MemoryAtlas

Cross-platform semantic memory layer for AI coding agents. Markdown files are the source of truth — memories are plain `.md` files that are human-readable, directly editable, and version-controllable. Milvus is used as a 'shadow index': a derived, rebuildable cache over the Markdown corpus. Ships a procedural 'Skills from Memory' layer that distils repeated workflows into installable reusable skills. Plugins cover Claude Code, OpenClaw, OpenCode, and Codex CLI; a single memory store is shared across all agents.

Storage
Daily Markdown files under `.memsearch/memory/` store conversation memories as flat text. Advanced maintenance tasks keep persistent `PROJECT.md` and `USER.md` summary notes current across sessions. All writes are append-style to Markdown; Milvus indexes are rebuilt on demand from the Markdown files, making the vector index a disposable cache rather than authoritative storage.
Retrieval
Three-layer progressive retrieval: (1) dense vector similarity in Milvus, (2) BM25 sparse search, and (3) RRF reranking to fuse both signals. A file watcher auto-indexes new Markdown content in real time. Content hashing (SHA-256) skips unchanged files. Retrieved context is injected via `/memory-recall` slash command or auto-invoked by the LLM when it senses a history question.
Self-host
Self-host: trivial
License
MIT
Pricing
Open-source MIT, free to self-host. Installs via pip or CLI. Milvus (the vector index) runs locally by default — no external cloud service required. No hosted tier found. · Free / OSS
GitHub stars
2,156
Last release
2026-06-22
Last commit
2026-06-22
First catalogued
2026-06-28

Strengths

  • Markdown-as-source-of-truth: vector index is a rebuildable cache, so memories survive index corruption or migration without data loss
  • True cross-agent memory: one shared store accessible from Claude Code, OpenClaw, OpenCode, and Codex CLI simultaneously
  • Skills-from-Memory procedural layer distils repeated workflows into installable skills — adds a third memory tier beyond episodic/semantic
  • MIT-licensed, actively maintained (v0.4.11 as of 2026-06-22); 2156+ stars

Watch out

  • Zilliz / Milvus vendor origin: while the core is MIT, Milvus is the recommended production vector backend — evaluate lock-in if Milvus adoption is not already planned
  • Snseam (independent curation source) flags: 'strong Milvus dependency, not equivalent to full memory governance layer' — verify that write/update/forget lifecycle is first-class for your use-case, not just retrieval
  • Primarily optimised for coding-agent workflows; suitability for general conversational or multi-domain agent memory is unverified

Best for

  • AI coding-agent setups (Claude Code, Cursor, OpenCode, Codex CLI) where cross-agent shared memory and Markdown inspectability are priorities
  • Teams that want persistent memory without a dedicated database server — the Milvus index is local and the Markdown files are the durable record

Benchmark results

No sourced results yet.

Sources

Last verified 2026-06-28 · updated by discover-frameworks