This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated. # Releases ## @langchain/langgraph-checkpoint@1.0.3 ### Patch Changes - [#2352](https://github.com/langchain-ai/langgraphjs/pull/2352) [`14f2a79`](https://github.com/langchain-ai/langgraphjs/commit/14f2a796912e81d7f52f0a4f16747f6d0a269209) Thanks [@Nagendhra-web](https://github.com/Nagendhra-web)! - fix(langgraph-checkpoint): block prototype pollution in MemorySaver via reserved storage keys `MemorySaver` previously embedded `thread_id`, `checkpoint_ns`, `checkpoint_id`, and `task_id` directly into property accesses on the nested plain objects `this.storage` and `this.writes`. A caller able to shape any of those fields (every quickstart, tutorial, and test fixture uses `MemorySaver` by default) could pass `"__proto__"`, `"constructor"`, or `"prototype"` and have the subsequent assignment mutate `Object.prototype`. From that point every plain object in the process inherits the injected property, breaking `for...in` loops, truthy short-circuits, and downstream serializers across unrelated code paths. CWE-1321. Adds an `assertSafeStorageKey` chokepoint applied at every public entry that touches `storage` or `writes` (`put`, `putWrites`, `deleteThread`, `getTuple`, `list`). The guard rejects non-string values, the empty string (unless explicitly opted-in for `checkpoint_ns`), and the three prototype-pollution keys. Behaviour for valid string identifiers is unchanged. ## @langchain/langgraph-checkpoint-redis@1.0.6 ### Patch Changes - [#2350](https://github.com/langchain-ai/langgraphjs/pull/2350) [`1e73c6b`](https://github.com/langchain-ai/langgraphjs/commit/1e73c6b4630bbc4aa976eea4bfc33c4f753b7ee9) Thanks [@Nagendhra-web](https://github.com/Nagendhra-web)! - fix(checkpoint-redis): block Redis KEYS / SCAN pattern injection via top-level identifiers `RedisSaver` and `ShallowRedisSaver` previously embedded `thread_id`, `checkpoint_ns`, `checkpoint_id`, and `task_id` directly into Redis keys and `client.keys(pattern)` calls with no validation. A caller able to shape any of those fields (multi-tenant SDK deployments where the `RunnableConfig` originates from request input, or webhook payloads that flow into a persisted thread) could promote a string identifier into a glob pattern (`*`, `?`, `[...]`) or escape character (`\`). The most severe sink is `deleteThread`: a `threadId` of `*` issues `client.keys("checkpoint:*:*")` followed by `client.del(...)`, deleting every checkpoint in the database across every tenant. `getTuple`, `list`, and `loadPendingWrites` are exposed to the same pattern via the fallback paths that bypass the existing `escapeRediSearchTagValue` defense. Adds a single `assertSafeKeyComponent` helper exported from `./utils.js` and applies it at every key-building site. The guard asserts the value is a non-empty string (the empty `checkpoint_ns` default is opt-in via `{ allowEmpty: true }`) and rejects the Redis pattern meta-characters `* ? [ ] \`. The `:` delimiter is intentionally permitted because LangGraph emits it as a legitimate part of `checkpoint_ns` for subgraphs / nested graphs, where it only ever appears as a literal in the key. Behavior for valid string identifiers is unchanged. ## @langchain/langgraph-api@1.2.3 ### Patch Changes - [#2447](https://github.com/langchain-ai/langgraphjs/pull/2447) [`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: fold forkFrom client-side and honor per-run multitaskStrategy The SDK now folds the ergonomic `forkFrom` option into `config.configurable.checkpoint_id` before sending `run.start`, so the agent server only ever accepts the single, legacy-compliant fork field (`forkFrom` no longer hits the wire). The protocol-v2 reference servers drop their top-level `forkFrom` normalization accordingly. The protocol-v2 servers now honor the caller's `multitaskStrategy` per run (one of `reject` \| `rollback` \| `interrupt` \| `enqueue`) instead of hardcoding it, falling back to `enqueue` when omitted or unrecognized. - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. - Updated dependencies \[[`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532)]: - @langchain/langgraph-ui@1.2.3 ## @langchain/langgraph-cli@1.2.3 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-api@1.2.3 ## @langchain/langgraph-ui@1.2.3 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. ## @langchain/langgraph-sdk@1.9.10 ### Patch Changes - [#2447](https://github.com/langchain-ai/langgraphjs/pull/2447) [`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: fold forkFrom client-side and honor per-run multitaskStrategy The SDK now folds the ergonomic `forkFrom` option into `config.configurable.checkpoint_id` before sending `run.start`, so the agent server only ever accepts the single, legacy-compliant fork field (`forkFrom` no longer hits the wire). The protocol-v2 reference servers drop their top-level `forkFrom` normalization accordingly. The protocol-v2 servers now honor the caller's `multitaskStrategy` per run (one of `reject` \| `rollback` \| `interrupt` \| `enqueue`) instead of hardcoding it, falling back to `enqueue` when omitted or unrecognized. - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. ## @langchain/angular@1.0.10 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 ## @langchain/react@1.0.10 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 ## @langchain/svelte@1.0.10 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 ## @langchain/vue@1.0.10 ### Patch Changes - [#2443](https://github.com/langchain-ai/langgraphjs/pull/2443) [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532) Thanks [@christian-bromann](https://github.com/christian-bromann)! - refactor(sdk): drop StreamSubmitOptions.command and simplify forkFrom Remove the misleading submit({ command }) surface from protocol-v2 StreamController; HITL resume is respond() only. Accept forkFrom as a plain checkpoint id string and align protocol-v2 servers and docs. - [#2448](https://github.com/langchain-ai/langgraphjs/pull/2448) [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d) Thanks [@christian-bromann](https://github.com/christian-bromann)! - protocol-v2: add `respondAll()` and run config/metadata on interrupt resume The stream controller (and the React/Angular/Svelte/Vue wrappers) gain a `respondAll(responsesById, options)` method to resume several interrupts pending at the same checkpoint in a single command — required for runs that pause on multiple interrupts at once (e.g. parallel tool-authorization prompts), which sequential `respond()` calls cannot handle. `respond()` now takes an options object (`{ interruptId?, namespace?, config?, metadata? }`) so a resumed run can carry the same run-level config (model, user context, …) and metadata (trigger source, test flags, …) a fresh `submit()` would. The protocol-v2 reference servers read the new `responses` batch and `config` / `metadata` fields leniently and fold them onto the run that services the `input.respond` command. - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 ## @example/ai-elements@0.1.25 ### Patch Changes - Updated dependencies \[[`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/react@1.0.10 ## @examples/assistant-ui-claude@0.1.25 ### Patch Changes - Updated dependencies \[[`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/react@1.0.10 ## @examples/ui-angular@0.0.35 ### Patch Changes - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 - @langchain/angular@1.0.10 ## @examples/ui-multimodal@0.0.11 ### Patch Changes - Updated dependencies \[[`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/react@1.0.10 ## @examples/ui-react@0.0.11 ### Patch Changes - Updated dependencies \[[`80c2806`](https://github.com/langchain-ai/langgraphjs/commit/80c2806cb2da93745a640664bd0cf603c2361da9), [`80a8c12`](https://github.com/langchain-ai/langgraphjs/commit/80a8c1200a240fd984edc4deb26a7787d08c7532), [`2c14b12`](https://github.com/langchain-ai/langgraphjs/commit/2c14b12a80c306578563e77595943037c7c4844d)]: - @langchain/langgraph-sdk@1.9.10 - @langchain/react@1.0.10 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
@langchain/svelte
Svelte 5 SDK for Deep Agents, LangChain and LangGraph.
useStream binds a LangGraph agent into a Svelte 5 component. Reactive fields are exposed as getters on a stable handle (stream.messages, stream.isLoading, …) so templates and $derived expressions track updates automatically — no stores, no $ prefix, no destructuring.
Installation
npm install @langchain/svelte @langchain/core
Peer dependencies: svelte ^5.0.0, @langchain/core ^1.0.1
Quick start
<script lang="ts">
import { useStream } from "@langchain/svelte";
const stream = useStream({
assistantId: "agent",
apiUrl: "http://localhost:2024",
});
</script>
{#each stream.messages as msg (msg.id)}
<div>{msg.content}</div>
{/each}
<button
disabled={stream.isLoading}
onclick={() =>
stream.submit({ messages: [{ type: "human", content: "Hello!" }] })}
>
Send
</button>
Note: Access fields through the live
streamhandle. Destructuring (const { messages } = stream) freezes the values at that moment — usestream.messagesin templates instead.
Highlights
- v2-native streaming protocol. Session-based transport with automatic re-attach on remount; no more
reconnectOnMount/joinStreamdance. - Always-on root projections.
values,messages,toolCalls, andinterruptsare reactive at the root with zero extra subscription cost. - Selector composables for scoped data. Per-subagent / per-subgraph messages, tool calls, and media stream only when a component actually mounts the matching composable, and release on unmount.
- Discriminated option bag. The hosted Agent Server path and the custom-adapter path are two arms of a single typed union — mixing them is a compile-time error.
- Reactive
threadId. PassthreadId: () => activeto drive in-place thread swaps without remounting. - Agent-brand type inference.
useStream<typeof agent>()unwraps state, tool calls, and subagent state maps from an agent brand. - Multimodal media streams. Built-in assembly for audio, images, video, and files — plus opinionated playback helpers.
- Headless tools. Register local tool implementations that auto-resolve server-emitted tool-call interrupts without a round-trip through the UI.
Documentation
In-depth guides live in docs/:
useStream— options, return shape, reactivethreadId- Selector composables (
useMessages,useToolCalls,useValues, …) - Interrupts,
respond(),stop(),hydrationPromise - Submission queue
- Stream context (
provideStream/getStream) - Headless tools
- Custom transport (
AgentServerAdapter,HttpAgentServerAdapter) - Media (images, audio, video, files) & playback helpers
- Type safety
Migrating from v0
@langchain/svelte v1 targets the v2 streaming protocol. The useStream import stays the same, but the option bag, return shape, and how you subscribe to scoped data all change. Most chat apps migrate in well under an hour — the full guide with line-by-line diffs lives in docs/v1-migration.md.
Playground
For full end-to-end examples, see the LangChain UI Playground.
License
MIT