Releases · Ultrathink-Solutions/openclaw-logfire

What's Changed

feat: subscribe to llm_input/llm_output hooks for token tracking and per-LLM-call spans by @namabile in #6

test: add hook handler unit tests, boost coverage 28% → 67% by @namabile in #7

chore: upgrade OpenTelemetry SDK from 1.x to 2.x by @namabile in #8

chore: bump dev dependencies to latest majors by @namabile in #9

chore: add tag-triggered npm publish workflow with provenance (ULT-578) by @namabile in #10

fix: use OIDC auth for npm publish instead of token-based .npmrc (ULT-578) by @namabile in #11

fix: clear NODE_AUTH_TOKEN and upgrade npm for OIDC publish (ULT-578) by @namabile in #12

Initial Release

Full agent lifecycle tracing with OTEL GenAI semantic conventions for Pydantic Logfire.

invoke_agent root spans with agent name, session key, workspace, and channel attribution
execute_tool child spans with tool arguments, duration, and result size
Error tracing with OTEL exception events (type, message, stack trace)
Token usage metrics (gen_ai.client.token.usage histogram)
Operation duration metrics (gen_ai.client.operation.duration histogram)
Pydantic Logfire OTLP HTTP export (US and EU regions)
Clickable Logfire trace URLs in agent logs
Secret redaction for API keys, platform tokens, JWTs, and credentials
W3C traceparent injection into HTTP commands for distributed tracing
W3C trace context extraction from inbound webhook headers
Configurable trace depth (tool input/output capture, message content, inference events)
Batch and simple span processor modes
Zero-config quickstart (just set LOGFIRE_TOKEN)

npm install @ultrathink/openclaw-logfire