feat(agentapi): stream coalesced tool-call events in SSE chat responses - #99
Open
fepfitra wants to merge 6 commits into
Open
feat(agentapi): stream coalesced tool-call events in SSE chat responses#99fepfitra wants to merge 6 commits into
fepfitra wants to merge 6 commits into
Conversation
Subscribe to tool_executed events, coalesce per session into a 5s window
(mirrors the Telegram channel pattern), and emit inline SSE events like
{"type":"tool","tool_name":"bash (2x)","has_error":false} so
streaming chat clients can show what the agent is doing.
The AgentAPI plugin authenticates via its own Bearer token, but the global enforce_auth before_request 302s every non-/api path to the login page — breaking /agentapi/v1/* for token-authenticated clients.
- flush pending coalesced tool events when the turn completes so the client sees them before the stream closes (the 5s window may not have elapsed) - actually unsubscribe on_tool_executed in the finally block (was leaking, off() mapped every event to on_chunk/on_turn_complete)
The queue loop exits after the sentinel, so a put_nowait'd tool item would never be consumed — pop the pending entry and yield it inline.
handle_message returns response=None immediately when message_buffer_seconds > 0 (buffered path). Non-streaming chat/completions then formatted content:null, breaking clients that parse choices[0].message.content (e.g. pitura-dashboard: 'could not read agent reply'). When response is empty, poll the session chatlog for the assistant reply anchored after this message's user row (120s deadline, 300ms interval).
handle_message returns None for buffered/async agents — the real reply arrives later via event_stream (llm_response_chunk/turn_complete). Only enqueue the sentinel when a real response was returned; otherwise stay subscribed or the reply events are lost (SSE shows only [DONE]).
fepfitra
force-pushed
the
feat/agentapi-tool-log-clean
branch
from
August 6, 2026 14:16
9de1ba5 to
fd59213
Compare
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Problem
The agentapi streaming chat endpoint (
/plugin/agentapi/v1/chat/completionswithstream: true) only forwardsllm_response_chunkandturn_completeevents. Streaming clients can show the agent's text but have NO visibility into what the agent is DOING (running bash, editing files, searching the web). Tool execution is a black box during long agent turns.Fix
Subscribe to the existing
tool_executedevent stream, coalesce tool calls per session into a 5-second window (mirroring the Telegram channel pattern from PR #95), and emit inline SSE events while the stream is open:data: {"type":"tool","tool_name":"bash (2x), python (1x)","has_error":false}has_errorsurfaces failures (client can style themtype: "tool"field distinguishes tool events from normalchoices[].delta.contentchunks.tool_executedis unsubscribed in the samefinallyblock as the other events.Verification
python -c 'import ast; ast.parse(...)'— syntax OK. Live-tested after restart: tool calls appear as compact chips in the chat while the agent works.