wtf is luthn?

Luthn is a self hosted tool that gives multiple AI coding assistants, like Codex or Claude Code, a shared memory that persists across sessions and projects. Normally each conversation with an AI agent starts fresh with no memory of past work, and Luthn is built to fix that by storing context in your own infrastructure using Docker and PostgreSQL, rather than sending your raw data to an outside company. The key idea is a data boundary. When an agent finishes a task, a small hook captures a limited summary of what happened, and Luthn classifies and redacts that summary before storing it. Only these reviewed, safe summaries are ever shown back to an agent, never your original private records, credentials, or unredacted messages. When a new task begins, Luthn automatically fetches a small relevant slice of past context so the agent can pick up where earlier work left off, without needing to ask for it every time. Setting it up involves connecting each coding agent to your Luthn installation with a simple command, which installs a hook for that agent, registers a background service the agent can talk to, and adds recall instructions to the agent's own configuration file. Codex and Claude Code can both be connected to the same Luthn installation at once, sharing the same safe memory while keeping their individual connections managed separately. The project documents this connection process for macOS, Linux, and Windows, with slightly different technical handling on Windows to avoid the agent cutting off the memory upload early. Luthn also keeps an audit trail of what was stored, shared, and retrieved, and treats publishing anything externally as a separate, explicit approval step rather than something that happens automatically. The self hosted runtime itself is licensed under AGPL-3.0, a copyleft license, while the software development kits, HTTP connectors, and public plugin templates use the more permissive Apache-2.0 license.