v0.1.9 - Complete Metrics Implementation

Summary

End‑to‑end metrics support for Traccia: standardized LLM + agent metrics across core SDK, integrations, and docs, with a central recorder, pricing awareness, and tests.

Highlights

Core metrics system

• Introduced a unified metrics layer (metrics.metrics, metrics.recorder) to emit:
  • Token usage
  • Operation cost (USD)
  • Operation duration / latency
• Added a global metrics recorder accessor so integrations can record metrics without tight coupling.

SDK & instrumentation

• Updated core instrumentation (auto.py, config.py, instrumentation/openai.py, instrumentation/anthropic.py, instrumentation/requests.py) to:
  • Attach pricing‑aware cost metrics to each LLM call.
  • Record token usage and duration consistently under gen_ai.client.*.
• Exposed metrics configuration and behavior at the top level (init.py) for easier imports and discoverability.

Framework integrations

• LangChain: Extended callback handler to emit standardized metrics (tokens, cost, duration) for LangChain LLM calls.
• CrewAI: Instrumented CrewAI integration to record LLM metrics via the central recorder.
• OpenAI Agents SDK: Enhanced processor to:
  • Track span start times.
  • Record generation metrics (tokens, cost, duration) on span end using the new metrics layer.

CLI & configuration

• Updated cli.py and config.py to:
  • Include metrics options in configuration handling.
  • Ensure traccia config init produces a config that is compatible with the new metrics system.

Testing & docs

• Added tests/test_metrics.py with coverage for:
  • Metrics recorder behavior.
  • Emission of token, cost, and duration metrics under expected names/attributes.
• Expanded README.md with:
  • An overview of built‑in metrics (LLM + agent).
  • How metrics relate to pricing and token usage.
  • Guidance on enabling metrics in typical setups.