memX
A self-learning memory plugin for AI coding agents that turns finished work into structured, searchable memory and injects only the evidence relevant to the current query, with native hooks for Codex, Claude Code, and OpenClaw.
memX turns completed agent work into structured, searchable, self-maintained memory, then injects only the evidence an agent needs for its current query. It connects natively to Codex, Claude Code, and OpenClaw through hooks, and reaches any MCP-compatible client through the same local memory layer.
Setup requires choosing an LLM provider, such as an OpenAI-compatible endpoint, Anthropic, Google, or Ollama, for memory compilation, recall planning, and maintenance, plus a base URL, model, and API key. Embeddings default to a local model, with an override for a different embedding provider. Native hooks are the default lifecycle path for Codex and Claude Code, handling automatic recall and turn capture while keeping MCP memory tools hidden by default to avoid duplicate writes.
The project reports internal benchmark results on long-context memory retrieval and a set of real engineering test cases, and requires a recent Node.js runtime.