MemoryAtlas

ai-memory

AkitaOnRails (Fabio Akita)

A single Rust server that gives coding agents a shared, persistent markdown wiki in a git repo. Lifecycle hooks fire-and-forget capture every prompt, tool call, and session boundary; at session end (or PreCompact) the relevant pages are consolidated into a coherent narrative, and the next agent that starts in the same directory receives a 'where you left off' handoff before its first prompt. Exposes MCP + an HTTP /web browser UI on the same axum server; a thin-client CLI talks to the server over HTTP.

Storage
Plain markdown in a git repo (grep-able, Obsidian-openable, rsync-backup-able), with a SQLite FTS5 index. Per-project isolation by construction (`<wiki_root>/<workspace_id>/<project_id>/…` keyed by stable UUIDs); supersession chain + git-versioned pages enable time-travel (`checkpoints`, `restore-page`, raw `git log`).
Retrieval
FTS5 full-text search over LLM-consolidated decision pages (not raw logs), fused with graph-neighbor RRF and — when an embedding provider (OpenAI/Voyage/Gemini) is configured — optional vector RRF, plus bounded raw-observation fallback for non-global searches. Zero-LLM mode still gives FTS5 search + rule-based summarisation; adding an LLM provider is opt-in and enables page consolidation, contradiction linting, and staged auto-improvement proposals.
Self-host
Self-host: trivial
License
MIT
Pricing
Open-source (MIT); self-hosted only, no paid or hosted tier. Runs local (loopback) or on a homelab/LAN/VPN box with bearer-token auth. · Free / OSS
GitHub stars
1,434
Last release
2026-08-07
Last commit
2026-08-07
First catalogued
2026-06-30

Strengths

  • Cross-agent handoff: quit Claude Code mid-task, resume in Codex (or Cursor, OpenCode, Gemini CLI…) hours later with a typed 'where you left off' block prepended automatically
  • Zero-friction capture via lifecycle hooks — you never type `write_note`
  • Plain markdown in git: grep-able, Obsidian-openable, rsync-backup-able; no vector database to babysit
  • Zero-LLM mode (FTS5 + rule-based summarisation) works with no provider; LLM is opt-in for consolidation/lint/auto-improve
  • Broad first-party client support: Claude Code, Codex, Devin CLI, OpenCode, Cursor, Gemini CLI, Claude Desktop, OpenClaw, Antigravity CLI, Grok Build CLI, Zero, Kimi Code, Oh My Pi, Pi, VS Code Copilot (MCP-only)

Watch out

  • Young, fast-moving project (repo created May 2026) — expect churn; evaluate long-term maintainership
  • It's a Rust server you run (single binary or Docker); primary targets are Linux and macOS, native Windows is experimental (WSL2 recommended)
  • Several clients (Grok Build CLI, Zero, Kimi Code) discard/ignore `SessionStart` hook stdout, so automatic handoff injection doesn't work there — recover via the MCP `memory_handoff_accept` tool instead
  • Hermes Agent support is a community-maintained plugin (`ai-memory-hermes-plugin`), not part of ai-memory's first-party install surface — review its compatibility and secret handling separately before use

Best for

  • Developers who switch between coding-agent CLIs/IDEs and want portable, git-native project memory with automatic session handoff
  • Teams wanting a self-hosted, grep-able wiki of decisions shared across agents and machines

How it integrates

Benchmark results

No sourced results yet.

Sources

Last verified 2026-08-01 · updated by refresh-framework-cards