Sign in Join the beta

Research Lab

Open tools we build while making Atlas.

Open specifications with reference SDKs in Python and TypeScript, shared conformance vectors and signed, verifiable outputs. Free for anyone.

  • v0.1.1 Python and TypeScript Apache-2.0

    Open Pseudonym

    Task-aware pseudonymisation for LLM traffic. Detected entities become consistent, realistic stand-ins; secrets become handles bound to the host they came from; every request can carry an Ed25519-signed, hash-chained egress receipt.

    • 88 / 88vectors
    • 11profiles
    • ~1 msto mask

  • v0.4.0 Python and TypeScript Apache-2.0

    Open Action Consent

    Signed previews, consents, mandates and hash-chained receipts for agent tool calls. JCS canonical JSON, compact JWS (EdDSA or ES256), JSON Schema 2020-12 and an MCP binding.

    • 241 / 241vectors
    • 4tiers
    • 0runtime deps

How we build them.

  • Specification first

    Each project is a written standard: data model, algorithms, wire formats, error codes. Text under CC BY 4.0, code under Apache-2.0.

  • Shared test vectors

    Fixed seeds and clocks, exact expected outputs. Both SDKs pass every vector, and a third implementation can check itself the same way.

  • Honest limits

    Each project publishes what it does not cover yet and which of its numbers are optimistic.

From the blog

Nov 7, 2026Introducing Open Pseudonym: keep personal data away from AI modelsAn open specification and SDK that swaps names, contact details, dates and secrets for realistic stand-ins before text reaches a model, and puts the real values back on your side.Nov 7, 2026Introducing Open Action Consent: a yes you can proveAn open standard for AI agents that ask before they act. The person's yes becomes a signed object bound to exactly what they saw, and every action leaves a receipt.

Built while making Atlas.

Atlas is the AI workspace these tools come from. Free during the private beta.

Join the beta