Lians
Memory for Claude, Codex, Cursor, and any AI agent
Start new thread

Lians v0.5 - Reconstruct what your AI knew when it acted

by•
Open-source, bitemporal memory and decision evidence for AI agents. Recall facts as they were knowable at a prior time, preserve provenance, and export verifiable decision receipts.

Add a comment

Replies

Best
Maker
📌
AI systems often make consequential decisions using facts, permissions, policies, and memory that later change. Ordinary logs show outputs. Lians reconstructs the state that was actually available when the decision happened. We built Lians around two primitives: bitemporal memory and verifiable evidence. You can record when a fact was true and when the system learned it, query with recall_at, and preserve the sources behind the result. The project is Apache 2.0 and available for Python, TypeScript, MCP, and local deployment. Our favorite demo shows how an agent memory layer can leak future information into a historical backtest, then fixes the contamination with point-in-time retrieval. We are looking for direct product criticism, integration feedback, and teams with one consequential workflow that must remain reconstructable.
Maker

Update: Lians Personal is now live for people who want durable memory without running the service themselves. The local and self-hosted version stays free. Personal is $10/month and includes a private managed workspace, inspection/correction/deletion/export controls, and email setup help. I’m personally helping the first few users connect Claude, Codex, Cursor, or another MCP client. Please only try the paid plan if managed setup genuinely saves you time: