Changelog

​

langgraph v1.2

• DeltaChannel (beta): A new channel type that stores only the incremental delta at each step rather than re-serializing the full accumulated value. Most useful for channels that grow large over time, for example a message list in a long-running thread. Use snapshot_frequency=K to write a full snapshot every K steps and bound read latency.
• Per-node timeouts: Pass timeout= to add_node to cap how long a single attempt may run. Set a hard wall-clock limit (run_timeout), an idle limit that resets on progress (idle_timeout), or both via TimeoutPolicy. When the limit fires, LangGraph raises NodeTimeoutError, clears writes from that attempt, and hands off to the retry policy. Async nodes only.
• Node-level error handlers: Pass error_handler= to add_node to run a recovery function after all retries are exhausted. The handler receives a typed NodeError and can return a Command to update state and route to a different node, useful for Saga/compensation patterns.
• Graceful shutdown: Stop an in-flight run cooperatively after the current superstep completes, and save a resumable checkpoint. Create a RunControl and call request_drain() from any thread; the run raises GraphDrained and can be resumed later with the same config.
• New event streaming API (beta): Pass version="v3" to stream_events() / astream_events() for a content-block-centric protocol with typed, per-channel projections (run.values, run.messages, run.lifecycle, run.subgraphs) plus opt-in transformers for updates, custom events, checkpoints, tasks, and debug. run.messages yields one ChatModelStream per LLM call with typed sub-projections for text, reasoning, tool calls, and usage. version="v1" and version="v2" are unchanged.

​

deepagents v0.5.0

• Async subagents: Deep Agents can launch non-blocking background tasks, so users can continue interacting with the agent while subagents work concurrently. Requires LangSmith Deployment for sub-agents.
• Multi-modal support: The read_file tool now supports PDFs, audio, and video files in addition to images.
• Backend changes: We’ve made backward-compatible changes to the Deep Agents backend protocol:
  • Updated the file format stored in State and Store backends to support binary files.
  • Improved error propagation from backends to tools.
  • You can now instantiate StateBackend() and StoreBackend() directly. Specifying with a factory (e.g., backend=(lambda rt: StateBackend(rt))) is deprecated.
• Anthropic prompt caching improvements: We’ve made some improvements to improve prompt caching performance for Anthropic models.

​

langgraph v1.1

• Type-safe streaming (version="v2"): Pass version="v2" to stream() / astream() for unified StreamPart output with type, ns, and data keys on every chunk. Each mode has its own TypedDict, all importable from langgraph.types. See streaming docs.
• Type-safe invoke (version="v2"): Pass version="v2" to invoke() / ainvoke() to get a GraphOutput object with .value and .interrupts attributes. See invoke docs.
• Pydantic and dataclass coercion: With version="v2", invoke() and values-mode stream output are automatically coerced to your declared Pydantic model or dataclass type.
• Fixed time travel with interrupts and subgraphs: Replays no longer reuse stale RESUME values, and subgraphs correctly restore the checkpoint for the parent’s historical state.
• Fully backwards compatible: version="v2" is opt-in. GraphOutput supports deprecated dict-style access for gradual migration.

​

deepagents v0.4

• New integration packages for pluggable sandboxes: langchain-modal, langchain-daytona, and langchain-runloop. See sandboxes guide and example data analysis tutorial.
• Changes to conversation history summarization:
  • Summarization now happens in the model node via wrap_model_call events. Due to this we retain the full message history in the graph state.
  • More accurate token counting.
  • Summarization will now automatically trigger if a chat model raises a ContextOverflowError (defined in langchain-core). Currently langchain-anthropic and langchain-openai support this.
• We now default to the Responses API for model strings prefixed with "openai:".
  Disable data retention with the Responses API

  from langchain.chat_models import init_chat_model
  
  agent = create_deep_agent(
      model=init_chat_model(
          "openai:...",
          use_responses_api=True,
          store=False,
          include=["reasoning.encrypted_content"],
      )
  )
  

​

langchain v1.2.0

• create_agent: Simplified support for provider-specific tool parameters and definitions via a new extras attribute on tools. Examples:
  • Provider-specific configuration such as Anthropic’s programmatic tool calling and tool search.
  • Built-in tools that are executed client-side, as supported by Anthropic, OpenAI, and other providers.
• Support for strict schema-adherence in agent response_format (see ProviderStrategy docs).

​

langchain-google-genai v4.0.0

​

langchain v1.1.0

• Model profiles: Chat models now expose supported features and capabilities through a .profile attribute. These data are derived from models.dev, an open source project providing model capability data.
• Summarization middleware: Updated to support flexible trigger points using model profiles for context-aware summarization.
• Structured output: ProviderStrategy support (native structured output) can now be inferred from model profiles.
• SystemMessage for create_agent: Support for passing SystemMessage instances directly to create_agent’s system_prompt parameter, enabling advanced features like cache control and structured content blocks.
• Model retry middleware: New middleware for automatically retrying failed model calls with configurable exponential backoff.
• Content moderation middleware: OpenAI content moderation middleware for detecting and handling unsafe content in agent interactions. Supports checking user input, model output, and tool results.

​

v1.0.0

​

langchain

• Release notes
• Migration guide

​

langgraph

• Release notes
• Migration guide