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@1.1.2 ### Patch Changes - [#2544](https://github.com/langchain-ai/langgraphjs/pull/2544) [`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): make concurrent DeltaChannel writes deterministic on replay Concurrent same-superstep writes to a `DeltaChannel` could reconstruct from a checkpoint differently than they were applied live, because live execution ordered them by task path while savers replayed them by task id. This fixes that divergence in two complementary ways: - Plain concurrent writes are now applied in the canonical `(task_id, idx)` order on both paths: `_applyWrites` orders them that way live, and the `getDeltaChannelHistory` walk enforces the same order so reconstruction matches live for every saver (Postgres, SQLite, MongoDB, Redis, and custom). - An `Overwrite` now wins its entire super-step: every sibling write in the same step — before AND after the `Overwrite` — is discarded, matching `BinaryOperatorAggregate`. This makes the result independent of the (unstable) ordering of concurrent fan-in writes; previously a plain write that landed after an `Overwrite` in the same step was still folded in. To keep reconstruction in sync with this `Overwrite` rule, any `DeltaChannel` that sees an `Overwrite` in a super-step is now force-snapshotted at the next checkpoint (and, under `"exit"` durability, in the final checkpoint). The post-overwrite value is materialized into `channel_values`, so a cold read seeds from that snapshot and never has to replay across the reset — making live and reconstructed state identical without changing the sparse-replay history shape. These delta-channel APIs remain Beta. - [#2531](https://github.com/langchain-ai/langgraphjs/pull/2531) [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): forward task metadata and name subagents via lc_agent_name `mapDebugTasks` now forwards filtered user-meaningful task config metadata (including `lc_agent_name`) onto `tasks` stream payloads. The lifecycle transformer uses that metadata to set subagent `graph_name` from `lc_agent_name` and recover `cause: { type: "toolCall", tool_call_id }` from parent tool-dispatch tasks. Adds the shared `EXCLUDED_METADATA_KEYS` constant to `@langchain/langgraph-checkpoint`. Ports langgraph#7928. ## @langchain/langgraph-checkpoint-redis@1.0.10 ### Patch Changes - [#2336](https://github.com/langchain-ai/langgraphjs/pull/2336) [`25907eb`](https://github.com/langchain-ai/langgraphjs/commit/25907eb0be25258c26327c6c68c72bc828ee1cff) Thanks [@MohMaherId](https://github.com/MohMaherId)! - fix(langgraph-checkpoint-redis): persist and reconstruct full `channel_values` across multi-node graphs. `RedisSaver.put()` delta-filters `channel_values` to only the channels written by the current node, but `getTuple()` had no reconstruction logic — unlike `PostgresSaver` — so any multi-node graph whose last node wrote a subset of channels silently lost the others. Each changed channel is now persisted as a version-keyed `checkpoint_blob:*` entry in `put()` and missing channels are reconstructed from those blobs on read. `deleteThread()` now also deletes the `checkpoint_blob:*` keys. Without this the blobs introduced above would orphan forever (memory growth) and thread deletion would be incomplete, matching `PostgresSaver.deleteThread()` parity. When `ttlConfig.refreshOnRead` is enabled, reads now refresh the TTL of the reconstructed `checkpoint_blob:*` keys alongside the checkpoint key. Otherwise a read would keep the checkpoint alive while the blobs it depends on expired, silently dropping reconstructed channels. On write, `put()` now refreshes the TTL of every blob the checkpoint references (the full `channel_versions` set), not just the channels changed by the current node, so carried-over blobs from earlier nodes expire in lockstep with the checkpoint doc. This write-side refresh is independent of `refreshOnRead`. The per-channel blob writes also now run in parallel. Known limitation: with TTL enabled, a carried-over blob can still be lost if it expires during an idle gap longer than `defaultTTL` (no read or write refreshed it in time). When that happens the channel is left cleanly absent on read rather than erroring. Fully closing this gap (re-persisting expired blobs) is tracked as a follow-up. ## @langchain/langgraph@1.4.3 ### Patch Changes - [#2544](https://github.com/langchain-ai/langgraphjs/pull/2544) [`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): make concurrent DeltaChannel writes deterministic on replay Concurrent same-superstep writes to a `DeltaChannel` could reconstruct from a checkpoint differently than they were applied live, because live execution ordered them by task path while savers replayed them by task id. This fixes that divergence in two complementary ways: - Plain concurrent writes are now applied in the canonical `(task_id, idx)` order on both paths: `_applyWrites` orders them that way live, and the `getDeltaChannelHistory` walk enforces the same order so reconstruction matches live for every saver (Postgres, SQLite, MongoDB, Redis, and custom). - An `Overwrite` now wins its entire super-step: every sibling write in the same step — before AND after the `Overwrite` — is discarded, matching `BinaryOperatorAggregate`. This makes the result independent of the (unstable) ordering of concurrent fan-in writes; previously a plain write that landed after an `Overwrite` in the same step was still folded in. To keep reconstruction in sync with this `Overwrite` rule, any `DeltaChannel` that sees an `Overwrite` in a super-step is now force-snapshotted at the next checkpoint (and, under `"exit"` durability, in the final checkpoint). The post-overwrite value is materialized into `channel_values`, so a cold read seeds from that snapshot and never has to replay across the reset — making live and reconstructed state identical without changing the sparse-replay history shape. These delta-channel APIs remain Beta. - [#2531](https://github.com/langchain-ai/langgraphjs/pull/2531) [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): merge instead of overwrite in `ensureLangGraphConfig` `ensureLangGraphConfig` now per-key merges `callbacks`, `tags`, `metadata`, and `configurable` across configs instead of last-write-wins, so values bound via `.withConfig({...})` survive when a later (e.g. invoke-time) config supplies other keys. The merged dicts are fresh objects, fixing a by-reference mutation of shared base configs. Also drops the now-redundant `combineCallbacks` workaround in `streamEvents`, which double-registered and double-fired graph-bound callbacks. - [#2531](https://github.com/langchain-ai/langgraphjs/pull/2531) [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): preserve namespace nesting for imperative graph invokes When a compiled graph is invoked from inside another graph's running task (e.g. a tool body calling `subAgent.invoke(...)`), the surrounding task context — including the langgraph-internal nesting keys (`__pregel_read`, `__pregel_stream`, `checkpoint_ns`, the checkpoint map) — is propagated implicitly via `AsyncLocalStorage`. The base `Runnable.stream` calls langchain-core's `ensureConfig`, which replaces the ambient `configurable` wholesale whenever the caller passes its own. Because `createAgent` always supplies a `configurable`, every tool-invoked sub-agent lost those keys, ran as a fresh root run, and had its streamed events flattened to the root namespace instead of nesting under the triggering task. `Pregel.stream` now merges the ambient `configurable` underneath the caller's (caller keys win per-key) when the ambient marks an active task (`__pregel_read` present) but the explicit `configurable` is missing it. Declared subgraph nodes (which already carry their own `__pregel_read`) and top-level runs are unaffected. - [#2537](https://github.com/langchain-ai/langgraphjs/pull/2537) [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): dispatch stream messages handler inline The v3 `messages` handler (`StreamProtocolMessagesHandler`, which powers `run.messages`) only performs a synchronous `push()` onto the run's stream, but its callbacks were dispatched on LangChain's background callback queue (the default `awaitHandlers === false`). A model or tool call inside a nested or parallel task could therefore flush its `messages` chunk _after_ the Pregel loop returned and sealed the stream, where `IterableReadableWritableStream.push` silently drops chunks once closed. This surfaced as empty per-message streams (`sub.messages`) for subagents dispatched in parallel from a single tools step. The handler now sets `awaitHandlers = true` so its callbacks run inline — every push happens during the originating model/chain call while the stream is still open. This avoids the global over-wait, fake-timer deadlock, and error-path unhandled rejections that a blanket `awaitAllCallbacks()` drain before close would have introduced. - [#2531](https://github.com/langchain-ai/langgraphjs/pull/2531) [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): forward task metadata and name subagents via lc_agent_name `mapDebugTasks` now forwards filtered user-meaningful task config metadata (including `lc_agent_name`) onto `tasks` stream payloads. The lifecycle transformer uses that metadata to set subagent `graph_name` from `lc_agent_name` and recover `cause: { type: "toolCall", tool_call_id }` from parent tool-dispatch tasks. Adds the shared `EXCLUDED_METADATA_KEYS` constant to `@langchain/langgraph-checkpoint`. Ports langgraph#7928. - [#2549](https://github.com/langchain-ai/langgraphjs/pull/2549) [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(langgraph): support DeltaChannel fields in StateSchema Add a `DeltaValue` state field (and a `MessagesDeltaValue` prebuilt) so a `DeltaChannel` can be declared via `StateSchema`, not just `Annotation.Root` or a raw channel map. `StateSchema` now maps `DeltaValue` to a `DeltaChannel` (forwarding `snapshotFrequency` and the value-schema default) and validates its inputs/`Overwrite` updates like `ReducedValue`. - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b)]: - @langchain/langgraph-checkpoint@1.1.2 - @langchain/langgraph-sdk@1.9.23 ## @langchain/langgraph-sdk@1.9.23 ### Patch Changes - [#2545](https://github.com/langchain-ai/langgraphjs/pull/2545) [`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236) Thanks [@christian-bromann](https://github.com/christian-bromann)! - fix(sdk): avoid scoped stream resubscribe churn Defer final projection disposal by one microtask so framework bindings that release and immediately reacquire the same scoped projection during reactive updates keep the existing stream subscription instead of rotating through root-only and scoped SSE filters. ## @langchain/angular@1.0.24 ### Patch Changes - Updated dependencies \[[`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236)]: - @langchain/langgraph-sdk@1.9.23 ## @langchain/react@1.0.24 ### Patch Changes - Updated dependencies \[[`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236)]: - @langchain/langgraph-sdk@1.9.23 ## @langchain/svelte@1.0.24 ### Patch Changes - Updated dependencies \[[`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236)]: - @langchain/langgraph-sdk@1.9.23 ## @langchain/vue@1.0.24 ### Patch Changes - Updated dependencies \[[`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236)]: - @langchain/langgraph-sdk@1.9.23 ## @example/ai-elements@0.1.39 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 - @langchain/react@1.0.24 ## @examples/assistant-ui-claude@0.1.39 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 - @langchain/react@1.0.24 ## @examples/ui-angular@0.0.49 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 - @langchain/langgraph-sdk@1.9.23 - @langchain/angular@1.0.24 ## @examples/ui-multimodal@0.0.25 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 - @langchain/react@1.0.24 ## @examples/ui-react@0.0.25 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`2134c8a`](https://github.com/langchain-ai/langgraphjs/commit/2134c8a2c0bc8dd2ebea33e1191c8dd0c4b83236), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 - @langchain/langgraph-sdk@1.9.23 - @langchain/react@1.0.24 ## langgraph@1.0.43 ### Patch Changes - Updated dependencies \[[`4487214`](https://github.com/langchain-ai/langgraphjs/commit/448721449f0801009ba76b03dd2e9c16f900bbba), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`be09666`](https://github.com/langchain-ai/langgraphjs/commit/be096663f42fe7ea9355d6c0def4854e657866d8), [`38cfe01`](https://github.com/langchain-ai/langgraphjs/commit/38cfe01ff02490ff6bcc86c66708ef671f2e0d4b), [`bc667a9`](https://github.com/langchain-ai/langgraphjs/commit/bc667a998ae9909d15795387dad45048e8947219)]: - @langchain/langgraph@1.4.3 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
@langchain/react
React SDK for building AI-powered applications with Deep Agents, LangChain and LangGraph.
@langchain/react v1 ships a v2-native useStream hook together with a small family of companion selector hooks. The root hook gives you always-on access to thread state, messages, tool calls, and interrupts; the selector hooks open ref-counted subscriptions for the things that aren't needed on every view (per-subagent messages, media streams, submission queue, message metadata, raw channels, …).
Highlights
- v2-native streaming protocol. Session-based transport with automatic re-attach on remount; no more
reconnectOnMount/joinStreamdance. - Selector-based subscriptions. Namespaced data (subagents, subgraphs, media) streams only when a component actually mounts the matching selector hook, and releases on unmount.
- Always-on root projections.
values,messages,toolCalls, andinterruptsare live at the root with zero per-subscription cost. - Agent-brand type inference.
useStream<typeof agent>()unwraps state, tool calls, and subagent state maps from an agent brand. - Discriminated options. 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.
- Multimodal media streams. Built-in assembly for audio, images, video, and files.
- Suspense integration.
useSuspenseStreamhands the initial hydration phase to<Suspense>and non-streaming errors to Error Boundaries.
Installation
npm install @langchain/react @langchain/core
Peer dependencies: react (^18 || ^19), @langchain/core (^1.1.27).
Quick Start
import { useStream } from "@langchain/react";
function Chat() {
const { messages, submit, isLoading } = useStream({
assistantId: "agent",
apiUrl: "http://localhost:2024",
});
return (
<div>
{messages.map((msg, i) => (
<div key={msg.id ?? i}>{String(msg.content)}</div>
))}
<button
disabled={isLoading}
onClick={() =>
void submit({
messages: [{ type: "human", content: "Hello!" }],
})
}
>
Send
</button>
</div>
);
}
Mental model
@langchain/react v1 splits the surface into two layers:
- Root hook (
useStream). Owns the thread lifecycle, the transport, and a handful of always-on projections (values,messages,toolCalls,interrupts,error,isLoading, discovery maps). Mount it once per thread. - Companion selector hooks. Each one opens a ref-counted subscription when the first component mounts it and releases it when the last consumer unmounts. Use them for anything scoped to a namespace, a subagent / subgraph, a specific message, a specific extension channel, or a media stream.
import {
useStream,
useMessages,
useToolCalls,
useSubmissionQueue,
} from "@langchain/react";
function Chat() {
const stream = useStream({ assistantId: "agent", apiUrl: "/api" });
// Root: free reads, no new subscription.
const messages = useMessages(stream); // same as stream.messages
// Scoped: opens a namespaced subscription on mount.
const queue = useSubmissionQueue(stream);
}
Documentation
Detailed guides live in ./docs. Start with the two files most apps need first:
useStream— options, return shape,submit(),stop(),respond(),hydrationPromise.- Companion selector hooks —
useValues,useMessages,useToolCalls,useMessageMetadata,useChannel,useExtension, and friends.
Feature-specific guides:
- Transports — SSE, WebSocket,
HttpAgentServerAdapter, customAgentServerAdapter. - Custom transports — implementing
AgentServerAdapteragainst your own backend, with a worked walkthrough ofexamples/ui-react-transport. - Interrupts & headless tools — pausing runs,
respond(),tools+onTool. - Fork / edit from a checkpoint —
useMessageMetadata+submit({ forkFrom }). - Submission queue —
multitaskStrategy: "enqueue"+useSubmissionQueue. - Subagents & subgraphs — discovery maps, scoped selector subscriptions.
- Multimodal media —
useAudio/useImages/useVideo/useFiles,useMediaURL, players. useSuspenseStream— Suspense + Error Boundary integration.StreamProvider/useStreamContext— share one stream across a subtree.- Type safety — agent-brand inference, prop-drilling, type helpers.
Migrating from v0 to v1
The useStream import name is unchanged, but the return shape, option bag, and protocol semantics all shifted. Most chat apps migrate in well under an hour — the full migration guide with line-by-line diffs lives in ./docs/v1-migration.md.
Legacy type aliases (UseStream, UseSuspenseStream, UseStreamOptions, UseStreamTransport, QueueEntry, GetToolCallsType, SubagentStream, …) and the legacy FetchStreamTransport class are no longer re-exported from @langchain/react. Apps still on the legacy surface can import directly from @langchain/langgraph-sdk/ui during their migration.
Playground
For complete end-to-end examples with full agentic UIs, visit the LangChain UI Playground.
License
MIT