agentmemory-0-8-9

name	agentmemory-0-8-9
description	Persistent memory engine for AI coding agents providing automatic cross-session context capture, hybrid search (BM25 + vector + knowledge graph), and multi-agent coordination via MCP server with 43 tools. Works with Claude Code, Cursor, Gemini CLI, OpenCode, Hermes, OpenClaw, and any MCP client without external database dependencies. Use when building AI coding agent workflows requiring persistent memory, semantic recall across sessions, or multi-agent coordination with shared context.
license	MIT
author	Tangled <[e-mail]>
version	0.1.0
tags	["ai-agents","memory","mcp-server","persistent-context","vector-search","hybrid-search","knowledge-graph","multi-agent"]
category	ai-agent-tools
external_references	["https://github.com/rohitg00/agentmemory/tree/v0.8.9","https://arxiv.org/abs/2410.10813","https://github.com/rohitg00/agentmemory/blob/v0.8.9/CHANGELOG.md","https://iii.dev/docs"]

agentmemory v0.8.9

Overview

agentmemory is a persistent memory system for AI coding agents. It runs as a background service that silently captures every tool use, file edit, and decision your agent makes — compresses it into searchable structured memory — and injects the right context when the next session starts. One command. Works across all major agents. No external databases required.

Built on iii-engine's three primitives (Worker/Function/Trigger), agentmemory replaces the traditional Express + Postgres + Redis stack with a single native binary that provides HTTP triggers, KV state, WebSocket streams, and worker management. The codebase is 118 TypeScript source files (~21,800 LOC) with 715 passing tests.

Key metrics:

• 95.2% retrieval recall at R@5 on LongMemEval-S (ICLR 2025 benchmark, 500 questions)
• 92% fewer tokens vs. pasting full context into conversation
• 43 MCP tools for memory operations
• 12 auto-capture hooks (zero manual effort)
• 0 external database dependencies (SQLite + in-memory vector index via iii-engine)

When to Use

• Building or operating AI coding agents that need persistent cross-session memory
• Setting up Claude Code, Cursor, Gemini CLI, OpenCode, Hermes, OpenClaw, Codex CLI, Cline, Goose, Kilo Code, Aider, Claude Desktop, Windsurf, Roo Code, or Claude SDK with memory
• Replacing built-in agent memory (CLAUDE.md, .cursorrules) that caps at ~200 lines
• Implementing hybrid semantic search (BM25 + vector embeddings + knowledge graph) for code context
• Coordinating memory across multiple AI agents working on the same project
• Needing real-time observability of agent memory via the built-in viewer (port 3113)
• Migrating from cloud-dependent memory systems (mem0, Letta, Hippo) to self-hosted alternatives

Core Concepts

The Memory Problem

Every AI coding agent forgets everything when the session ends. You waste the first 5 minutes of every session re-explaining your stack, architecture decisions, and bug fixes. Built-in memory files (CLAUDE.md, .cursorrules) cap out at ~200 lines and go stale. agentmemory solves this by running as a background service that captures, compresses, indexes, and retrieves context automatically.

How It Works — The Memory Pipeline

1. Capture — 12 hooks fire on every agent event (tool use, file edit, session start/end, errors)
2. Dedup — SHA-256 deduplication with a 5-minute window prevents duplicate observations
3. Privacy Filter — API keys, secrets, and <private> tags are stripped before storage