Agent OS Python SDK

Python SDK for Agent OS — AI agent observability, experimentation, and security. "PostHog for AI Agents."

Install

pip install agentos

Quick Start

from agentos import wrap_openai
import openai

client = openai.OpenAI()
client = wrap_openai(client, api_key="aos_...", agent_id="my-agent")

# Every LLM call is now automatically traced
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)

That's it. Open your Agent OS dashboard to see traces, token usage, costs, and latency.

Other Ways to Instrument

@observe decorator (Langfuse-compatible)

from agentos import observe

@observe()
def my_agent(query: str) -> str:
    result = search(query)       # auto-traced as child span
    return summarize(result)     # auto-traced as child span

@observe(as_type="generation")
def summarize(text: str) -> str:
    return openai_client.chat.completions.create(...)

Manual event capture

from agentos import AgentOS

client = AgentOS(api_key="aos_...")

client.llm_call("my-agent", model="gpt-4o", system="openai", input_tokens=100, output_tokens=50)
client.tool_call("my-agent", tool_name="web-search", status="success", duration_ms=120)
client.eval("my-agent", eval_name="accuracy", score=0.95)
client.business_event("my-agent", event_name="task_completed")

client.flush()

Environment variable (zero-code)

export AGENTOS_API_KEY=aos_...
export AGENTOS_ENABLED=true
python my_agent.py

Framework Integrations

Migrating From Another Platform?

Change one import and keep your existing code:

Documentation

Full docs, guides, and API reference: docs.agentos.dev

License

MIT