memeX · Documentation

memeX — Cognitive Memory for AI Agents

memeX is an open-source MCP server giving AI coding tools and agents persistent, graph-structured cognitive memory. Free, Apache 2.0 licensed, local-first, zero telemetry.

v1.0 GAApache 2.0Free foreverZero telemetry

Persistent cognitive memory for AI coding tools. Your AI coding tool has amnesia. memeX fixes it. Local-first, MCP-native, ships with curated skills any AI can install and validate against. Also acts as an MCP gateway — one connection, two tool families: memeX’s native memory tools plus every upstream MCP server you wire in.

In one diagram

              ┌─────────────────────────────────────┐
              │         memeX core engine            │
              │  storage + retrieval + lifecycle     │
              │  + working set + ML — protocol-blind │
              └─────────────────────────────────────┘
                   ▲          ▲          ▲
                   │          │          │
        ┌──────────┴──────┐ ┌─┴────┐ ┌───┴────┐
        │   MCP frontend  │ │ HTTP │ │  CLI   │
        │   (stdio)       │ │ API  │ │        │
        └─────────────────┘ └──────┘ └────────┘

In three sentences

memeX is a local daemon giving Claude Code, Cursor, Windsurf, Cline, and any HTTP-capable AI tool a shared persistent memory — graph-structured, budget-bounded, provenance-stamped. It also ships skills: installable validation bundles (security, money, concurrency, crypto, SQL, untrusted input) the AI must self-attest against before generating sensitive code. Every action is logged, every confidence decays over time, every retrieval is bounded by tokens — modeled jointly on RAM hardware and human memory.

Where to go next

Status

v1.0 — generally available. Apache 2.0 licensed, no telemetry, runs entirely on your machine. File issues on github.com/queflyhq/memex.