abmind

Persistent cross-session memory for AI agents.

abmind gives any AI tool long-term memory — facts, preferences, conversation history, and emotional context that persists across sessions and survives restarts.

How it works

Every conversation turn is recorded. A background "sleep" cycle extracts facts, detects patterns, consolidates knowledge, and prunes stale memories. On the next session, relevant memories are recalled and injected into the agent's context.

Documentation

• Why abmind? — motivation, design philosophy
• Installation — setup guide
• Memory System — how storage, extraction, consolidation, and darwinism work
• Recall Pipeline — the 4-layer search algorithm, ranking, deduplication
• Classification — NATO Admiralty Codes, trust/integrity/credibility, access control
• Configuration — all ABMIND_* env vars with defaults
• CLI Reference — all commands and flags
• Integration — embedding abmind in host CLIs and agent frameworks
• Session Context — how context is assembled per-turn
• Sleep Cycles — overnight processing, extraction, consolidation
• Backup & Restore — encrypted backups, scheduling, restore modes
• Security — encryption, permissions, multi-user isolation
• Troubleshooting — common issues and fixes

Use it with

• abTARS — in-process memory for the autonomous bridge
• Kiro CLI — native hooks + MCP server (abmind install-host kiro)
• Claude Code — hooks + MCP server (abmind install-host claude)
• Gemini CLI — hooks + MCP server (abmind install-host gemini)
• OpenAI Codex CLI — hooks + MCP server (abmind install-host codex)
• Hermes-Agent — memory provider plugin
• OpenClaw — ContextEngine plugin
• Any MCP client — abmind mcp starts a stdio MCP server

Quick start

npm install -g abmind
abmind install
abmind install-host kiro      # or: claude, gemini, codex

Community

• Discord: Join our server
• Email: aksikatwo@gmail.com
• GitHub: aksika/abmind