feat:add_request_level_token_callback#1362
Open
fengju0213 wants to merge 14 commits into
Open
Conversation
Collaborator
Author
Screen.Recording.2026-02-24.at.14.47.55.mp4The video shows that the token is updated with each tool call (request). |
Collaborator
Author
|
pending camel-ai/camel#3872 |
Pakchoioioi
requested review from
Wendong-Fan and
bytecii
and removed request for
Wendong-Fan and
bytecii
Wendong-Fan
reviewed
Mar 4, 2026
Wendong-Fan
reviewed
Mar 4, 2026
Contributor
There was a problem hiding this comment.
seems in streaming mode _stream_chunks / _astream_chunks call _send_agent_deactivate(accumulated_content, total_tokens) in their finally block with the full step token count, while request_usage events have already reported those same tokens, the frontend adds both, resulting in inflated token counts?
added pr #1439
Resolve conflict in listen_chat_agent.py: combine main's model-reload / auth-retry logic with the request-level token callback. - Move api_task_id/agent_name assignment before super().__init__ (needed by the _on_request_usage callback) while keeping main's _model_reload_callback / _model_reload_lock setup after super(). - Keep both _on_request_usage (request-level usage) and the auth-retry reload methods from main. - _send_agent_deactivate zeroes tokens when CAMEL exposes on_request_usage, so streaming (_stream_chunks/_astream_chunks) no longer double-counts. - Consolidate astep's inline deactivate to call _send_agent_deactivate, matching sync step() and removing the duplicated zeroing branch. - Bump camel-ai to main's 0.2.91a3 (still exposes on_request_usage). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…zeroed Request-level token reporting relies on CAMEL's on_request_usage callback, but OpenAI-family providers omit usage from streamed responses unless `stream_options.include_usage` is set. Since the step-level deactivate is zeroed once request-level reporting is active, streaming steps (the task_agent runs streaming by default) were counted as 0 tokens. Inject `stream_options.include_usage=True` in agent_model when streaming, except for native-SDK platforms (Anthropic, Bedrock, etc.) that reject the OpenAI-specific param and report streaming usage on their own. Verified against a real model (openai-compatible, streaming): - before: request_usage fires with 0 tokens -> step counted 0 - after: request_usage tokens=366, deactivate_agent tokens=0 (exact, no double count) - anthropic task_agent: stream=True, no stream_options injected (unchanged) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Once request-level usage reporting is active, _send_agent_deactivate zeroes
deactivate_agent.tokens. Two consumers relied on that per-event field:
- single-agent SSE: _action_to_sse did not forward Action.request_usage, so
in single-agent mode request-level events were consumed and dropped while
the deactivate fallback was zeroed -> no real-time token updates (only the
final end.tokens backfilled the total). Forward request_usage like
chat_service does.
- question_confirm quick-reply detection (chatStore): read the zeroed
deactivate_agent.tokens, so hasTokens was always false and the quick-reply
"mark trigger execution Completed" branch never ran. Base it on the task's
accumulated total (getTokens) instead.
Verified:
- _action_to_sse(ActionRequestUsageData(tokens=123)) now emits
`data: {"step": "request_usage", ...}` (was None/dropped)
- backend tests: test_listen_chat_agent (15 passed, 1 skipped),
test_single_agent_service (1 passed)
- frontend chatStore tests unchanged (5 pre-existing getProjectById
failures, request_usage case passes)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
4 participants
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.
Related Issue
Closes #https://github.com/eigent-ai/eigent_server/issues/46
Description
Summary
This PR adapts Eigent to CAMEL request-level usage reporting for non-stream
mode by introducing a dedicated SSE event
request_usage, enabling real-timetoken accumulation during a step loop.
Background
Previously, Eigent consumed token usage mainly via
deactivate_agent(step-level). This is too coarse for long multi-request loops inside a single step
and can delay downstream billing/accounting actions.
What Changed
Action.request_usageActionRequestUsageDataListenChatAgent:ChatAgentsupportson_request_usagerequest_usageper model request with:tokens(request total tokens)request_indexresponse_idstep_total_tokenschat_servicenow emitsrequest_usageAgentStep.REQUEST_USAGEchatStoreconsumesrequest_usageand callsaddTokens(...)immediately
deactivate_agent.tokensis set to0request_usage+deactivate_agent)request_usageincrements tokens in real timedeactivate_agent(tokens=0)does not over-countCompatibility
on_request_usage, Eigent falls back toprevious step-level behavior automatically.
Validation
python3 -m py_compile backend/app/agent/listen_chat_agent.py backend/app/ service/task.py backend/app/service/chat_service.pynpm run test -- test/unit/store/chatStore.test.tsScope
Testing Evidence (REQUIRED)
What is the purpose of this pull request?
Contribution Guidelines Acknowledgement