We built callback handlers for LangChain (TypeScript and Python) that automatically create portable, cryptographically signed departure and admission records as agents run.
The idea: when an AI agent leaves one platform and arrives at another, there should be a verifiable record of that transition. Think of it as passport stamps for AI. The handler hooks into LangChain’s lifecycle events to create signed EXIT markers on departure, and verify them on arrival.
Why this matters: if you’re running agents across multiple platforms or environments, there’s currently no standard way to track where they’ve been or verify their history. These records give you an auditable trail without depending on any single platform’s logs.
Quick example (Python):
from exit_door_langchain import ExitCallbackHandler
from exit_door import generate_keypair
keys = generate_keypair()
handler = ExitCallbackHandler(
keys=keys,
subject="did:key:z6Mk...",
origin="platform-a"
)
Supports Ed25519 and P-256 (FIPS-compliant path), offline verification, and a policy engine with 7 admission presets for the receiving side.
The packages are on npm (@cellar-door/langchain) and PyPI (exit-door-langchain). Part of the Passage Protocol, which was submitted to the NIST AI Agent Standards Initiative. Apache 2.0, open source.
Happy to answer questions about the design or take feedback. The project is at cellar-door.dev and the GitHub org is CellarDoorExits.