roomy-observability 0.1.1

Local-first LangChain agent observability: traces, context segments, tokens, CLI & web UI

Roomy

Roomy is a local-first observability toolkit for LangChain agents. It records what entered the model (messages, tools, retrievers), token usage, latency, and structured traces in SQLite, with a CLI and an optional web UI for inspection.

Install

pip install roomy-observability

• PyPI package: roomy-observability
• Python import: import roomy
• CLI: roomy (e.g. roomy serve, roomy sessions list)

Requires Python 3.11+.

Quick start

from roomy import end_session, instrument_langchain

bindings = instrument_langchain(app_name="my-agent", db_path="./traces.db")
result = chain.invoke(
    {"topic": "Hello"},
    config={"callbacks": bindings.callbacks},
)
end_session(bindings.manager)

Use wrap_agent(chain, bindings.manager) if you prefer binding callbacks on a Runnable.

Optional extras

Extra	Purpose
api	FastAPI + Uvicorn (roomy serve for the inspector API)
openai	tiktoken for better token estimates
dev	Tests, ruff, mypy, build, twine

pip install "roomy-observability[api,openai]"

Documentation & source

Full monorepo (sample agents, web app, architecture docs): github.com/Sarang-Pramode/Roomy

Author: Sarang Pramode · License: MIT