Paxm

Paxm gives AI coding assistants persistent memory of project decisions, conventions and past work. It's for developers who want later sessions or another coding agent to recover context they've already explained.

• Automatic recall and capture. Codex, Claude Code, OpenCode, Pi, TRAE, Kimi Code, ZCode, Cline and the Kiro paxm agent can recall context before model requests and save completed turns afterward.
• Local SQLite memory. The default backend stores turn-level memories and returns excerpts relevant to a query without calling an external LLM or embedding service.
• Explicit memory tools. CLI and MCP access let agents save memories, retrieve context and inspect history; Cursor also supports automatic capture, but doesn't support automatic recall.
• Choice of storage. Connect self-hosted Mem0, MemOS or OpenViking, on-premises Team Memory, or a private memory system through JSON-RPC, and use multiple providers without replacing the agent integrations.

Automatic writes enter a durable local queue before delivery. Slow or unavailable providers retry in the background without blocking the coding session, and recall can return partial results when some providers time out.

Paxm is open source under Apache 2.0. Its default SQLite memory layer runs locally and needs no account or API key. Mem0 Cloud and MemOS Cloud provide hosted storage, and Zep is also supported. Memories sent to remote providers leave your machine; local memory storage doesn't make the connected coding assistant's model requests local.