[PR #5887] feat(prebuilt): 1.0 implementation updates #4700

Closed
opened 2026-02-20 17:50:36 -05:00 by yindo · 0 comments
Owner

Original Pull Request: https://github.com/langchain-ai/langgraph/pull/5887

State: closed
Merged: No


deferred to next agent iteration, @sydney-runkle small tasks

  • New ToolRuntime pattern for tools
  • prebuilts for pre and post model hook nodes (forcing a stop after X tool calls, a certain tool, etc)
  • Add prepare_call API, removing dynamic model pattern currently available (@sydney-runkle)
  • Maybe remove interrupt stuff that's specific to the agent inbox?
  • Drop Python 3.9 syntax (https://github.com/langchain-ai/langgraph/pull/5983)
  • Simplify tool output / native output bindings in the interest of a sleek private API
  • Get rid of remaining steps in favor of model call count not stored in state (@nfcampos approved)
  • don't think we need to require tools?
  • Clean up prompt signature and handling

Considered:

Internal:

  • Clean up injection logic -- should just support state, config, and context (no more store explicitly, this should be deprecated)
**Original Pull Request:** https://github.com/langchain-ai/langgraph/pull/5887 **State:** closed **Merged:** No --- - [x] BREAKING: disabled support for binding tools and other things to a chat model, simplifying the signature of `create_agent`'s model param (https://github.com/langchain-ai/langgraph/pull/5912 and https://github.com/langchain-ai/langgraph/pull/5958) - [x] BREAKING: removing structured output generation node - [x] BREAKING: rename `agent` node to `model`: https://github.com/langchain-ai/langgraph/pull/5971 - [x] Remove `version` arg, no more "v1" logic (https://github.com/langchain-ai/langgraph/pull/5960) - [x] Remove config schema deprecation + deprecated kwargs stuff: https://github.com/langchain-ai/langgraph/pull/5970 - [x] Simplify prompt logic, done to make to provide prepared messages via `prepare_call` api: https://github.com/langchain-ai/langgraph/pull/5909 - [x] Refactor internal implementation of react agent to use `_AgentBuilder` class - [x] Remove pydantic state + consolidate AgentState (w/ structured response): https://github.com/langchain-ai/langgraph/pull/5973 - [x] Add strategies for "optimized" structured response to react agent (https://github.com/langchain-ai/langgraph/pull/5961) - [x] ToolOutput (basic) https://github.com/langchain-ai/langgraph/pull/5887/commits/b58a7fb2fe85abd6fd6877fa94033a0ee754bedc - [x] ToolOutput parameterize w/ different techniques to change history - [x] NativeOutput - [x] figure out retry policy / error handling policy for std outputs: https://github.com/langchain-ai/langgraph/pull/6002 - [x] update default error handling for tool errors: https://github.com/langchain-ai/langgraph/pull/6002 deferred to next agent iteration, @sydney-runkle small tasks - [ ] New `ToolRuntime` pattern for tools - [ ] prebuilts for pre and post model hook nodes (forcing a stop after X tool calls, a certain tool, etc) - [ ] Add `prepare_call` API, removing dynamic model pattern currently available (@sydney-runkle) - [ ] Maybe remove interrupt stuff that's specific to the agent inbox? - [ ] Drop Python 3.9 syntax (https://github.com/langchain-ai/langgraph/pull/5983) - [ ] Simplify tool output / native output bindings in the interest of a sleek private API - [ ] Get rid of remaining steps in favor of model call count not stored in state (@nfcampos approved) - [ ] don't think we need to require tools? - [ ] Clean up prompt signature and handling Considered: * Allowing support for binding other things, but vetoed in the interest of a simple API * Split tool nodes into individual nodes (https://github.com/langchain-ai/langgraph/pull/5888/files) Internal: - Clean up injection logic -- should just support state, config, and context (no more store explicitly, this should be deprecated)
yindo added the pull-request label 2026-02-20 17:50:36 -05:00
yindo closed this issue 2026-02-20 17:50:36 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langchain-ai/langgraph#4700