool_result"))
+      messages[messages.length - 1] = { role: "user", content: [...previous.content, ...content] }
+    else messages.push({ role: "user", content })
   }
 
   return messages
@@ -1031,4 +1032,4 @@ export const route = Route.make({
   headers: () => ({ "anthropic-version": "2023-06-01" }),
 })
 
-export * as AnthropicMessages from "./anthropic-messages"
+export * as AnthropicMessages from "./anthropic-messages.js"

packages/ai/src/protocols/bedrock-converse.ts

@@ -1,7 +1,7 @@
 import { Effect, Schema } from "effect"
-import { Route } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Protocol } from "../route/protocol"
+import { Route } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Protocol } from "../route/protocol.js"
 import {
   AIError,
   LLMEvent,
@@ -17,20 +17,20 @@ import {
   type ToolCallPart,
   type ToolDefinition,
   type ToolResultPart,
-} from "../schema"
-import { BedrockEventStream } from "./bedrock-event-stream"
-import { classifyProviderFailure } from "../provider-error"
-import { JsonObject, optionalArray, ProviderShared } from "./shared"
-import { BedrockAuth } from "./utils/bedrock-auth"
-import { BedrockCache } from "./utils/bedrock-cache"
-import { BedrockMedia } from "./utils/bedrock-media"
-import { Lifecycle } from "./utils/lifecycle"
-import { ToolSchemaProjection } from "./utils/tool-schema"
-import { ToolStream } from "./utils/tool-stream"
+} from "../schema/index.js"
+import { BedrockEventStream } from "./bedrock-event-stream.js"
+import { classifyProviderFailure } from "../provider-error.js"
+import { JsonObject, optionalArray, ProviderShared } from "./shared.js"
+import { BedrockAuth } from "./utils/bedrock-auth.js"
+import { BedrockCache } from "./utils/bedrock-cache.js"
+import { BedrockMedia } from "./utils/bedrock-media.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
+import { ToolStream } from "./utils/tool-stream.js"
 
 const ADAPTER = "bedrock-converse"
 
-export type { Credentials as BedrockCredentials } from "./utils/bedrock-auth"
+export type { Credentials as BedrockCredentials } from "./utils/bedrock-auth.js"
 
 // =============================================================================
 // Request Body Schema
@@ -348,7 +348,10 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
           continue
         }
       }
-      messages.push({ role: "user", content })
+      const previous = messages.at(-1)
+      if (previous?.role === "user")
+        messages[messages.length - 1] = { role: "user", content: [...previous.content, ...content] }
+      else messages.push({ role: "user", content })
       continue
     }
 
@@ -392,7 +395,10 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
       const cachePoint = BedrockCache.block(breakpoints, part.cache)
       if (cachePoint) content.push(cachePoint)
     }
-    messages.push({ role: "user", content })
+    const previous = messages.at(-1)
+    if (previous?.role === "user")
+      messages[messages.length - 1] = { role: "user", content: [...previous.content, ...content] }
+    else messages.push({ role: "user", content })
   }
 
   return messages
@@ -731,4 +737,4 @@ export const route = Route.make({
 
 export const sigV4Auth = BedrockAuth.sigV4
 
-export * as BedrockConverse from "./bedrock-converse"
+export * as BedrockConverse from "./bedrock-converse.js"

packages/ai/src/protocols/bedrock-event-stream.ts

@@ -1,8 +1,8 @@
 import { EventStreamCodec } from "@smithy/eventstream-codec"
 import { fromUtf8, toUtf8 } from "@smithy/util-utf8"
 import { Effect, Stream } from "effect"
-import { Framing } from "../route/framing"
-import { ProviderShared } from "./shared"
+import { Framing } from "../route/framing.js"
+import { ProviderShared } from "./shared.js"
 
 // Bedrock streams responses using the AWS event stream binary protocol — each
 // frame is `[length:4][headers-length:4][prelude-crc:4][headers][payload][crc:4]`.
@@ -98,4 +98,4 @@ export const framing = (route: string): Framing.Definition<object> => ({
   frame: (bytes) => bytes.pipe(Stream.mapAccumEffect(() => initialFrameBuffer, consumeFrames(route))),
 })
 
-export * as BedrockEventStream from "./bedrock-event-stream"
+export * as BedrockEventStream from "./bedrock-event-stream.js"

packages/ai/src/protocols/gemini.ts

@@ -1,10 +1,10 @@
 import { Effect, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
-import { Route } from "../route/client"
-import { Auth } from "../route/auth"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { Protocol } from "../route/protocol"
+import { Route } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { Protocol } from "../route/protocol.js"
 import {
   LLMEvent,
   Usage,
@@ -17,16 +17,28 @@ import {
   type TextPart,
   type ToolCallPart,
   type ToolDefinition,
-} from "../schema"
-import { JsonObject, optionalArray, ProviderShared } from "./shared"
-import { GeminiToolSchema } from "./utils/gemini-tool-schema"
-import { Lifecycle } from "./utils/lifecycle"
-import { ToolSchemaProjection } from "./utils/tool-schema"
+} from "../schema/index.js"
+import { JsonObject, optionalArray, ProviderShared } from "./shared.js"
+import { GeminiToolSchema } from "./utils/gemini-tool-schema.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
 
 const ADAPTER = "gemini"
 const MEDIA_MIMES = new Set<string>(ProviderShared.MEDIA_MIMES)
+// Google documents this sentinel for replaying Gemini 3 function calls after their original signature was lost.
+const SKIP_THOUGHT_SIGNATURE_VALIDATOR = "skip_thought_signature_validator"
 export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"
 
+// Gemini 3 rejects replayed function calls without a thought signature. Google's SDKs avoid that in normal chats by
+// retaining complete model responses, but OpenCode reconstructs durable history and may encounter an unsigned call
+// from an older or external session. Model IDs are open-ended, so unknown Gemini aliases inherit current behavior.
+const requiresThoughtSignatureFallback = (modelID: string) => {
+  if (!/(^|\/)gemini-/i.test(modelID)) return false
+  if (/(^|\/)gemini-(?:1|2)(?:[.-]|$)/i.test(modelID)) return false
+  if (/(^|\/)gemini-pro(?:-vision)?$/i.test(modelID)) return false
+  return !/(^|\/)gemini-robotics-er-1\.5(?:[.-]|$)/i.test(modelID)
+}
+
 export interface OptionsInput {
   readonly [key: string]: unknown
   readonly cachedContent?: string
@@ -145,6 +157,9 @@ const GeminiGenerationConfig = Schema.Struct({
   temperature: Schema.optional(Schema.Number),
   topP: Schema.optional(Schema.Number),
   topK: Schema.optional(Schema.Number),
+  frequencyPenalty: Schema.optional(Schema.Number),
+  presencePenalty: Schema.optional(Schema.Number),
+  seed: Schema.optional(Schema.Number),
   stopSequences: optionalArray(Schema.String),
   thinkingConfig: Schema.optional(GeminiThinkingConfig),
 })
@@ -202,11 +217,13 @@ interface ParserState {
 //    keys on non-object scalars. Mirrors OpenCode's historical Gemini rules.
 //
 // 2. Project — lossy mapping from JSON Schema to Gemini's schema dialect:
-//    drop empty objects, derive `nullable: true` from `type: [..., "null"]`,
-//    coerce `const` to `[const]` enum, recurse properties/items, propagate
+//    drop empty root parameter schemas while preserving nested empty objects,
+//    expand type arrays into `anyOf`, derive `nullable: true` from null members,
+//    coerce `const` to `[const]` enum, recurse properties/items, and propagate
 //    only an allowlisted set of keys (description, required, format, type,
-//    properties, items, allOf, anyOf, oneOf, minLength). Anything outside the
-//    allowlist (e.g. `additionalProperties`, `$ref`) is silently dropped.
+//    nullable, enum, properties, items, allOf, anyOf, oneOf, minLength).
+//    Anything outside the allowlist (e.g. `additionalProperties`, `$ref`) is
+//    silently dropped.
 //
 // Sanitize runs first, then project. The implementation lives in
 // `utils/gemini-tool-schema` so this protocol keeps the same shape as the other
@@ -282,6 +299,8 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
 
     if (message.role === "assistant") {
       const parts: Array<Schema.Schema.Type<typeof GeminiContentPart>> = []
+      // Parallel Gemini 3 calls may carry one signature on the first call; unsigned sibling calls are valid.
+      let hasSignedToolCall = false
       for (const part of message.content) {
         if (!ProviderShared.supportsContent(part, ["text", "reasoning", "tool-call"]))
           return yield* ProviderShared.unsupportedContent("Gemini", "assistant", ["text", "reasoning", "tool-call"])
@@ -294,7 +313,17 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
           continue
         }
         if (part.type === "tool-call") {
-          parts.push(lowerToolCall(part))
+          const lowered = lowerToolCall(part)
+          const signature = lowered.thoughtSignature
+          parts.push({
+            ...lowered,
+            thoughtSignature:
+              signature ??
+              (requiresThoughtSignatureFallback(request.model.id) && !hasSignedToolCall
+                ? SKIP_THOUGHT_SIGNATURE_VALIDATOR
+                : undefined),
+          })
+          if (signature !== undefined) hasSignedToolCall = true
           continue
         }
       }
@@ -388,6 +417,9 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
     temperature: generation?.temperature,
     topP: generation?.topP,
     topK: generation?.topK,
+    frequencyPenalty: generation?.frequencyPenalty,
+    presencePenalty: generation?.presencePenalty,
+    seed: generation?.seed,
     stopSequences: generation?.stop,
     thinkingConfig: options.thinkingConfig,
   }
@@ -611,4 +643,4 @@ export const route = Route.make({
   framing: Framing.sse,
 })
 
-export * as Gemini from "./gemini"
+export * as Gemini from "./gemini.js"

packages/ai/src/protocols/google-images.ts

@@ -7,8 +7,8 @@ import {
   type ImageInput,
   type ImageRequestFor,
   type ImageRoute,
-} from "../image"
-import { Auth, type Definition as AuthDefinition } from "../route/auth"
+} from "../image.js"
+import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
 import {
   InvalidProviderOutputReason,
   AIError,
@@ -17,9 +17,9 @@ import {
   mergeJsonRecords,
   type HttpOptions,
   type ProviderMetadata,
-} from "../schema"
-import { ProviderShared } from "./shared"
-import { ImageInputs } from "./utils/image-input"
+} from "../schema/index.js"
+import { ProviderShared } from "./shared.js"
+import { ImageInputs } from "./utils/image-input.js"
 
 const ADAPTER = "google-images"
 export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"

packages/ai/src/protocols/index.ts

@@ -1,9 +1,10 @@
-export * as AnthropicMessages from "./anthropic-messages"
-export * as BedrockConverse from "./bedrock-converse"
-export * as Gemini from "./gemini"
-export * as OpenAIChat from "./openai-chat"
-export * as OpenAIImages from "./openai-images"
-export * as OpenAICompatibleChat from "./openai-compatible-chat"
-export * as OpenAICompatibleResponses from "./openai-compatible-responses"
-export * as OpenAIResponses from "./openai-responses"
-export * as OpenResponses from "./open-responses"
+export * as AnthropicMessages from "./anthropic-messages.js"
+export * as BedrockConverse from "./bedrock-converse.js"
+export * as Gemini from "./gemini.js"
+export * as OpenAIChat from "./openai-chat.js"
+export * as OpenAIImages from "./openai-images.js"
+export * as OpenAICompatibleChat from "./openai-compatible-chat.js"
+export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
+export * as OpenAIResponses from "./openai-responses.js"
+export * as OpenResponses from "./open-responses.js"
+export * as OpenResponsesChannel from "./open-responses-channel.js"

packages/ai/src/protocols/open-responses-channel.ts

@@ -0,0 +1,191 @@
+import { Effect, Schema, Stream } from "effect"
+import { Headers } from "effect/unstable/http"
+import { Framing } from "../route/framing.js"
+import {
+  HttpTransport,
+  WebSocketTransport,
+  type Transport,
+  type WebSocketChannelDriver,
+  type WebSocketChannelExchange,
+} from "../route/transport/index.js"
+import * as ProviderShared from "./shared.js"
+import { OpenResponses } from "./open-responses.js"
+
+const WebSocketResponseCreate = Schema.StructWithRest(Schema.Struct({ type: Schema.tag("response.create") }), [
+  Schema.Record(Schema.String, Schema.Unknown),
+])
+const decodeMessage = ProviderShared.validateWith(Schema.decodeUnknownEffect(WebSocketResponseCreate))
+const encodeMessage = Schema.encodeSync(Schema.fromJsonString(WebSocketResponseCreate))
+const decodeEvent = Schema.decodeUnknownEffect(OpenResponses.protocol.stream.event)
+
+export interface Options {
+  readonly id: string
+  readonly name: string
+  readonly rotateAfterMs?: number
+  readonly headers?: (headers: Headers.Headers) => Headers.Headers
+  readonly driver?: (input: {
+    readonly request: Readonly<Record<string, unknown>>
+    readonly message: string
+    readonly base: WebSocketChannelDriver
+  }) => WebSocketChannelDriver
+}
+
+export interface Prepared {
+  readonly http: HttpTransport.HttpPrepared<string>
+  readonly channel?: {
+    readonly url: string
+    readonly headers: Headers.Headers
+    readonly rotateAfterMs?: number
+    readonly driver: WebSocketChannelDriver
+  }
+}
+
+const message = (body: unknown) =>
+  Effect.gen(function* () {
+    if (!ProviderShared.isRecord(body))
+      return yield* ProviderShared.invalidRequest("Open Responses WebSocket body must be a JSON object")
+    const { stream: _stream, stream_options: _streamOptions, background: _background, ...request } = body
+    const decoded = yield* decodeMessage({ ...request, type: "response.create" })
+    return { request: decoded, message: encodeMessage(decoded) }
+  })
+
+const driver = (options: Options, body: string): WebSocketChannelDriver => {
+  let responseID: string | undefined
+  let terminal = false
+  return {
+    create: () =>
+      Effect.sync(() => {
+        responseID = undefined
+        terminal = false
+        return { message: body, mode: "full" }
+      }),
+    observe: (_create, frame) =>
+      Effect.gen(function* () {
+        const event = yield* decodeEvent(frame).pipe(
+          Effect.mapError(() =>
+            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame),
+          ),
+        )
+        if (terminal)
+          return yield* ProviderShared.eventError(
+            options.id,
+            `${options.name} emitted ${event.type} after a terminal event`,
+            frame,
+          )
+        if (event.type === "error") {
+          terminal = true
+          yield* OpenResponses.decodeKnownErrorEvent(event).pipe(
+            Effect.mapError(() =>
+              ProviderShared.eventError(options.id, `${options.name} returned a malformed error event`, frame),
+            ),
+          )
+          return {
+            type: "provider-failure",
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} stream error`),
+          }
+        }
+        if (event.type === "response.failed") {
+          terminal = true
+          if (responseID && event.response?.id && event.response.id !== responseID)
+            return yield* ProviderShared.eventError(
+              options.id,
+              `${options.name} response ID changed during execution`,
+              frame,
+            )
+          return {
+            type: "provider-failure",
+            error: OpenResponses.providerFailure(options.id, event, `${options.name} response failed`),
+          }
+        }
+        if (event.type === "response.created") {
+          const created = event.response?.id
+          if (responseID)
+            return yield* ProviderShared.eventError(
+              options.id,
+              `${options.name} emitted duplicate response.created`,
+              frame,
+            )
+          if (!created)
+            return yield* ProviderShared.eventError(
+              options.id,
+              `${options.name} response.created is missing response.id`,
+              frame,
+            )
+          responseID = created
+          return { type: "frame", frame }
+        }
+        if (!responseID)
+          return yield* ProviderShared.eventError(
+            options.id,
+            `${options.name} emitted ${event.type} before response.created`,
+            frame,
+          )
+        if (event.response?.id && event.response.id !== responseID)
+          return yield* ProviderShared.eventError(
+            options.id,
+            `${options.name} response ID changed during execution`,
+            frame,
+          )
+        if (event.type === "response.completed") {
+          terminal = true
+          return { type: "completed", frame }
+        }
+        if (event.type === "response.incomplete") {
+          terminal = true
+          return { type: "incomplete", frame }
+        }
+        return { type: "frame", frame }
+      }),
+  }
+}
+
+export const transport = <Body>(options: Options): Transport<Body, Prepared, string> => {
+  const http = HttpTransport.sseJson.with<Body>()
+  return {
+    id: http.id,
+    prepare: (input) =>
+      Effect.gen(function* () {
+        const parts = yield* HttpTransport.jsonRequestParts(input)
+        const headers = Headers.remove(options.headers?.(parts.headers) ?? parts.headers, "content-length")
+        const channel = input.webSocket
+          ? yield* Effect.gen(function* () {
+              const create = yield* message(parts.jsonBody)
+              const base = driver(options, create.message)
+              return {
+                url: yield* WebSocketTransport.toWebSocketUrl(parts.url),
+                headers,
+                rotateAfterMs: options.rotateAfterMs,
+                driver: options.driver?.({ request: create.request, message: create.message, base }) ?? base,
+              }
+            })
+          : undefined
+        return {
+          http: {
+            request: ProviderShared.jsonPost({ url: parts.url, body: parts.bodyText, headers: parts.headers }),
+            framing: Framing.sse,
+            middleware: input.middleware,
+          },
+          channel,
+        }
+      }),
+    execute: (prepared, request, runtime, executeOptions) => {
+      if (!executeOptions?.webSocket || !prepared.channel) return http.execute(prepared.http, request, runtime)
+      const exchange: WebSocketChannelExchange = {
+        id: request.id ?? "request",
+        connect: {
+          url: prepared.channel.url,
+          headers: prepared.channel.headers,
+          rotateAfterMs: prepared.channel.rotateAfterMs,
+        },
+        fallback: () =>
+          Stream.unwrap(
+            http.execute(prepared.http, request, runtime).pipe(Effect.map((execution) => execution.frames)),
+          ),
+        driver: prepared.channel.driver,
+      }
+      return executeOptions.webSocket.execute(exchange)
+    },
+  }
+}
+
+export const OpenResponsesChannel = { transport } as const

packages/ai/src/protocols/open-responses.ts

@@ -1,7 +1,7 @@
 import { Effect, Schema } from "effect"
 import type { Content } from "@opencode-ai/schema/tool"
-import { HttpTransport } from "../route/transport"
-import { Protocol } from "../route/protocol"
+import { HttpTransport } from "../route/transport/index.js"
+import { Protocol } from "../route/protocol.js"
 import {
   AIError,
   LLMEvent,
@@ -16,13 +16,13 @@ import {
   type ToolCallPart,
   type ToolDefinition,
   type ToolResultPart,
-} from "../schema"
-import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
-import { classifyProviderFailure } from "../provider-error"
-import { OpenResponsesOptions } from "./utils/open-responses-options"
-import { Lifecycle } from "./utils/lifecycle"
-import { ToolSchemaProjection } from "./utils/tool-schema"
-import { ToolStream } from "./utils/tool-stream"
+} from "../schema/index.js"
+import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
+import { classifyProviderFailure } from "../provider-error.js"
+import { OpenResponsesOptions } from "./utils/open-responses-options.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
+import { ToolStream } from "./utils/tool-stream.js"
 
 const ADAPTER = "open-responses"
 const NAME = "Open Responses"
@@ -211,11 +211,43 @@ export type StreamItem = Schema.Schema.Type<typeof StreamItem>
 // event-level `error` envelope, so accept all three shapes here.
 // https://www.openresponses.org/specification
 const OpenResponsesErrorPayload = Schema.Struct({
+  type: optionalNull(Schema.String),
   code: optionalNull(Schema.String),
   message: optionalNull(Schema.String),
   param: optionalNull(Schema.String),
 })
 
+const WebSocketErrorHeader = Schema.Union([Schema.String, Schema.Number, Schema.Boolean])
+export const WebSocketErrorEvent = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.tag("error"),
+    status: Schema.optional(Schema.Number),
+    status_code: Schema.optional(Schema.Number),
+    code: optionalNull(Schema.String),
+    message: Schema.optional(Schema.String),
+    param: optionalNull(Schema.String),
+    error: optionalNull(OpenResponsesErrorPayload),
+    headers: Schema.optional(Schema.Record(Schema.String, WebSocketErrorHeader)),
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
+const decodeWebSocketErrorEvent = Schema.decodeUnknownEffect(WebSocketErrorEvent)
+
+export const decodeKnownErrorEvent = (event: Event) =>
+  decodeWebSocketErrorEvent({
+    ...event,
+    status: typeof event.status === "number" ? event.status : undefined,
+    status_code: typeof event.status_code === "number" ? event.status_code : undefined,
+    headers: ProviderShared.isRecord(event.headers)
+      ? Object.fromEntries(
+          Object.entries(event.headers).filter(
+            (entry): entry is [string, string | number | boolean] =>
+              typeof entry[1] === "string" || typeof entry[1] === "number" || typeof entry[1] === "boolean",
+          ),
+        )
+      : undefined,
+  })
+
 export const Event = Schema.StructWithRest(
   Schema.Struct({
     type: Schema.String,
@@ -240,6 +272,9 @@ export const Event = Schema.StructWithRest(
     message: Schema.optional(Schema.String),
     param: optionalNull(Schema.String),
     error: optionalNull(OpenResponsesErrorPayload),
+    status: Schema.optional(Schema.Unknown),
+    status_code: Schema.optional(Schema.Unknown),
+    headers: Schema.optional(Schema.Unknown),
   }),
   [Schema.Record(Schema.String, Schema.Unknown)],
 )
@@ -358,7 +393,7 @@ const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
   return { type: "input_image" as const, image_url: media.dataUrl }
 })
 
-const lowerUserContent = Effect.fn("OpenResponses.lowerUserContent")(function* (
+const lowerUserContent = Effect.fnUntraced(function* (
   part: LLMRequest["messages"][number]["content"][number],
   request: LLMRequest,
   extension: Extension,
@@ -370,7 +405,7 @@ const lowerUserContent = Effect.fn("OpenResponses.lowerUserContent")(function* (
 
 // Tool results may carry structured text, images, and files. Keep media as provider-native
 // content instead of JSON-stringifying base64 into a prompt string.
-const lowerToolResultContentItem = Effect.fn("OpenResponses.lowerToolResultContentItem")(function* (
+const lowerToolResultContentItem = Effect.fnUntraced(function* (
   item: Content,
   request: LLMRequest,
   extension: Extension,
@@ -383,7 +418,7 @@ const lowerToolResultContentItem = Effect.fn("OpenResponses.lowerToolResultConte
   )
 })
 
-const lowerToolResultOutput = Effect.fn("OpenResponses.lowerToolResultOutput")(function* (
+const lowerToolResultOutput = Effect.fnUntraced(function* (
   part: ToolResultPart,
   request: LLMRequest,
   extension: Extension,
@@ -539,7 +574,7 @@ const lowerOptions = (request: LLMRequest) => {
   return {
     ...(options.instructions ? { instructions: options.instructions } : {}),
     ...(options.store !== undefined ? { store: options.store } : {}),
-    ...(options.promptCacheKey ? { prompt_cache_key: options.promptCacheKey } : {}),
+    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
     ...(options.include ? { include: options.include } : {}),
     ...(options.reasoningEffort || options.reasoningSummary
       ? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }
@@ -632,9 +667,9 @@ export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []
 
 // `response.completed` / `response.incomplete` are clean finishes that emit a
-// `finish` event; `response.failed` is a hard failure. All three end the stream,
-// so keep this set aligned with `step` and the protocol's terminal predicate.
-const TERMINAL_TYPES = new Set(["response.completed", "response.incomplete", "response.failed"])
+// `finish` event; `response.failed` and `error` are hard failures. All four end
+// the stream, so keep this set aligned with `step` and the protocol's terminal predicate.
+const TERMINAL_TYPES = new Set(["error", "response.completed", "response.incomplete", "response.failed"])
 export const terminal = (event: Event) => TERMINAL_TYPES.has(event.type)
 
 const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepResult => {
@@ -966,16 +1001,24 @@ const providerErrorMessage = (event: Event, fallback: string): string => {
   return message || code || fallback
 }
 
-const providerError = (state: ParserState, event: Event, fallback: string) => {
+export const providerFailure = (id: string, event: Event, fallback: string) => {
   const code = event.code || event.error?.code || event.response?.error?.code || undefined
   const message = providerErrorMessage(event, fallback)
+  const status =
+    typeof event.status === "number"
+      ? event.status
+      : typeof event.status_code === "number"
+        ? event.status_code
+        : undefined
   return new AIError({
-    module: state.id,
+    module: id,
     method: "stream",
-    reason: classifyProviderFailure({ message, code }),
+    reason: classifyProviderFailure({ message, code, status }),
   })
 }
 
+const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)
+
 export const step = (state: ParserState, event: Event) => {
   if (event.type === "response.output_text.delta" || event.type === "response.output_text.done") {
     if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
@@ -1015,7 +1058,11 @@ export const step = (state: ParserState, event: Event) => {
   if (event.type === "response.completed" || event.type === "response.incomplete")
     return Effect.succeed(onResponseFinish(state, event))
   if (event.type === "response.failed") return providerError(state, event, `${state.name} response failed`)
-  if (event.type === "error") return providerError(state, event, `${state.name} stream error`)
+  if (event.type === "error")
+    return decodeKnownErrorEvent(event).pipe(
+      Effect.mapError(() => ProviderShared.eventError(state.id, `${state.name} returned a malformed error event`)),
+      Effect.flatMap(() => providerError(state, event, `${state.name} stream error`)),
+    )
   return Effect.succeed<StepResult>([state, NO_EVENTS])
 }
 
@@ -1061,4 +1108,4 @@ export const protocol = Protocol.make({
 
 export const httpTransport = HttpTransport.sseJson.with<OpenResponsesBody>()
 
-export * as OpenResponses from "./open-responses"
+export * as OpenResponses from "./open-responses.js"

packages/ai/src/protocols/openai-chat.ts

@@ -1,10 +1,10 @@
 import { Effect, Schema } from "effect"
 import { Tool } from "@opencode-ai/schema/tool"
-import { Route } from "../route/client"
-import { Auth } from "../route/auth"
-import { Endpoint } from "../route/endpoint"
-import { HttpTransport } from "../route/transport"
-import { Protocol } from "../route/protocol"
+import { Route } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import { Endpoint } from "../route/endpoint.js"
+import { HttpTransport } from "../route/transport/index.js"
+import { Protocol } from "../route/protocol.js"
 import {
   AIError,
   LLMEvent,
@@ -19,13 +19,13 @@ import {
   type TextPart,
   type ToolCallPart,
   type ToolDefinition,
-} from "../schema"
-import { classifyProviderFailure } from "../provider-error"
-import { isRecord, JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
-import { OpenAIOptions } from "./utils/openai-options"
-import { Lifecycle } from "./utils/lifecycle"
-import { ToolSchemaProjection } from "./utils/tool-schema"
-import { ToolStream } from "./utils/tool-stream"
+} from "../schema/index.js"
+import { classifyProviderFailure } from "../provider-error.js"
+import { isRecord, JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
+import { OpenAIOptions } from "./utils/openai-options.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
+import { ToolStream } from "./utils/tool-stream.js"
 
 const ADAPTER = "openai-chat"
 const IMAGE_MIMES = new Set<string>(ProviderShared.IMAGE_MIMES)
@@ -67,6 +67,12 @@ const OpenAIChatAssistantToolCall = Schema.Struct({
 })
 type OpenAIChatAssistantToolCall = Schema.Schema.Type<typeof OpenAIChatAssistantToolCall>
 
+// Intentionally omit Gemini's provider-specific `extra_content.google.thought_signature`
+// extension until direct Google OpenAI-compatible routing is supported here:
+// https://github.com/vercel/ai/issues/11590
+// https://github.com/vercel/ai/pull/11745
+// https://ai.google.dev/gemini-api/docs/thought-signatures#openai
+
 const OpenAIChatUserContent = Schema.Union([
   Schema.Struct({
     type: Schema.Literal("text"),
@@ -126,6 +132,7 @@ export const bodyFields = {
   stream: Schema.Literal(true),
   stream_options: Schema.optional(Schema.Struct({ include_usage: Schema.Boolean })),
   store: Schema.optional(Schema.Boolean),
+  prompt_cache_key: Schema.optional(Schema.String),
   reasoning_effort: Schema.optional(OpenAIOptions.OpenAIReasoningEffort),
   max_completion_tokens: Schema.optional(Schema.Number),
   max_tokens: Schema.optional(Schema.Number),
@@ -145,22 +152,33 @@ export type OpenAIChatBody = Schema.Schema.Type<typeof OpenAIChatBody>
 // The event schema is one decoded SSE `data:` payload. `Framing.sse` splits the
 // byte stream into strings, then `Protocol.jsonEvent` decodes each string into
 // this provider-native event shape.
-const OpenAIChatUsage = Schema.Struct({
-  prompt_tokens: Schema.optional(Schema.Number),
-  completion_tokens: Schema.optional(Schema.Number),
-  total_tokens: Schema.optional(Schema.Number),
-  prompt_tokens_details: optionalNull(
-    Schema.Struct({
-      cached_tokens: Schema.optional(Schema.Number),
-      cache_write_tokens: Schema.optional(Schema.Number),
-    }),
-  ),
-  completion_tokens_details: optionalNull(
-    Schema.Struct({
-      reasoning_tokens: Schema.optional(Schema.Number),
-    }),
-  ),
-})
+const OpenAIChatUsage = Schema.StructWithRest(
+  Schema.Struct({
+    prompt_tokens: optionalNull(Schema.Number),
+    completion_tokens: optionalNull(Schema.Number),
+    total_tokens: optionalNull(Schema.Number),
+    prompt_tokens_details: optionalNull(
+      Schema.StructWithRest(
+        Schema.Struct({
+          cached_tokens: optionalNull(Schema.Number),
+          cache_write_tokens: optionalNull(Schema.Number),
+        }),
+        [Schema.Record(Schema.String, Schema.Unknown)],
+      ),
+    ),
+    completion_tokens_details: optionalNull(
+      Schema.StructWithRest(
+        Schema.Struct({
+          reasoning_tokens: optionalNull(Schema.Number),
+          accepted_prediction_tokens: optionalNull(Schema.Number),
+          rejected_prediction_tokens: optionalNull(Schema.Number),
+        }),
+        [Schema.Record(Schema.String, Schema.Unknown)],
+      ),
+    ),
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
 
 const OpenAIChatToolCallDeltaFunction = Schema.Struct({
   name: optionalNull(Schema.String),
@@ -168,7 +186,7 @@ const OpenAIChatToolCallDeltaFunction = Schema.Struct({
 })
 
 const OpenAIChatToolCallDelta = Schema.Struct({
-  index: Schema.Number,
+  index: optionalNull(Schema.Number),
   id: optionalNull(Schema.String),
   function: optionalNull(OpenAIChatToolCallDeltaFunction),
 })
@@ -222,6 +240,8 @@ export interface ParserState {
   readonly reasoningDetails: Array<unknown>
   readonly reasoningDetailsObserved: boolean
   readonly reasoningEmitted: boolean
+  readonly latestToolIndex?: number
+  readonly nextToolIndex: number
 }
 
 // =============================================================================
@@ -351,12 +371,13 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
     return text
   })()
   const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
+  const cacheControl = options.cacheControl?.(cached && "cache" in cached ? cached.cache : undefined)
   const result = {
     role: "assistant" as const,
-    content: content.length === 0 ? null : ProviderShared.joinText(content),
-    tool_calls: toolCalls.length === 0 ? undefined : toolCalls,
-    reasoning_details: details,
-    cache_control: options.cacheControl?.(cached && "cache" in cached ? cached.cache : undefined),
+    content: content.length > 0 ? content.map((part) => part.text).join("") : toolCalls.length > 0 ? null : "",
+    ...(toolCalls.length > 0 ? { tool_calls: toolCalls } : {}),
+    ...(details !== undefined ? { reasoning_details: details } : {}),
+    ...(cacheControl !== undefined ? { cache_control: cacheControl } : {}),
   }
   if (field === undefined || reasoningText === undefined) return result
   return { ...result, [field]: reasoningText }
@@ -490,6 +511,7 @@ const lowerOptions = (request: LLMRequest) => {
   const options = OpenAIOptions.resolve(request)
   return {
     ...(options.store !== undefined ? { store: options.store } : {}),
+    ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
     ...(options.reasoningEffort ? { reasoning_effort: options.reasoningEffort } : {}),
   }
 }
@@ -559,22 +581,34 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // satisfied on both sides.
 const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
   if (!usage) return undefined
-  const cached = usage.prompt_tokens_details?.cached_tokens
-  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens
-  const reasoning = usage.completion_tokens_details?.reasoning_tokens
-  const nonCached = ProviderShared.subtractTokens(usage.prompt_tokens, ProviderShared.sumTokens(cached, cacheWrite))
+  const input = usage.prompt_tokens ?? undefined
+  const output = usage.completion_tokens ?? undefined
+  const cached = usage.prompt_tokens_details?.cached_tokens ?? undefined
+  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
+  const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
+  const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
   return new Usage({
-    inputTokens: usage.prompt_tokens,
-    outputTokens: usage.completion_tokens,
+    inputTokens: input,
+    outputTokens: output,
     nonCachedInputTokens: nonCached,
     cacheReadInputTokens: cached,
     cacheWriteInputTokens: cacheWrite,
     reasoningTokens: reasoning,
-    totalTokens: ProviderShared.totalTokens(usage.prompt_tokens, usage.completion_tokens, usage.total_tokens),
+    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
     providerMetadata: { openai: usage },
   })
 }
 
+const toolIndexByID = (
+  tools: ParserState["tools"],
+  pendingTools: ParserState["pendingTools"],
+  id: string | undefined,
+) => {
+  if (!id) return undefined
+  const entry = Object.entries({ ...pendingTools, ...tools }).find(([, tool]) => tool?.id === id)
+  return entry ? Number(entry[0]) : undefined
+}
+
 const reasoningDelta = (
   delta: Schema.Schema.Type<typeof OpenAIChatDelta> | null | undefined,
   configuredField?: string,
@@ -657,25 +691,39 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     const events: LLMEvent[] = []
     const usage = mapUsage(event.usage) ?? state.usage
     const choice = event.choices?.[0]
-    const finishReason = choice?.finish_reason
-      ? { normalized: mapFinishReason(choice.finish_reason), raw: choice.native_finish_reason ?? choice.finish_reason }
-      : state.finishReason
+    const rawFinishReason = choice?.finish_reason
+    const finishReason =
+      rawFinishReason !== undefined && rawFinishReason !== null
+        ? { normalized: mapFinishReason(rawFinishReason), raw: choice?.native_finish_reason ?? rawFinishReason }
+        : state.finishReason
     const delta = choice?.delta
     const toolDeltas = delta?.tool_calls ?? []
     let tools = state.tools
     let pendingTools = state.pendingTools
+    let latestToolIndex = state.latestToolIndex
+    let nextToolIndex = state.nextToolIndex
 
     let lifecycle = state.lifecycle
 
     const reasoning = reasoningDelta(delta, state.reasoningField)
-    const reasoningField = state.reasoningField ?? (!state.lifecycle.text.has("text-0") ? reasoning?.field : undefined)
+    const hasLateContent =
+      Boolean(delta?.content) ||
+      reasoning !== undefined ||
+      (Array.isArray(delta?.reasoning_details) && delta.reasoning_details.length > 0) ||
+      toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
+    if (state.finishReason !== undefined) {
+      if (hasLateContent)
+        return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat received content after the finish reason")
+      return [{ ...state, usage }, events] as const
+    }
+
+    const reasoningField = state.reasoningField ?? reasoning?.field
     const detailDelta = Array.isArray(delta?.reasoning_details) ? delta.reasoning_details : undefined
     if (detailDelta !== undefined) appendReasoningDetails(state.reasoningDetails, detailDelta)
     const reasoningDetailsObserved = state.reasoningDetailsObserved || detailDelta !== undefined
     const deltaMetadata = reasoningMetadata(reasoningField)
     const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
-    if (!state.lifecycle.text.has("text-0") && text !== undefined)
-      lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
+    if (text !== undefined) lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
     else if (
       reasoningDetailsObserved &&
       !lifecycle.reasoning.has("reasoning-0") &&
@@ -694,24 +742,36 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
       lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
     }
 
-    for (const tool of toolDeltas) {
-      const current = tools[tool.index]
-      const pending = pendingTools[tool.index]
+    // Compatible providers may omit indexes. Prefer durable identity, then use
+    // batch position for parallel deltas or the latest call for sparse chunks.
+    for (const [position, tool] of toolDeltas.entries()) {
+      const matched = toolIndexByID(tools, pendingTools, tool.id || undefined)
+      const fallback = toolDeltas.length > 1 ? position : (latestToolIndex ?? position)
+      const fallbackTool = tools[fallback] ?? pendingTools[fallback]
+      const index =
+        tool.index ?? matched ?? (tool.id && fallbackTool?.id && fallbackTool.id !== tool.id ? nextToolIndex : fallback)
+      const current = tools[index]
+      const pending = pendingTools[index]
       const id = current?.id ?? pending?.id ?? (tool.id || undefined)
       const name = current?.name ?? pending?.name ?? (tool.function?.name || undefined)
       const text = `${pending?.input ?? ""}${tool.function?.arguments ?? ""}`
+      latestToolIndex = index
+      nextToolIndex = Math.max(nextToolIndex, index + 1)
       if (!current && (!id || !name)) {
-        pendingTools = { ...pendingTools, [tool.index]: { id: id || undefined, name: name || undefined, input: text } }
+        pendingTools = {
+          ...pendingTools,
+          [index]: { id: id || undefined, name: name || undefined, input: text },
+        }
         continue
       }
       if (pending) {
         pendingTools = { ...pendingTools }
-        delete pendingTools[tool.index]
+        delete pendingTools[index]
       }
       const result = ToolStream.appendOrStart(
         ADAPTER,
         tools,
-        tool.index,
+        index,
         { id: id || undefined, name: name || undefined, text },
         "OpenAI Chat tool call delta is missing id or name",
       )
@@ -743,6 +803,8 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
         reasoningDetails: state.reasoningDetails,
         reasoningDetailsObserved,
         reasoningEmitted,
+        latestToolIndex,
+        nextToolIndex,
       },
       events,
     ] as const
@@ -750,14 +812,18 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
 
 const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
   const events: LLMEvent[] = []
-  const hasToolCalls = state.toolCallEvents.length > 0
+  const toolCallEvents =
+    state.finishReason === undefined && Object.keys(state.tools).length > 0
+      ? Effect.runSync(ToolStream.finishAll(ADAPTER, state.tools)).events
+      : state.toolCallEvents
+  const hasToolCalls = toolCallEvents.length > 0
   const reason = state.finishReason
     ? {
         ...state.finishReason,
         normalized:
           state.finishReason.normalized === "stop" && hasToolCalls ? "tool-calls" : state.finishReason.normalized,
       }
-    : undefined
+    : { normalized: hasToolCalls ? ("tool-calls" as const) : ("unknown" as const) }
   const metadata = reasoningMetadata(
     state.reasoningField,
     state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
@@ -767,9 +833,9 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
       ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
       : state.lifecycle
   const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
-  const lifecycle = state.toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
-  events.push(...state.toolCallEvents)
-  if (reason) Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
+  const lifecycle = toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
+  events.push(...toolCallEvents)
+  Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
   return events
 }
 
@@ -799,6 +865,7 @@ export const protocol = Protocol.make({
       reasoningDetails: [],
       reasoningDetailsObserved: false,
       reasoningEmitted: false,
+      nextToolIndex: 0,
     }),
     step,
     onHalt: finishEvents,
@@ -817,4 +884,4 @@ export const route = Route.make({
   transport: httpTransport,
 })
 
-export * as OpenAIChat from "./openai-chat"
+export * as OpenAIChat from "./openai-chat.js"

packages/ai/src/protocols/openai-compatible-chat.ts

@@ -1,7 +1,7 @@
-import { Route, type RouteRoutedLanguageModelInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import * as OpenAIChat from "./openai-chat"
+import { Route, type RouteRoutedLanguageModelInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import * as OpenAIChat from "./openai-chat.js"
 
 const ADAPTER = "openai-compatible-chat"
 
@@ -22,4 +22,4 @@ export const route = Route.make({
   framing: Framing.sse,
 })
 
-export * as OpenAICompatibleChat from "./openai-compatible-chat"
+export * as OpenAICompatibleChat from "./openai-compatible-chat.js"

packages/ai/src/protocols/openai-compatible-responses.ts

@@ -1,6 +1,6 @@
-import { Route, type RouteRoutedLanguageModelInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { OpenResponses } from "./open-responses"
+import { Route, type RouteRoutedLanguageModelInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { OpenResponses } from "./open-responses.js"
 
 const ADAPTER = "openai-compatible-responses"
 
@@ -19,4 +19,4 @@ export const route = Route.make({
   transport: OpenResponses.httpTransport,
 })
 
-export * as OpenAICompatibleResponses from "./openai-compatible-responses"
+export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"

packages/ai/src/protocols/openai-images.ts

@@ -7,8 +7,8 @@ import {
   type ImageInput,
   type ImageRequestFor,
   type ImageRoute,
-} from "../image"
-import { Auth, type Definition as AuthDefinition } from "../route/auth"
+} from "../image.js"
+import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
 import {
   InvalidProviderOutputReason,
   AIError,
@@ -16,10 +16,10 @@ import {
   mergeHttpOptions,
   mergeJsonRecords,
   type HttpOptions,
-} from "../schema"
-import { ProviderShared } from "./shared"
-import { ImageInputs } from "./utils/image-input"
-import { OpenAIImage } from "./utils/openai-image"
+} from "../schema/index.js"
+import { ProviderShared } from "./shared.js"
+import { ImageInputs } from "./utils/image-input.js"
+import { OpenAIImage } from "./utils/openai-image.js"
 
 const ADAPTER = "openai-images"
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"

packages/ai/src/protocols/openai-responses-channel.ts

@@ -0,0 +1,164 @@
+import { AIError, TransportReason } from "../schema/index.js"
+import type { ChannelCheckpoint, ChannelObservation, WebSocketChannelDriver } from "../route/transport/index.js"
+import { Effect, Option, Schema } from "effect"
+import * as ProviderShared from "./shared.js"
+import { OpenResponses } from "./open-responses.js"
+
+const PROTOCOL = "openai-responses.websocket.v1"
+const VERSION = 1
+const decodeEvent = Schema.decodeUnknownEffect(OpenResponses.protocol.stream.event)
+
+interface CheckpointValue {
+  readonly version: typeof VERSION
+  readonly responseID: string
+  readonly request: Readonly<Record<string, unknown>>
+  readonly output: ReadonlyArray<unknown>
+}
+
+export interface DriverInput {
+  readonly id: string
+  readonly name: string
+  readonly request: Readonly<Record<string, unknown>>
+  readonly message: string
+  readonly base: WebSocketChannelDriver
+}
+
+const checkpointValue = (checkpoint: ChannelCheckpoint | undefined): CheckpointValue | undefined => {
+  if (checkpoint?.protocol !== PROTOCOL || !ProviderShared.isRecord(checkpoint.value)) return undefined
+  if (checkpoint.value.version !== VERSION) return undefined
+  if (typeof checkpoint.value.responseID !== "string" || checkpoint.value.responseID.trim().length === 0)
+    return undefined
+  if (!ProviderShared.isRecord(checkpoint.value.request) || !Array.isArray(checkpoint.value.output)) return undefined
+  return {
+    version: VERSION,
+    responseID: checkpoint.value.responseID,
+    request: checkpoint.value.request,
+    output: checkpoint.value.output,
+  }
+}
+
+const canonical = (value: unknown): string => {
+  if (value === undefined) return "undefined"
+  if (Array.isArray(value)) return `[${value.map(canonical).join(",")}]`
+  if (!ProviderShared.isRecord(value)) return ProviderShared.encodeJson(value)
+  return `{${Object.keys(value)
+    .sort()
+    .map((key) => `${ProviderShared.encodeJson(key)}:${canonical(value[key])}`)
+    .join(",")}}`
+}
+
+const json = (value: unknown) => {
+  if (typeof value !== "string") return value
+  return Option.getOrElse(Schema.decodeUnknownOption(ProviderShared.Json)(value), () => value)
+}
+
+const comparable = (value: unknown) => {
+  if (!ProviderShared.isRecord(value)) return value
+  if (value.type === "message" && value.role === "assistant")
+    return {
+      role: "assistant",
+      content: value.content,
+      ...(value.phase === undefined ? {} : { phase: value.phase }),
+    }
+  if (value.type === "function_call")
+    return {
+      type: value.type,
+      call_id: value.call_id,
+      name: value.name,
+      arguments: json(value.arguments),
+    }
+  if (value.type === "reasoning")
+    return {
+      type: value.type,
+      summary: value.summary,
+      encrypted_content: value.encrypted_content,
+    }
+  return value
+}
+
+const invariant = (request: Readonly<Record<string, unknown>>) => {
+  const { type: _type, input: _input, previous_response_id: _previousResponseID, ...rest } = request
+  return rest
+}
+
+const incremental = (
+  request: Readonly<Record<string, unknown>>,
+  checkpoint: CheckpointValue,
+): ReadonlyArray<unknown> | undefined => {
+  const input = request.input
+  const previousInput = checkpoint.request.input
+  if (!Array.isArray(input) || !Array.isArray(previousInput)) return undefined
+  if (canonical(invariant(request)) !== canonical(invariant(checkpoint.request))) return undefined
+  const baseline = [...previousInput, ...checkpoint.output]
+  if (input.length <= baseline.length) return undefined
+  if (!baseline.every((item, index) => canonical(comparable(item)) === canonical(comparable(input[index]))))
+    return undefined
+  return input.slice(baseline.length)
+}
+
+const code = (event: OpenResponses.Event) => event.code || event.error?.code || event.response?.error?.code || undefined
+
+const rejected = (
+  input: DriverInput,
+  observation: Extract<ChannelObservation, { readonly type: "provider-failure" }>,
+  recovery: "retry-full" | "rotate-and-retry-full",
+): ChannelObservation => ({
+  type: "rejected",
+  recovery,
+  error: new AIError({
+    module: input.id,
+    method: "stream",
+    reason: new TransportReason({
+      message: observation.error.message,
+      transport: "websocket",
+      operation: "read",
+      phase: "receive",
+      delivery: "rejected",
+      recovery,
+    }),
+  }),
+})
+
+export const driver = (input: DriverInput): WebSocketChannelDriver => {
+  const { previous_response_id: _previousResponseID, ...request } = input.request
+  let output: unknown[] = []
+  return {
+    create: (checkpoint) =>
+      Effect.sync(() => {
+        output = []
+        const previous = checkpointValue(checkpoint)
+        const delta = previous ? incremental(request, previous) : undefined
+        if (!previous || !delta) return { message: ProviderShared.encodeJson(request), mode: "full" as const }
+        return {
+          message: ProviderShared.encodeJson({ ...request, input: delta, previous_response_id: previous.responseID }),
+          mode: "incremental" as const,
+        }
+      }),
+    observe: (create, frame) =>
+      Effect.gen(function* () {
+        const event = yield* decodeEvent(frame).pipe(
+          Effect.mapError(() => ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame)),
+        )
+        const observation = yield* input.base.observe(create, frame)
+        if (event.type === "response.output_item.done" && event.item) output.push(event.item)
+        if (observation.type === "provider-failure") {
+          const rejection = code(event)
+          if (rejection === "previous_response_not_found") return rejected(input, observation, "retry-full")
+          if (rejection === "websocket_connection_limit_reached")
+            return rejected(input, observation, "rotate-and-retry-full")
+        }
+        if (observation.type !== "completed") return observation
+        const responseID = event.response?.id
+        if (!responseID || responseID.trim().length === 0) return observation
+        return {
+          ...observation,
+          checkpoint: {
+            protocol: PROTOCOL,
+            value: { version: VERSION, responseID, request, output: output.slice() } satisfies CheckpointValue,
+          },
+        }
+      }),
+  }
+}
+
+export const OpenAIResponsesChannel = { driver } as const

packages/ai/src/protocols/openai-responses.ts

@@ -1,18 +1,23 @@
 import { Effect, Encoding, Schema } from "effect"
-import { Route } from "../route/client"
-import { Auth } from "../route/auth"
-import { Endpoint } from "../route/endpoint"
-import { Protocol } from "../route/protocol"
-import { HttpTransport, WebSocketTransport } from "../route/transport"
-import { LLMEvent, LLMRequest, type JsonSchema, type ToolDefinition } from "../schema"
-import { OpenResponses } from "./open-responses"
-import { optionalArray, ProviderShared } from "./shared"
-import { Lifecycle } from "./utils/lifecycle"
-import { OpenAIImage } from "./utils/openai-image"
-import { ToolSchemaProjection } from "./utils/tool-schema"
+import { Headers } from "effect/unstable/http"
+import { Route } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Protocol } from "../route/protocol.js"
+import { HttpTransport } from "../route/transport/index.js"
+import { LLMEvent, LLMRequest, type JsonSchema, type ToolDefinition } from "../schema/index.js"
+import { OpenResponses } from "./open-responses.js"
+import { optionalArray, ProviderShared } from "./shared.js"
+import { Lifecycle } from "./utils/lifecycle.js"
+import { OpenAIImage } from "./utils/openai-image.js"
+import { ToolSchemaProjection } from "./utils/tool-schema.js"
+import { OpenResponsesChannel } from "./open-responses-channel.js"
+import { OpenAIResponsesChannel } from "./openai-responses-channel.js"
 
 const ADAPTER = "openai-responses"
 const NAME = "OpenAI Responses"
+const WEBSOCKET_PROTOCOL_HEADER = "responses_websockets=2026-02-06"
+const WEBSOCKET_ROTATE_AFTER_MS = 55 * 60 * 1000
 export const DEFAULT_BASE_URL = "https://api.openai.com/v1"
 export const PATH = OpenResponses.PATH
 
@@ -57,16 +62,6 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>
 
-const OpenAIResponsesWebSocketMessage = Schema.StructWithRest(
-  Schema.Struct({
-    type: Schema.tag("response.create"),
-    ...OpenAIResponsesCoreFields,
-  }),
-  [Schema.Record(Schema.String, Schema.Unknown)],
-)
-type OpenAIResponsesWebSocketMessage = Schema.Schema.Type<typeof OpenAIResponsesWebSocketMessage>
-const encodeWebSocketMessage = Schema.encodeSync(Schema.fromJsonString(OpenAIResponsesWebSocketMessage))
-
 const extension = {
   id: ADAPTER,
   name: NAME,
@@ -249,6 +244,13 @@ const endpoint = Endpoint.path<OpenAIResponsesBody>(PATH, { baseURL: DEFAULT_BAS
 const auth = Auth.none
 
 export const httpTransport = HttpTransport.sseJson.with<OpenAIResponsesBody>()
+export const transport = OpenResponsesChannel.transport<OpenAIResponsesBody>({
+  id: ADAPTER,
+  name: NAME,
+  rotateAfterMs: WEBSOCKET_ROTATE_AFTER_MS,
+  headers: (headers) => Headers.set(headers, "openai-beta", headers["openai-beta"] ?? WEBSOCKET_PROTOCOL_HEADER),
+  driver: (input) => OpenAIResponsesChannel.driver({ id: ADAPTER, name: NAME, ...input }),
+})
 
 export const route = Route.make({
   id: ADAPTER,
@@ -257,37 +259,8 @@ export const route = Route.make({
   protocol,
   endpoint,
   auth,
-  transport: httpTransport,
+  transport,
   defaults: { providerOptions: { openai: { store: false } } },
 })
 
-const decodeWebSocketMessage = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenAIResponsesWebSocketMessage))
-
-const webSocketMessage = (body: OpenAIResponsesBody | Record<string, unknown>) =>
-  Effect.gen(function* () {
-    if (!ProviderShared.isRecord(body))
-      return yield* ProviderShared.invalidRequest("OpenAI Responses WebSocket body must be a JSON object")
-    const { stream: _stream, ...message } = body
-    return yield* decodeWebSocketMessage({ ...message, type: "response.create" })
-  })
-
-export const webSocketTransport = WebSocketTransport.jsonTransport.with<
-  OpenAIResponsesBody,
-  OpenAIResponsesWebSocketMessage
->({
-  toMessage: webSocketMessage,
-  encodeMessage: encodeWebSocketMessage,
-})
-
-export const webSocketRoute = Route.make({
-  id: `${ADAPTER}-websocket`,
-  provider: "openai",
-  providerMetadataKey: "openai",
-  protocol,
-  endpoint,
-  auth,
-  transport: webSocketTransport,
-  defaults: { providerOptions: { openai: { store: false } } },
-})
-
-export * as OpenAIResponses from "./openai-responses"
+export * as OpenAIResponses from "./openai-responses.js"

packages/ai/src/protocols/shared.ts

@@ -12,8 +12,8 @@ import {
   type MediaPart,
   type TextPart,
   type ToolResultPart,
-} from "../schema"
-import { isRecord } from "../utils/record"
+} from "../schema/index.js"
+import { isRecord } from "../utils/record.js"
 export { isRecord }
 
 export const Json = Schema.fromJsonString(Schema.Unknown)
@@ -324,4 +324,4 @@ export const jsonPost = (input: { readonly url: string; readonly body: string; r
     HttpClientRequest.bodyText(input.body, "application/json"),
   )
 
-export * as ProviderShared from "./shared"
+export * as ProviderShared from "./shared.js"

packages/ai/src/protocols/utils/bedrock-auth.ts

@@ -1,8 +1,8 @@
 import { AwsV4Signer } from "aws4fetch"
 import { Effect } from "effect"
 import { Headers } from "effect/unstable/http"
-import { Auth, type AuthInput } from "../../route/auth"
-import { ProviderShared } from "../shared"
+import { Auth, type AuthInput } from "../../route/auth.js"
+import { ProviderShared } from "../shared.js"
 
 /**
  * AWS credentials for SigV4 signing. Bedrock also supports Bearer API key auth,
@@ -74,4 +74,4 @@ export const sigV4 = (
 /** Bedrock route auth defaults to SigV4 and expects credentials from route configuration. */
 export const auth = sigV4(undefined)
 
-export * as BedrockAuth from "./bedrock-auth"
+export * as BedrockAuth from "./bedrock-auth.js"

packages/ai/src/protocols/utils/bedrock-cache.ts

@@ -1,6 +1,6 @@
 import { Schema } from "effect"
-import type { CacheHint } from "../../schema"
-import { newBreakpoints, ttlBucket, type Breakpoints } from "./cache"
+import type { CacheHint } from "../../schema/index.js"
+import { newBreakpoints, ttlBucket, type Breakpoints } from "./cache.js"
 
 // Bedrock cache markers are positional: emit a `cachePoint` block immediately
 // after the content the caller wants treated as a cacheable prefix. Bedrock
@@ -18,7 +18,7 @@ export type CachePointBlock = Schema.Schema.Type<typeof CachePointBlock>
 // budget is respected across `system`, `messages`, and `tools`.
 export const BEDROCK_BREAKPOINT_CAP = 4
 
-export type { Breakpoints } from "./cache"
+export type { Breakpoints } from "./cache.js"
 export const breakpoints = () => newBreakpoints(BEDROCK_BREAKPOINT_CAP)
 
 const DEFAULT_5M: CachePointBlock = { cachePoint: { type: "default" } }
@@ -34,4 +34,4 @@ export const block = (breakpoints: Breakpoints, cache: CacheHint | undefined): C
   return ttlBucket(cache.ttlSeconds) === "1h" ? DEFAULT_1H : DEFAULT_5M
 }
 
-export * as BedrockCache from "./bedrock-cache"
+export * as BedrockCache from "./bedrock-cache.js"

packages/ai/src/protocols/utils/bedrock-media.ts

@@ -1,6 +1,6 @@
 import { Effect, Schema } from "effect"
-import type { MediaPart } from "../../schema"
-import { ProviderShared } from "../shared"
+import type { MediaPart } from "../../schema/index.js"
+import { ProviderShared } from "../shared.js"
 
 // Bedrock Converse accepts image `format` as the file extension and
 // `source.bytes` as base64 in the JSON wire format.
@@ -89,4 +89,4 @@ export const lower = Effect.fn("BedrockMedia.lower")(function* (part: MediaPart)
   return yield* ProviderShared.invalidRequest(`Bedrock Converse does not support media type ${part.mediaType}`)
 })
 
-export * as BedrockMedia from "./bedrock-media"
+export * as BedrockMedia from "./bedrock-media.js"

packages/ai/src/protocols/utils/gemini-tool-schema.ts

@@ -1,4 +1,4 @@
-import { isRecord } from "../../utils/record"
+import { isRecord } from "../../utils/record.js"
 
 // Gemini accepts a JSON Schema-like dialect for tool parameters, but rejects a
 // handful of common JSON Schema shapes. Keep this projection isolated so the
@@ -61,39 +61,59 @@ const emptyObjectSchema = (schema: Record<string, unknown>) =>
   (!isRecord(schema.properties) || Object.keys(schema.properties).length === 0) &&
   !schema.additionalProperties
 
-const projectNode = (schema: unknown): Record<string, unknown> | undefined => {
+const projectNode = (schema: unknown, nested = false): Record<string, unknown> | undefined => {
   if (!isRecord(schema)) return undefined
-  if (emptyObjectSchema(schema)) return undefined
-  return Object.fromEntries(
+  if (!nested && emptyObjectSchema(schema)) return undefined
+  const types = Array.isArray(schema.type) ? schema.type.filter((type) => type !== "null") : undefined
+  const anyOf = Array.isArray(schema.anyOf) ? schema.anyOf : undefined
+  const hasNullAnyOf = anyOf?.some((item) => isRecord(item) && item.type === "null") ?? false
+  const anyOfTypes = hasNullAnyOf ? anyOf?.filter((item) => !isRecord(item) || item.type !== "null") : anyOf
+  const flattenedAnyOf = hasNullAnyOf && anyOfTypes?.length === 1 ? projectNode(anyOfTypes[0], true) : undefined
+  const result = Object.fromEntries(
     [
       ["description", schema.description],
       ["required", schema.required],
       ["format", schema.format],
-      ["type", Array.isArray(schema.type) ? schema.type.filter((type) => type !== "null")[0] : schema.type],
-      ["nullable", Array.isArray(schema.type) && schema.type.includes("null") ? true : undefined],
+      ["type", types ? (types.length === 0 ? "null" : undefined) : schema.type],
+      [
+        "nullable",
+        (Array.isArray(schema.type) && schema.type.includes("null") && types && types.length > 0) || hasNullAnyOf
+          ? true
+          : undefined,
+      ],
       ["enum", schema.const !== undefined ? [schema.const] : schema.enum],
       [
         "properties",
         isRecord(schema.properties)
-          ? Object.fromEntries(Object.entries(schema.properties).map(([key, value]) => [key, projectNode(value)]))
+          ? Object.fromEntries(Object.entries(schema.properties).map(([key, value]) => [key, projectNode(value, true)]))
           : undefined,
       ],
       [
         "items",
         Array.isArray(schema.items)
-          ? schema.items.map(projectNode)
+          ? schema.items.map((item) => projectNode(item, true))
           : schema.items === undefined
             ? undefined
-            : projectNode(schema.items),
+            : projectNode(schema.items, true),
       ],
-      ["allOf", Array.isArray(schema.allOf) ? schema.allOf.map(projectNode) : undefined],
-      ["anyOf", Array.isArray(schema.anyOf) ? schema.anyOf.map(projectNode) : undefined],
-      ["oneOf", Array.isArray(schema.oneOf) ? schema.oneOf.map(projectNode) : undefined],
+      ["allOf", Array.isArray(schema.allOf) ? schema.allOf.map((item) => projectNode(item, true)) : undefined],
+      [
+        "anyOf",
+        anyOfTypes
+          ? hasNullAnyOf && anyOfTypes.length === 1
+            ? undefined
+            : anyOfTypes.map((item) => projectNode(item, true))
+          : types && types.length > 0
+            ? types.map((type) => ({ type }))
+            : undefined,
+      ],
+      ["oneOf", Array.isArray(schema.oneOf) ? schema.oneOf.map((item) => projectNode(item, true)) : undefined],
       ["minLength", schema.minLength],
     ].filter((entry) => entry[1] !== undefined),
   )
+  return flattenedAnyOf ? { ...result, ...flattenedAnyOf } : result
 }
 
 export const convert = (schema: unknown) => projectNode(sanitizeNode(schema))
 
-export * as GeminiToolSchema from "./gemini-tool-schema"
+export * as GeminiToolSchema from "./gemini-tool-schema.js"

packages/ai/src/protocols/utils/image-input.ts

@@ -1,6 +1,6 @@
 import { Effect, Encoding } from "effect"
-import type { ImageInput } from "../../image"
-import { InvalidRequestReason, AIError } from "../../schema"
+import type { ImageInput } from "../../image.js"
+import { InvalidRequestReason, AIError } from "../../schema/index.js"
 
 const invalid = (module: string, message: string) =>
   new AIError({

packages/ai/src/protocols/utils/lifecycle.ts

@@ -1,4 +1,4 @@
-import { LLMEvent, type FinishReasonDetails, type ProviderMetadata, type Usage } from "../../schema"
+import { LLMEvent, type FinishReasonDetails, type ProviderMetadata, type Usage } from "../../schema/index.js"
 
 export interface State {
   readonly stepStarted: boolean
@@ -102,4 +102,4 @@ export const finish = (
   return { ...stepped, stepStarted: false }
 }
 
-export * as Lifecycle from "./lifecycle"
+export * as Lifecycle from "./lifecycle.js"

packages/ai/src/protocols/utils/open-responses-options.ts

@@ -1,5 +1,5 @@
 import { Schema } from "effect"
-import { TextVerbosity, type LLMRequest } from "../../schema"
+import { TextVerbosity, type LLMRequest } from "../../schema/index.js"
 
 export const ResponseIncludables = [
   "file_search_call.results",
@@ -33,7 +33,6 @@ export const ServiceTierSchema = Schema.Literals(ServiceTiers)
 export interface Resolved {
   readonly instructions?: string
   readonly store?: boolean
-  readonly promptCacheKey?: string
   readonly reasoningEffort?: string
   readonly reasoningSummary?: "auto" | "concise" | "detailed"
   readonly include?: ReadonlyArray<ResponseIncludable>
@@ -50,7 +49,6 @@ export const resolve = (request: LLMRequest): Resolved => {
   return {
     instructions: typeof input?.instructions === "string" ? input.instructions : undefined,
     store: typeof input?.store === "boolean" ? input.store : undefined,
-    promptCacheKey: typeof input?.promptCacheKey === "string" ? input.promptCacheKey : undefined,
     reasoningEffort: typeof input?.reasoningEffort === "string" ? input.reasoningEffort : undefined,
     reasoningSummary:
       reasoningSummary === "auto" || reasoningSummary === "concise" || reasoningSummary === "detailed"
@@ -62,4 +60,4 @@ export const resolve = (request: LLMRequest): Resolved => {
   }
 }
 
-export * as OpenResponsesOptions from "./open-responses-options"
+export * as OpenResponsesOptions from "./open-responses-options.js"

packages/ai/src/protocols/utils/openai-options.ts

@@ -1,5 +1,5 @@
-import { ReasoningEfforts } from "../../schema"
-import { OpenResponsesOptions } from "./open-responses-options"
+import { ReasoningEfforts } from "../../schema/index.js"
+import { OpenResponsesOptions } from "./open-responses-options.js"
 
 export const OpenAIReasoningEfforts = ReasoningEfforts
 export type OpenAIReasoningEffort = string
@@ -20,4 +20,4 @@ export const isReasoningEffort = (effort: unknown): effort is OpenAIReasoningEff
 
 export const resolve = OpenResponsesOptions.resolve
 
-export * as OpenAIOptions from "./openai-options"
+export * as OpenAIOptions from "./openai-options.js"

packages/ai/src/protocols/utils/tool-schema.ts

@@ -1,6 +1,6 @@
-import type { JsonSchema, LanguageModelToolSchemaCompatibility } from "../../schema"
-import { isRecord } from "../../utils/record"
-import { GeminiToolSchema } from "./gemini-tool-schema"
+import type { JsonSchema, LanguageModelToolSchemaCompatibility } from "../../schema/index.js"
+import { isRecord } from "../../utils/record.js"
+import { GeminiToolSchema } from "./gemini-tool-schema.js"
 
 const removeNullSchemas = (value: unknown): unknown => {
   if (Array.isArray(value)) return value.map(removeNullSchemas)

packages/ai/src/protocols/utils/tool-stream.ts

@@ -1,6 +1,6 @@
 import { Effect } from "effect"
-import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema"
-import { eventError, parseToolInput, type ToolAccumulator } from "../shared"
+import { AIError, LLMEvent, type ProviderMetadata, type ToolCall, type ToolInputError } from "../../schema/index.js"
+import { eventError, parseToolInput, type ToolAccumulator } from "../shared.js"
 
 type StreamKey = string | number
 
@@ -223,4 +223,4 @@ export const finishAll = <K extends StreamKey>(route: string, tools: State<K>) =
     }
   })
 
-export * as ToolStream from "./tool-stream"
+export * as ToolStream from "./tool-stream.js"

packages/ai/src/protocols/xai-images.ts

@@ -1,7 +1,7 @@
 import { Effect, Encoding, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
-import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image"
-import { Auth, type Definition as AuthDefinition } from "../route/auth"
+import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
+import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
 import {
   InvalidProviderOutputReason,
   AIError,
@@ -9,9 +9,9 @@ import {
   mergeHttpOptions,
   mergeJsonRecords,
   type HttpOptions,
-} from "../schema"
-import { ProviderShared, optionalNull } from "./shared"
-import { ImageInputs } from "./utils/image-input"
+} from "../schema/index.js"
+import { ProviderShared, optionalNull } from "./shared.js"
+import { ImageInputs } from "./utils/image-input.js"
 
 const ADAPTER = "xai-images"
 export const DEFAULT_BASE_URL = "https://api.x.ai/v1"

packages/ai/src/protocols/zai-images.ts

@@ -1,10 +1,16 @@
 import { Effect, Schema } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
-import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image"
-import { Auth, type Definition as AuthDefinition } from "../route/auth"
-import { InvalidProviderOutputReason, AIError, mergeHttpOptions, mergeJsonRecords, type HttpOptions } from "../schema"
-import { ProviderShared } from "./shared"
-import { ImageInputs } from "./utils/image-input"
+import { GeneratedImage, ImageModel, ImageResponse, type ImageRequestFor, type ImageRoute } from "../image.js"
+import { Auth, type Definition as AuthDefinition } from "../route/auth.js"
+import {
+  InvalidProviderOutputReason,
+  AIError,
+  mergeHttpOptions,
+  mergeJsonRecords,
+  type HttpOptions,
+} from "../schema/index.js"
+import { ProviderShared } from "./shared.js"
+import { ImageInputs } from "./utils/image-input.js"
 
 const ADAPTER = "zai-images"
 export const DEFAULT_BASE_URL = "https://api.z.ai/api/paas/v4"

packages/ai/src/provider-error.ts

@@ -12,7 +12,7 @@ import {
   type HttpContext,
   type HttpRateLimitDetails,
   type ProviderMetadata,
-} from "./schema"
+} from "./schema/index.js"
 
 const patterns = [
   /prompt is too long/i,
@@ -67,6 +67,7 @@ const SERVER_CODES = new Set([
   "overloaded_error",
   "server_error",
   "server_is_overloaded",
+  "slow_down",
   "serviceunavailableexception",
 ])
 const INVALID_REQUEST_CODES = new Set(["invalid_prompt", "invalid_request_error", "validationexception"])

packages/ai/src/provider-package.ts

@@ -1,4 +1,4 @@
-import type { LanguageModel, ProviderOptions } from "./schema"
+import type { LanguageModel, ProviderOptions } from "./schema/index.js"
 
 export interface Settings extends Readonly<Record<string, unknown>> {
   readonly baseURL?: string
@@ -18,4 +18,4 @@ export interface Definition<
   readonly model: (modelID: string, settings: ProviderSettings) => LanguageModel<Options>
 }
 
-export * as ProviderPackage from "./provider-package"
+export * as ProviderPackage from "./provider-package.js"

packages/ai/src/provider.ts

@@ -1,4 +1,4 @@
-import type { LanguageModel, ModelID, ProviderID } from "./schema"
+import type { LanguageModel, ModelID, ProviderID } from "./schema/index.js"
 
 export type LanguageModelOptions = Pick<LanguageModel.Input, "defaults" | "compatibility">
 
@@ -33,4 +33,4 @@ export const make = <DefinitionType extends DefinitionShape>(
   definition: NoExtraFields<DefinitionType, DefinitionShape>,
 ) => definition
 
-export * as Provider from "./provider"
+export * as Provider from "./provider.js"

packages/ai/src/providers.ts

@@ -1 +1 @@
-export * from "./providers/index"
+export * from "./providers/index.js"

packages/ai/src/providers/amazon-bedrock-mantle.ts

@@ -1,11 +1,11 @@
-import { Auth } from "../route/auth"
-import type { Route as RouteDef, RouteDefaultsInput } from "../route/client"
-import type { ProviderPackage } from "../provider-package"
-import { OpenAIChat } from "../protocols/openai-chat"
-import { OpenAIResponses } from "../protocols/openai-responses"
-import { BedrockAuth, type Credentials } from "../protocols/utils/bedrock-auth"
-import { ProviderID, type ModelID } from "../schema"
-import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
+import { Auth } from "../route/auth.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAIChat } from "../protocols/openai-chat.js"
+import { OpenAIResponses } from "../protocols/openai-responses.js"
+import { BedrockAuth, type Credentials } from "../protocols/utils/bedrock-auth.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("amazon-bedrock")
 

packages/ai/src/providers/amazon-bedrock.ts

@@ -1,9 +1,9 @@
-import type { RouteDefaultsInput } from "../route/client"
-import { Auth } from "../route/auth"
-import type { ProviderPackage } from "../provider-package"
-import { ProviderID, type ModelID } from "../schema"
-import * as BedrockConverse from "../protocols/bedrock-converse"
-import type { BedrockCredentials } from "../protocols/bedrock-converse"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import * as BedrockConverse from "../protocols/bedrock-converse.js"
+import type { BedrockCredentials } from "../protocols/bedrock-converse.js"
 
 export const id = ProviderID.make("amazon-bedrock")
 

packages/ai/src/providers/amazon-bedrock/mantle.ts

@@ -1,2 +1,2 @@
-export { chatModel as model } from "../amazon-bedrock-mantle"
-export type { Settings } from "../amazon-bedrock-mantle"
+export { chatModel as model } from "../amazon-bedrock-mantle.js"
+export type { Settings } from "../amazon-bedrock-mantle.js"

packages/ai/src/providers/amazon-bedrock/mantle/chat.ts

@@ -1,2 +1,2 @@
-export { chatModel as model } from "../../amazon-bedrock-mantle"
-export type { Settings } from "../../amazon-bedrock-mantle"
+export { chatModel as model } from "../../amazon-bedrock-mantle.js"
+export type { Settings } from "../../amazon-bedrock-mantle.js"

packages/ai/src/providers/amazon-bedrock/mantle/responses.ts

@@ -1,2 +1,2 @@
-export { responsesModel as model } from "../../amazon-bedrock-mantle"
-export type { Settings } from "../../amazon-bedrock-mantle"
+export { responsesModel as model } from "../../amazon-bedrock-mantle.js"
+export type { Settings } from "../../amazon-bedrock-mantle.js"

packages/ai/src/providers/anthropic-compatible.ts

@@ -1,9 +1,9 @@
-import type { ProviderPackage } from "../provider-package"
-import { AnthropicMessages } from "../protocols/anthropic-messages"
-import { Auth } from "../route/auth"
-import type { ProviderAuthOption } from "../route/auth-options"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
+import type { ProviderPackage } from "../provider-package.js"
+import { AnthropicMessages } from "../protocols/anthropic-messages.js"
+import { Auth } from "../route/auth.js"
+import type { ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
 
 export type AnthropicOptionsInput = AnthropicMessages.OptionsInput
 export type AnthropicProviderOptionsInput = AnthropicMessages.ProviderOptionsInput
@@ -74,4 +74,4 @@ export const model: ProviderPackage.Definition<Settings, AnthropicMessages.Provi
   }).model(modelID)
 }
 
-export * as AnthropicCompatible from "./anthropic-compatible"
+export * as AnthropicCompatible from "./anthropic-compatible.js"

packages/ai/src/providers/anthropic.ts

@@ -1,10 +1,10 @@
-import type { RouteDefaultsInput } from "../route/client"
-import { Auth } from "../route/auth"
-import type { ProviderAuthOption } from "../route/auth-options"
-import type { ProviderPackage } from "../provider-package"
-import { ProviderID, type ModelID } from "../schema"
-import { AnthropicMessages } from "../protocols/anthropic-messages"
-import { AnthropicCompatible } from "./anthropic-compatible"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import type { ProviderAuthOption } from "../route/auth-options.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { AnthropicMessages } from "../protocols/anthropic-messages.js"
+import { AnthropicCompatible } from "./anthropic-compatible.js"
 
 export type AnthropicOptionsInput = AnthropicMessages.OptionsInput
 export type AnthropicProviderOptionsInput = AnthropicMessages.ProviderOptionsInput

packages/ai/src/providers/azure.ts

@@ -1,11 +1,12 @@
-import { Auth } from "../route/auth"
-import { type AtLeastOne, type ProviderAuthOption } from "../route/auth-options"
-import type { Route as RouteDef, RouteDefaultsInput } from "../route/client"
-import type { ProviderPackage } from "../provider-package"
-import { ProviderID, type ModelID } from "../schema"
-import * as OpenAIChat from "../protocols/openai-chat"
-import * as OpenAIResponses from "../protocols/openai-responses"
-import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
+import { Auth } from "../route/auth.js"
+import { type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import * as OpenAIChat from "../protocols/openai-chat.js"
+import * as OpenAIResponses from "../protocols/openai-responses.js"
+import { ProviderShared } from "../protocols/shared.js"
+import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("azure")
 const routeAuth = Auth.remove("authorization")
@@ -19,7 +20,7 @@ export type LanguageModelOptions = AzureURL &
   ProviderAuthOption<"optional"> & {
     readonly apiVersion?: string
     readonly queryParams?: Record<string, string>
-    readonly useCompletionUrls?: boolean
+    readonly useDeploymentBasedUrls?: boolean
     readonly providerOptions?: OpenAIProviderOptionsInput
   }
 export type Config = LanguageModelOptions
@@ -29,27 +30,22 @@ export type Settings = ProviderPackage.Settings &
     readonly apiKey?: string
     readonly apiVersion?: string
     readonly queryParams?: Readonly<Record<string, string>>
+    readonly useDeploymentBasedUrls?: boolean
     readonly providerOptions?: OpenAIProviderOptionsInput
   }
 
-const resourceBaseURL = (resourceName: string) => `https://${resourceName.trim()}.openai.azure.com/openai/v1`
+const resourceBaseURL = (resourceName: string) => `https://${resourceName.trim()}.openai.azure.com/openai`
 
 const responsesRoute = OpenAIResponses.route.with({
   id: "azure-openai-responses",
   provider: id,
   auth: routeAuth,
-  endpoint: {
-    query: { "api-version": "v1" },
-  },
 })
 
 const chatRoute = OpenAIChat.route.with({
   id: "azure-openai-chat",
   provider: id,
   auth: routeAuth,
-  endpoint: {
-    query: { "api-version": "v1" },
-  },
 })
 
 export const routes = [responsesRoute, chatRoute]
@@ -59,7 +55,7 @@ const defaults = (input: Config) => {
     apiKey: _,
     apiVersion: _apiVersion,
     resourceName: _resourceName,
-    useCompletionUrls: _useCompletionUrls,
+    useDeploymentBasedUrls: _useDeploymentBasedUrls,
     baseURL: _baseURL,
     queryParams: _queryParams,
     ...rest
@@ -80,37 +76,39 @@ const auth = (input: Config) => {
   )
 }
 
-const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) =>
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config, modelID: string | ModelID) =>
   route.with({
     auth: auth(input),
-    endpoint: {
-      // AtLeastOne guarantees at least one is set; baseURL wins if both are.
-      baseURL: input.baseURL ?? resourceBaseURL(input.resourceName!),
-      query: {
-        ...(input.apiVersion ? { "api-version": input.apiVersion } : {}),
-        ...input.queryParams,
-      },
-    },
+    endpoint: endpoint(input, modelID),
   })
 
+function endpoint(input: Config, modelID: string | ModelID) {
+  const baseURL = ProviderShared.trimBaseUrl(input.baseURL ?? resourceBaseURL(input.resourceName!))
+  const query = { "api-version": input.apiVersion ?? "v1", ...input.queryParams }
+
+  if (input.useDeploymentBasedUrls) return { baseURL: `${baseURL}/deployments/${modelID}`, query }
+  if (input.baseURL !== undefined && !new URL(input.baseURL).hostname.endsWith(".openai.azure.com")) {
+    return { baseURL, query: input.queryParams }
+  }
+  return { baseURL: `${baseURL}/v1`, query }
+}
+
 export const configure = (input: Config) => {
-  const configuredResponsesRoute = configuredRoute(responsesRoute, input)
-  const configuredChatRoute = configuredRoute(chatRoute, input)
   const modelDefaults = defaults(input)
 
   const responses = (modelID: string | ModelID) =>
-    configuredResponsesRoute
+    configuredRoute(responsesRoute, input, modelID)
       .with(withOpenAIOptions(modelID, modelDefaults))
       .model<OpenAIProviderOptionsInput>({ id: modelID })
 
   const chat = (modelID: string | ModelID) =>
-    configuredChatRoute
+    configuredRoute(chatRoute, input, modelID)
       .with(withOpenAIOptions(modelID, modelDefaults))
       .model<OpenAIProviderOptionsInput>({ id: modelID })
 
   return {
     id,
-    model: (modelID: string | ModelID) => (input.useCompletionUrls === true ? chat(modelID) : responses(modelID)),
+    model: responses,
     responses,
     chat,
     configure,
@@ -131,6 +129,7 @@ const config = (settings: Settings): Config => {
     limits: settings.limits,
     providerOptions: settings.providerOptions,
     queryParams: settings.queryParams === undefined ? undefined : { ...settings.queryParams },
+    useDeploymentBasedUrls: settings.useDeploymentBasedUrls,
   }
   if (settings.baseURL !== undefined) return { ...common, baseURL: settings.baseURL }
   if (settings.resourceName !== undefined) return { ...common, resourceName: settings.resourceName }

packages/ai/src/providers/azure/chat.ts

@@ -1,2 +1,2 @@
-export { chatModel as model } from "../azure"
-export type { Settings } from "../azure"
+export { chatModel as model } from "../azure.js"
+export type { Settings } from "../azure.js"

packages/ai/src/providers/azure/responses.ts

@@ -1,2 +1,2 @@
-export { responsesModel as model } from "../azure"
-export type { Settings } from "../azure"
+export { responsesModel as model } from "../azure.js"
+export type { Settings } from "../azure.js"

packages/ai/src/providers/cloudflare.ts

@@ -1,10 +1,10 @@
 import type { Config, Redacted } from "effect"
-import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat"
-import { Auth } from "../route/auth"
-import { AuthOptions, type AtLeastOne, type ProviderAuthOption } from "../route/auth-options"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
-import type { OpenAIProviderOptionsInput } from "./openai-options"
+import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
+import { Auth } from "../route/auth.js"
+import { AuthOptions, type AtLeastOne, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const aiGatewayID = ProviderID.make("cloudflare-ai-gateway")
 export const workersAIID = ProviderID.make("cloudflare-workers-ai")

packages/ai/src/providers/github-copilot.ts

@@ -1,71 +0,0 @@
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
-import * as OpenAIChat from "../protocols/openai-chat"
-import * as OpenAIResponses from "../protocols/openai-responses"
-import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
-
-export const id = ProviderID.make("github-copilot")
-
-// GitHub Copilot has no canonical public URL — callers (opencode, etc.) must
-// supply `baseURL` explicitly.
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL: string
-    readonly endpoint?: "chat" | "responses"
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export const shouldUseResponsesApi = (modelID: string | ModelID, endpoint?: LanguageModelOptions["endpoint"]) => {
-  if (endpoint) return endpoint === "responses"
-  const model = String(modelID)
-  const match = /^gpt-(\d+)/.exec(model)
-  if (!match) return false
-  return Number(match[1]) >= 5 && !model.startsWith("gpt-5-mini")
-}
-
-export const routes = [OpenAIResponses.route, OpenAIChat.route]
-
-const chatRoute = OpenAIChat.route.with({ provider: id })
-const responsesRoute = OpenAIResponses.route.with({ provider: id })
-
-const defaults = (options: LanguageModelOptions) => {
-  const { apiKey: _, auth: _auth, baseURL: _baseURL, endpoint: _endpoint, ...rest } = options
-  return rest
-}
-
-const configuredResponsesRoute = (options: LanguageModelOptions) =>
-  responsesRoute.with({
-    endpoint: { baseURL: options.baseURL },
-    auth: AuthOptions.bearer(options, []),
-  })
-
-const configuredChatRoute = (options: LanguageModelOptions) =>
-  chatRoute.with({
-    endpoint: { baseURL: options.baseURL },
-    auth: AuthOptions.bearer(options, []),
-  })
-
-export const configure = (options: LanguageModelOptions) => {
-  const responsesRoute = configuredResponsesRoute(options)
-  const chatRoute = configuredChatRoute(options)
-  const responses = (modelID: string | ModelID) =>
-    responsesRoute
-      .with(withOpenAIOptions(modelID, defaults(options)))
-      .model<OpenAIProviderOptionsInput>({ id: modelID })
-  const chat = (modelID: string | ModelID) =>
-    chatRoute.with(withOpenAIOptions(modelID, defaults(options))).model<OpenAIProviderOptionsInput>({ id: modelID })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      shouldUseResponsesApi(modelID, options.endpoint) ? responses(modelID) : chat(modelID),
-    responses,
-    chat,
-    configure,
-  }
-}
-
-export const provider = {
-  id,
-  configure,
-}

packages/ai/src/providers/google-vertex-chat.ts

@@ -1,9 +1,9 @@
-import type { ProviderPackage } from "../provider-package"
-import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
-import { GoogleVertexShared } from "./google-vertex-shared"
-import type { OpenAIProviderOptionsInput } from "./openai-options"
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { GoogleVertexShared } from "./google-vertex-shared.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("google-vertex")
 

packages/ai/src/providers/google-vertex-messages.ts

@@ -1,13 +1,13 @@
 import { Effect, Schema, Struct } from "effect"
-import type { ProviderPackage } from "../provider-package"
-import { AnthropicMessages } from "../protocols/anthropic-messages"
-import { Auth } from "../route/auth"
-import { Route, type RouteDefaultsInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { Protocol } from "../route/protocol"
-import { ProviderID, type ModelID } from "../schema"
-import { GoogleVertexShared } from "./google-vertex-shared"
+import type { ProviderPackage } from "../provider-package.js"
+import { AnthropicMessages } from "../protocols/anthropic-messages.js"
+import { Auth } from "../route/auth.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { Protocol } from "../route/protocol.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { GoogleVertexShared } from "./google-vertex-shared.js"
 
 export type AnthropicOptionsInput = AnthropicMessages.OptionsInput
 export type AnthropicProviderOptionsInput = AnthropicMessages.ProviderOptionsInput
@@ -15,8 +15,7 @@ export type AnthropicThinkingInput = AnthropicMessages.ThinkingInput
 
 const VERSION = "vertex-2023-10-16" as const
 
-// models.dev uses this provider id even though the API contract is Anthropic Messages.
-export const id = ProviderID.make("google-vertex-anthropic")
+export const id = ProviderID.make("google-vertex")
 
 export type Config = RouteDefaultsInput &
   GoogleVertexShared.OAuthOptions & {

packages/ai/src/providers/google-vertex-responses.ts

@@ -1,9 +1,9 @@
-import type { ProviderPackage } from "../provider-package"
-import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
-import { GoogleVertexShared } from "./google-vertex-shared"
-import type { OpenResponsesProviderOptionsInput } from "./open-responses-options"
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { GoogleVertexShared } from "./google-vertex-shared.js"
+import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
 
 export const id = ProviderID.make("google-vertex")
 

packages/ai/src/providers/google-vertex-shared.ts

@@ -1,6 +1,6 @@
 import type { AnyAuthClient } from "google-auth-library"
 import { Effect, Redacted } from "effect"
-import { Auth, MissingCredentialError } from "../route/auth"
+import { Auth, MissingCredentialError } from "../route/auth.js"
 
 const SCOPE = "https://www.googleapis.com/auth/cloud-platform"
 
@@ -74,4 +74,4 @@ export const oauth = (input: OAuthOptions, project?: string) => {
   return adc(project)
 }
 
-export * as GoogleVertexShared from "./google-vertex-shared"
+export * as GoogleVertexShared from "./google-vertex-shared.js"

packages/ai/src/providers/google-vertex.ts

@@ -1,11 +1,11 @@
-import type { ProviderPackage } from "../provider-package"
-import { Gemini } from "../protocols/gemini"
-import { Auth } from "../route/auth"
-import { Route, type RouteDefaultsInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { ProviderID, type ModelID } from "../schema"
-import { GoogleVertexShared } from "./google-vertex-shared"
+import type { ProviderPackage } from "../provider-package.js"
+import { Gemini } from "../protocols/gemini.js"
+import { Auth } from "../route/auth.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { GoogleVertexShared } from "./google-vertex-shared.js"
 
 export type GeminiOptionsInput = Gemini.OptionsInput
 export type GeminiProviderOptionsInput = Gemini.ProviderOptionsInput

packages/ai/src/providers/google-vertex/chat.ts

@@ -1,2 +1,2 @@
-export { model } from "../google-vertex-chat"
-export type { Settings } from "../google-vertex-chat"
+export { model } from "../google-vertex-chat.js"
+export type { Settings } from "../google-vertex-chat.js"

packages/ai/src/providers/google-vertex/gemini.ts

@@ -1,2 +1,2 @@
-export { model } from "../google-vertex"
-export type { Settings } from "../google-vertex"
+export { model } from "../google-vertex.js"
+export type { Settings } from "../google-vertex.js"

packages/ai/src/providers/google-vertex/messages.ts

@@ -1,2 +1,2 @@
-export { model } from "../google-vertex-messages"
-export type { Settings } from "../google-vertex-messages"
+export { model } from "../google-vertex-messages.js"
+export type { Settings } from "../google-vertex-messages.js"

packages/ai/src/providers/google-vertex/responses.ts

@@ -1,2 +1,2 @@
-export { model } from "../google-vertex-responses"
-export type { Settings } from "../google-vertex-responses"
+export { model } from "../google-vertex-responses.js"
+export type { Settings } from "../google-vertex-responses.js"

packages/ai/src/providers/google.ts

@@ -1,12 +1,12 @@
-import type { RouteDefaultsInput } from "../route/client"
-import { Auth } from "../route/auth"
-import type { ProviderAuthOption } from "../route/auth-options"
-import type { ProviderPackage } from "../provider-package"
-import { HttpOptions, ProviderID, mergeHttpOptions, type ModelID } from "../schema"
-import { Gemini } from "../protocols/gemini"
-import { GoogleImages } from "../protocols/google-images"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { Auth } from "../route/auth.js"
+import type { ProviderAuthOption } from "../route/auth-options.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { HttpOptions, ProviderID, mergeHttpOptions, type ModelID } from "../schema/index.js"
+import { Gemini } from "../protocols/gemini.js"
+import { GoogleImages } from "../protocols/google-images.js"
 
-export type { GoogleImageOptions } from "../protocols/google-images"
+export type { GoogleImageOptions } from "../protocols/google-images.js"
 export type GeminiOptionsInput = Gemini.OptionsInput
 export type GeminiProviderOptionsInput = Gemini.ProviderOptionsInput
 

packages/ai/src/providers/index.ts

@@ -1,19 +1,18 @@
-export * as Anthropic from "./anthropic"
-export * as AnthropicCompatible from "./anthropic-compatible"
-export * as AmazonBedrock from "./amazon-bedrock"
-export * as AmazonBedrockMantle from "./amazon-bedrock-mantle"
-export * as Azure from "./azure"
-export * as Cloudflare from "./cloudflare"
-export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare"
-export * as GitHubCopilot from "./github-copilot"
-export * as Google from "./google"
-export * as GoogleVertex from "./google-vertex"
-export * as GoogleVertexChat from "./google-vertex-chat"
-export * as GoogleVertexMessages from "./google-vertex-messages"
-export * as GoogleVertexResponses from "./google-vertex-responses"
-export * as OpenAI from "./openai"
-export * as OpenAICompatible from "./openai-compatible"
-export * as OpenAICompatibleResponses from "./openai-compatible-responses"
-export * as OpenRouter from "./openrouter"
-export * as XAI from "./xai"
-export * as ZAI from "./zai"
+export * as Anthropic from "./anthropic.js"
+export * as AnthropicCompatible from "./anthropic-compatible.js"
+export * as AmazonBedrock from "./amazon-bedrock.js"
+export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
+export * as Azure from "./azure.js"
+export * as Cloudflare from "./cloudflare.js"
+export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
+export * as Google from "./google.js"
+export * as GoogleVertex from "./google-vertex.js"
+export * as GoogleVertexChat from "./google-vertex-chat.js"
+export * as GoogleVertexMessages from "./google-vertex-messages.js"
+export * as GoogleVertexResponses from "./google-vertex-responses.js"
+export * as OpenAI from "./openai.js"
+export * as OpenAICompatible from "./openai-compatible.js"
+export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
+export * as OpenRouter from "./openrouter.js"
+export * as XAI from "./xai.js"
+export * as ZAI from "./zai.js"

packages/ai/src/providers/open-responses-options.ts

@@ -1,11 +1,10 @@
-import type { ResponseIncludable, ServiceTier } from "../protocols/utils/open-responses-options"
-import type { ProviderOptions, ReasoningEffort, TextVerbosity } from "../schema"
+import type { ResponseIncludable, ServiceTier } from "../protocols/utils/open-responses-options.js"
+import type { ProviderOptions, ReasoningEffort, TextVerbosity } from "../schema/index.js"
 
 export interface OpenResponsesOptionsInput {
   readonly [key: string]: unknown
   readonly instructions?: string
   readonly store?: boolean
-  readonly promptCacheKey?: string
   readonly reasoningEffort?: ReasoningEffort
   readonly reasoningSummary?: "auto" | "concise" | "detailed"
   readonly include?: ReadonlyArray<ResponseIncludable>
@@ -17,4 +16,4 @@ export type OpenResponsesProviderOptionsInput = ProviderOptions & {
   readonly openresponses?: OpenResponsesOptionsInput
 }
 
-export * as OpenResponsesProviderOptions from "./open-responses-options"
+export * as OpenResponsesProviderOptions from "./open-responses-options.js"

packages/ai/src/providers/openai-compatible-responses.ts

@@ -1,11 +1,11 @@
-import type { ProviderPackage } from "../provider-package"
-import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import type { RouteDefaultsInput } from "../route/client"
-import { ProviderID, type ModelID } from "../schema"
-import type { OpenResponsesProviderOptionsInput } from "./open-responses-options"
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
 
-export type { OpenResponsesOptionsInput, OpenResponsesProviderOptionsInput } from "./open-responses-options"
+export type { OpenResponsesOptionsInput, OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
 
 export const id = ProviderID.make("openai-compatible")
 

packages/ai/src/providers/openai-compatible.ts

@@ -1,10 +1,10 @@
-import { ProviderID, type ModelID } from "../schema"
-import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat"
-import type { RouteDefaultsInput } from "../route/client"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import type { ProviderPackage } from "../provider-package"
-import { profiles, type OpenAICompatibleProfile } from "./openai-compatible-profile"
-import type { OpenAIProviderOptionsInput } from "./openai-options"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { profiles, type OpenAICompatibleProfile } from "./openai-compatible-profile.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
 
 export const id = ProviderID.make("openai-compatible")
 

packages/ai/src/providers/openai-compatible/responses.ts

@@ -1 +1 @@
-export * from "../openai-compatible-responses"
+export * from "../openai-compatible-responses.js"

packages/ai/src/providers/openai-options.ts

@@ -1,8 +1,8 @@
-import type { ProviderOptions } from "../schema"
-import { mergeProviderOptions } from "../schema"
-import type { OpenResponsesOptionsInput } from "./open-responses-options"
+import type { ProviderOptions } from "../schema/index.js"
+import { mergeProviderOptions } from "../schema/index.js"
+import type { OpenResponsesOptionsInput } from "./open-responses-options.js"
 
-export type { OpenAIResponseIncludable, OpenAIServiceTier } from "../protocols/utils/openai-options"
+export type { OpenAIResponseIncludable, OpenAIServiceTier } from "../protocols/utils/openai-options.js"
 
 export type OpenAIOptionsInput = OpenResponsesOptionsInput
 
@@ -17,7 +17,6 @@ const openAIProviderOptions = (options: OpenAIOptionsInput | undefined): Provide
   const openai = Object.fromEntries(
     definedEntries({
       store: options?.store,
-      promptCacheKey: options?.promptCacheKey,
       reasoningEffort: options?.reasoningEffort,
       reasoningSummary: options?.reasoningSummary,
       include: options?.include,
@@ -68,4 +67,4 @@ export const withOpenAIOptions = <Options extends { readonly providerOptions?: O
   }
 }
 
-export * as OpenAIProviderOptions from "./openai-options"
+export * as OpenAIProviderOptions from "./openai-options.js"

packages/ai/src/providers/openai.ts

@@ -1,18 +1,18 @@
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import type { Route, RouteDefaultsInput } from "../route/client"
-import type { ProviderPackage } from "../provider-package"
-import { HttpOptions, ProviderID, ToolDefinition, mergeHttpOptions, type ModelID } from "../schema"
-import * as OpenAIChat from "../protocols/openai-chat"
-import * as OpenAIResponses from "../protocols/openai-responses"
-import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options"
-import { OpenAIImages, type OpenAIImageString } from "../protocols/openai-images"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { Route, RouteDefaultsInput } from "../route/client.js"
+import type { ProviderPackage } from "../provider-package.js"
+import { HttpOptions, ProviderID, ToolDefinition, mergeHttpOptions, type ModelID } from "../schema/index.js"
+import * as OpenAIChat from "../protocols/openai-chat.js"
+import * as OpenAIResponses from "../protocols/openai-responses.js"
+import { withOpenAIOptions, type OpenAIProviderOptionsInput } from "./openai-options.js"
+import { OpenAIImages, type OpenAIImageString } from "../protocols/openai-images.js"
 
-export type { OpenAIOptionsInput, OpenAIResponseIncludable } from "./openai-options"
-export type { OpenAIImageOptions } from "../protocols/openai-images"
+export type { OpenAIOptionsInput, OpenAIResponseIncludable } from "./openai-options.js"
+export type { OpenAIImageOptions } from "../protocols/openai-images.js"
 
 export const id = ProviderID.make("openai")
 
-export const routes = [OpenAIResponses.route, OpenAIResponses.webSocketRoute, OpenAIChat.route]
+export const routes = [OpenAIResponses.route, OpenAIChat.route]
 
 // This provider facade wraps the lower-level Responses and Chat model factories
 // with OpenAI-specific conveniences: typed options, API-key sugar, env fallback,
@@ -63,7 +63,6 @@ export interface Settings extends ProviderPackage.Settings {
   readonly organization?: string
   readonly project?: string
   readonly queryParams?: Readonly<Record<string, string>>
-  readonly transport?: "http" | "websocket"
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
@@ -82,17 +81,12 @@ const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Co
 
 export const configure = (input: Config = {}) => {
   const responsesRoute = configuredRoute(OpenAIResponses.route, input)
-  const responsesWebSocketRoute = configuredRoute(OpenAIResponses.webSocketRoute, input)
   const chatRoute = configuredRoute(OpenAIChat.route, input)
   const modelDefaults = defaults(input)
   const responses = (id: string | ModelID) =>
     responsesRoute
       .with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true }))
       .model<OpenAIProviderOptionsInput>({ id })
-  const responsesWebSocket = (id: string | ModelID) =>
-    responsesWebSocketRoute
-      .with(withOpenAIOptions(id, modelDefaults, { textVerbosity: true }))
-      .model<OpenAIProviderOptionsInput>({ id })
   const chat = (id: string | ModelID) =>
     chatRoute.with(withOpenAIOptions(id, modelDefaults)).model<OpenAIProviderOptionsInput>({ id })
   const image = (modelID: string | ModelID) =>
@@ -111,7 +105,6 @@ export const configure = (input: Config = {}) => {
     id,
     model: responses,
     responses,
-    responsesWebSocket,
     chat,
     image,
     configure,
@@ -138,10 +131,7 @@ const config = (settings: Settings): Config => {
 }
 
 export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) => {
-  const configured = configure(config(settings))
-  if (settings.transport === undefined || settings.transport === "http") return configured.responses(modelID)
-  if (settings.transport === "websocket") return configured.responsesWebSocket(modelID)
-  throw new Error(`Unsupported OpenAI Responses transport: ${String(settings.transport)}`)
+  return configure(config(settings)).responses(modelID)
 }
 
 export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (
@@ -149,6 +139,5 @@ export const chatModel: ProviderPackage.Definition<Settings, OpenAIProviderOptio
   settings,
 ) => configure(config(settings)).chat(modelID)
 export const responses = provider.responses
-export const responsesWebSocket = provider.responsesWebSocket
 export const chat = provider.chat
 export const image = provider.image

packages/ai/src/providers/openai/chat.ts

@@ -1,2 +1,2 @@
-export { chatModel as model } from "../openai"
-export type { Settings } from "../openai"
+export { chatModel as model } from "../openai.js"
+export type { Settings } from "../openai.js"

packages/ai/src/providers/openai/responses.ts

@@ -1,2 +1,2 @@
-export { model } from "../openai"
-export type { Settings } from "../openai"
+export { model } from "../openai.js"
+export type { Settings } from "../openai.js"

packages/ai/src/providers/openrouter.ts

@@ -1,15 +1,15 @@
 import { Effect, Schema } from "effect"
-import { Route, type RouteDefaultsInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { Protocol } from "../route/protocol"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import { ProviderID, type CacheHint, type ModelID, type ProviderOptions } from "../schema"
-import type { ProviderPackage } from "../provider-package"
-import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
-import * as OpenAIChat from "../protocols/openai-chat"
-import { newBreakpoints, ttlBucket } from "../protocols/utils/cache"
-import { isRecord } from "../protocols/shared"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { Framing } from "../route/framing.js"
+import { Protocol } from "../route/protocol.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import { ProviderID, type CacheHint, type ModelID, type ProviderOptions } from "../schema/index.js"
+import type { ProviderPackage } from "../provider-package.js"
+import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
+import * as OpenAIChat from "../protocols/openai-chat.js"
+import { newBreakpoints, ttlBucket } from "../protocols/utils/cache.js"
+import { isRecord } from "../protocols/shared.js"
 
 export const profile = OpenAICompatibleProfiles.profiles.openrouter
 export const id = ProviderID.make(profile.provider)
@@ -55,7 +55,6 @@ export interface OpenRouterOptions {
   readonly debug?: Readonly<{ echo_upstream_body?: boolean }>
   readonly models?: ReadonlyArray<string>
   readonly plugins?: ReadonlyArray<OpenRouterPlugin>
-  readonly promptCacheKey?: string
   readonly provider?: OpenRouterProviderRouting
   readonly reasoning?: Readonly<{
     enabled?: boolean
@@ -122,6 +121,7 @@ export const protocol = Protocol.make({
             ...body,
             messages,
             ...bodyOptions(request.providerOptions?.openrouter),
+            ...(request.promptCacheKey ? { prompt_cache_key: request.promptCacheKey } : {}),
           } as OpenRouterBody
         }),
       ),
@@ -161,7 +161,6 @@ const bodyOptions = (input: unknown) => {
     ...(isRecord(debug) ? { debug } : {}),
     ...(typeof user === "string" ? { user } : {}),
     ...(isRecord(reasoning) ? { reasoning } : {}),
-    ...(typeof promptCacheKey === "string" ? { prompt_cache_key: promptCacheKey } : {}),
   }
 }
 

packages/ai/src/providers/xai.ts

@@ -1,14 +1,14 @@
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import { Route, type RouteDefaultsInput } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { HttpOptions, ProviderID, type ModelID, type ProviderOptions } from "../schema"
-import * as OpenAICompatibleProfiles from "./openai-compatible-profile"
-import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat"
-import * as OpenAIChat from "../protocols/openai-chat"
-import * as OpenAIResponses from "../protocols/openai-responses"
-import { XAIImages } from "../protocols/xai-images"
-import type { OpenAIOptionsInput } from "./openai-options"
-import type { ProviderPackage } from "../provider-package"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
+import { Endpoint } from "../route/endpoint.js"
+import { HttpOptions, ProviderID, type ModelID, type ProviderOptions } from "../schema/index.js"
+import * as OpenAICompatibleProfiles from "./openai-compatible-profile.js"
+import * as OpenAICompatibleChat from "../protocols/openai-compatible-chat.js"
+import * as OpenAIChat from "../protocols/openai-chat.js"
+import * as OpenAIResponses from "../protocols/openai-responses.js"
+import { XAIImages } from "../protocols/xai-images.js"
+import type { OpenAIOptionsInput } from "./openai-options.js"
+import type { ProviderPackage } from "../provider-package.js"
 
 export const id = ProviderID.make("xai")
 
@@ -28,7 +28,7 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: XAIProviderOptionsInput
 }
 
-export type { XAIImageOptions } from "../protocols/xai-images"
+export type { XAIImageOptions } from "../protocols/xai-images.js"
 
 const responsesRoute = Route.make({
   id: "openai-responses",
@@ -47,6 +47,8 @@ const chatRoute = Route.make({
   protocol: OpenAIChat.protocol,
   endpoint: Endpoint.path("/chat/completions", { baseURL: OpenAICompatibleProfiles.profiles.xai.baseURL }),
   transport: OpenAICompatibleChat.route.transport,
+  headers: ({ request }): Record<string, string> =>
+    request.promptCacheKey ? { "x-grok-conv-id": request.promptCacheKey } : {},
 })
 
 export const routes = [responsesRoute, chatRoute]

packages/ai/src/providers/zai.ts

@@ -1,6 +1,6 @@
-import { ZAIImages } from "../protocols/zai-images"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options"
-import { HttpOptions, ProviderID, type ModelID } from "../schema"
+import { ZAIImages } from "../protocols/zai-images.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import { HttpOptions, ProviderID, type ModelID } from "../schema/index.js"
 
 export const id = ProviderID.make("zai")
 
@@ -10,7 +10,7 @@ export type Config = ProviderAuthOption<"optional"> & {
   readonly http?: HttpOptions.Input
 }
 
-export type { ZAIImageOptions } from "../protocols/zai-images"
+export type { ZAIImageOptions } from "../protocols/zai-images.js"
 
 const auth = (options: ProviderAuthOption<"optional">) => AuthOptions.bearer(options, "ZAI_API_KEY")
 

packages/ai/src/route.ts

@@ -1 +1 @@
-export * from "./route/index"
+export * from "./route/index.js"