agenttrail-sdk 0.1.0

AgentTrail Python SDK — observability for agentic AI systems

AgentTrail Python SDK

Client library for tracing agent and LLM workloads: it records spans and sends them to your ingestion endpoint (POST {base_url}/v1/spans). It does not include a server, database, or UI—only instrumentation and export.

Install (PyPI)

pip install agenttrail-sdk

LangGraph, OpenAI, and Anthropic clients are dependencies of this package—customers install once; new framework support ships in new SDK releases, not separate extras.

Developers can use uv with this repo: see pyproject.toml and uv.lock.

Quick usage

Set base_url to your span ingestion service, or set the environment variable AGENTTRAIL_BASE_URL (default if both omitted: http://127.0.0.1:8080).

from agenttrail import init, shutdown

init(
    api_key="your-api-key",
    project="my-agent",
    base_url="https://ingest.example.com",
)
try:
    ...  # app / agents
finally:
    shutdown()

By default, init() auto-instruments LangGraph (invoke / ainvoke / stream / astream), OpenAI chat.completions.create, and Anthropic messages.create. Use auto_instrument=False to disable.

Docs

Links use the GitHub repository so they work from PyPI and from this checkout.

Doc	Purpose
QUICKSTART	Examples (LangGraph, OpenAI, Anthropic, manual spans)
CONNECTOR_SPEC	Building third-party connectors
TESTING	Automated checks and manual smoke tests
PUBLISHING	PyPI release (uv build, uv publish)
SDK_Plan	Internal roadmap / program context

Development

cd /path/to/repo   # repository root containing pyproject.toml
uv sync
uv run pytest
uv run ruff check src tests
uv run ruff format --check src tests
uv run mypy src

License

Apache License 2.0 — see LICENSE and NOTICE.