Add an API for streaming tool calls performed by HandleResponseNode #831
+157
−62
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dmontagu
force-pushed
the
dmontagu/stream-tool-calls
branch
from
1443b49 to
c4e9180
Compare
|
@samuelcolvin I think we should discuss the ergonomics of |
|
Closing in favor of an approach based on #833, though I need to check if this has relevant implementation |
|
Actually 833 was based on this, so I guess we should merge it first |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds a
run_streammethod toHandleResponseNodein the Agent graph. That method is an AsyncContextManager yielding a stream (the pattern we use to ensure async-generator-wrapping spans get closed), and that stream yields events with information about tool calls being performed by the agent.(We can eventually add other events if desired, including possibly a way for users to emit custom events in their tools through a method on the GraphContext, but I think this is more than enough to start with.)
Once we expose a way to call the
Agentgraph'snextmethod, I believe this will be usable as a way to handle full agent event streaming — you'll usenextto transition between nodes, and nodes with interesting events will have arun_stream(or similar) method for streaming those events.I'm okay not merging this until we have a more fully-baked implementation of agent streaming (which I'll work on now), but I'm opening this PR now to get some early feedback.
I'll note that if we want to provide a way to inject control over the operation of the agent as part of the streaming (e.g., being able to see the list of tool calls and programmatically modify it before they are executed), instead of just providing a stream of events we could make use of a nested graph and expose its
nextcall to the user. But I don't currently see a compelling use case for that (at least forpydantic_ai.Agent; I could imagine it being useful in more sophisticated graph-based applications), and I don't think it's worth the extra complexity.