feat(langchain): add deepagents middleware and prebuilt #33405
Conversation
…memory, and file search Middleware Classes Text Editor Tools - StateClaudeTextEditorToolMiddleware: In-memory text editor using agent state - FilesystemClaudeTextEditorToolMiddleware: Text editor operating on real filesystem Implementing Claude's text editor tools https://docs.claude.com/en/docs/agents-and-tools/tool-use/text-editor-tool Operations: view, create, str_replace, insert Memory Tools - StateClaudeMemoryToolMiddleware: Memory persistence in agent state - FilesystemClaudeMemoryToolMiddleware: Memory persistence on filesystem Implementing Claude's memory tools https://docs.claude.com/en/docs/agents-and-tools/tool-use/memory-tool Operations: Same as text editor plus delete and rename File Search Tools - StateFileSearchMiddleware: Search state-based files - FilesystemFileSearchMiddleware: Search real filesystem Provides Glob and Grep tools with same schema as used by Claude Code (but compatible with any model) - Glob: Pattern matching (e.g., **/*.py, src/**/*.ts), sorted by modification time - Grep: Regex content search with output modes (files_with_matches, content, count) Usage ``` from langchain.agents import create_agent from langchain.agents.middleware import ( StateTextEditorToolMiddleware, StateFileSearchMiddleware, ) agent = create_agent( model=model, tools=[], middleware=[ StateTextEditorToolMiddleware(), StateFileSearchMiddleware(), ], ) ```
github-actions
bot
added
langchain
…nto nh/subagent-middleware
libs/langchain_v1/tests/integration_tests/agents/middleware/test_subagent_middleware.py
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Deepagents are super exciting! Really looking forward to these new middlewares, especially the subagents one.
I have some concern about introducing the create_deep_agent factory here, imo it has a significantly lower ceiling than create_agent because under the hood, we decide on all of the middleware being used. That means we decide on the config, prompts, etc. And it's not easy to plug and play w/ the current API.
When on a call w/ a customer using deepagents recently, they mentioned they really liked the deepagent middleware, but they were going to have to copy all of the deepagents code and then make their own customizations to the middleware stack (ex, different settings for summarization, etc).
I would be more comfortable if we introduced just the middleware for now + really leaned into documenting how to create a "deep agent" (it's really just a few lines of code right, basically the list of middleware + the base agent prompt).
We could have a whole deepagents guide or even conceptual page that talks about all the ways you might want to customize your deep agent :)
| def wrap_model_call( | ||
| self, | ||
| request: ModelRequest, | ||
| handler: Callable[[ModelRequest], AIMessage], |
There was a problem hiding this comment.
note to myself we should have a type alias here, I can work on this over the weekend :)
|
Another note -- I've updated the HITL API as discussed, will want to update this accordingly. |
nfcampos
force-pushed
the
nc/9oct/file-tools-middleware
branch
3 times, most recently
from
255b316 to
4d48faa
Compare
nfcampos
force-pushed
the
nc/9oct/file-tools-middleware
branch
from
4d48faa to
6b6eeae
Compare
4 participants
tasktool which allows the main agent to create subagentsTODO