parlot-instrumentation-langgraph

MIT OpenTelemetry instrumentation for LangGraph / LangChain agents — the text/webchat path into Parlot.

Parlot’s lead product is production voice AI observability on LiveKit. Use this package when you already run voice and want the same debugger for LangGraph / LangChain sessions (same turn model — not a separate observability product). Prefer parlot-instrumentation-livekit for voice agents.

pip install parlot-instrumentation-langgraph
# or: uv add parlot-instrumentation-langgraph
# or: pip install "parlot[langgraph]"

from parlot.instrumentation.langgraph import parlotize

parlotize(agent_id="support-bot", version="0.1.0", channel="webchat")

# Normal LangGraph usage — no per-call callbacks required
graph.invoke(
    {"messages": [("user", "Hello")]},
    config={"configurable": {"thread_id": "demo-1"}},
)

If thread_id is not provided, Parlot mints one automatically (anon-…) so contract spans still attach to a session.

Pass channel (e.g. webchat, sms) when LangGraph owns the session so ingest does not assume voice. Optional modality overrides the default derived from channel (webchat/sms/whatsapp → text, voice → voice).

Set PARLOT_ENDPOINT and PARLOT_API_KEY. Generative AI content capture (message bodies / tool payloads; default on) is documented in Concepts. Override with parlotize(capture_genai_content=False).

Session application logs (Python logging, default on) follow Settings → Logs / parlotize(capture_logs=…, log_level=…). Not print(). Treat content like stdout for PII.

When used inside a LiveKit voice agent that already called parlot.instrumentation.livekit.parlotize(), this package nests GenAI operational spans under the active LiveKit session and does not emit duplicate parlot.session / parlot.turn spans — LiveKit owns channel and modality.

Docs: LangGraph guide · Concepts · parlot.ai