LangChain

Fact0 ships a LangChain callback that records every chain, tool, and LLM call as a telemetry span, with optional audit events for sensitive actions.

Install

pip install fact0-sdk langchain

Quick example

from langchain_openai import ChatOpenAI
from fact0.integrations.langchain import Fact0CallbackHandler
import fact0

client = fact0.Client(api_key="f0_live_...")

handler = Fact0CallbackHandler(
    client=client,
    agent_id="research-bot",
    audit_sensitive_actions=True,
)

llm = ChatOpenAI(model="gpt-4o", callbacks=[handler])
response = llm.invoke("Summarize Q3 results")

What gets captured

Callback event	Telemetry	Audit (when `audit_sensitive_actions=True`)
LLM start/end	`MODEL_INVOCATION` span	-
Tool start/end	`TOOL_CALL` span	`agent.tool.call`
Chain start/end	execution lifecycle	`agent.run.start` / `agent.run.end`

Configuration

handler = Fact0CallbackHandler(
    client=client,
    agent_id="support-bot",
    execution_metadata={"tenant": "acme", "env": "prod"},
)

Option	Purpose
`agent_id`	Stable agent identifier shown in the dashboard
`execution_metadata`	Tags applied to every execution from this handler
`audit_sensitive_actions`	Also write `agent.tool.call` audit events

Telemetry client for manual span control
LangGraph for stateful graph agents