Home / About

Built for agents that must answer for themselves.

Lians is open infrastructure for one hard question: what did the agent know when it decided, and can you prove it?

What exists today

Not a deck. A shipping system.

Open-source engineApache 2.0

Bitemporal store, deterministic supersession, admission control, SHA-256 audit chain, crypto-shred erasure, relationship graph, all inspectable at github.com/Lians-ai/Lians ↗.

Five SDKs + MCP

Python, TypeScript, Go, Java, C, released in lock-step, plus a native MCP server on the official registry for Claude Desktop, Cursor, and VS Code.

Managed console

API keys, live playground, and the Memory Governor review queues at lians.ai/login.

A benchmark you can re-run

The regulated-memory eval executes live against the engine and scores five competitors through real-SDK adapters. No adjectives.

Why we built this

A founder note.

Most agent memory accumulates facts and ranks by similarity. That breaks when an answer moves money, touches a patient, or faces review.

Regulated reality has two clocks: when something was true, and when you learned it. Lians makes both queryable. It is open source because trust should be verifiable. · Ethan Beirne, founder

Founding cohort · 7 openings

Founding design partners get the direct line.

Three implementation openings and four technical evaluation openings for teams putting agents on time-sensitive, audited, or confidential data.

Contact

Talk to a human.

Sales & partnerssales@lians.ai
Security reportssecurity@lians.ai
Everything elseGitHub issues ↗

Seven clearly scoped openings. Regulated teams first.

See the cohort →