agentmemory: what it is, what problem it solves & why it's gaining traction

What it solves

Coding agents often suffer from "memory loss" between sessions, forcing users to re-explain architectures, preferences, and previously discovered bugs. While static files like CLAUDE.md or .cursorrules exist, they are limited in size and quickly become outdated. agentmemory provides a persistent, shared memory layer that automatically captures and recalls relevant context across different AI coding agents.

How it works

Built on the iii engine, agentmemory acts as a memory server that integrates with agents via the Model Context Protocol (MCP), REST API, or hooks. It silently captures agent activity, compresses it into searchable memory, and injects the correct context into new sessions. It uses a hybrid search approach—combining BM25, Vector, and Graph search with Reciprocal Rank Fusion (RRF)—to retrieve information. To maintain efficiency, it employs a 4-tier consolidation process with decay and auto-forgetting mechanisms to prevent memory bloat.

Who it’s for

Developers using AI coding agents such as Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, and others. It is designed for those who want their agents to remember project-specific details across sessions and across different tools without manual copy-pasting or manual memory management.

Highlights

• Broad Compatibility: Works with any agent that supports MCP, REST API, or hooks (e.g., Claude Code, Cursor, OpenCode, Aider).
• High Retrieval Accuracy: Achieves 95.2% R@5 on the LongMemEval-S benchmark.
• Zero External Dependencies: Uses SQLite and the iii-engine locally, requiring no external vector databases.
• Automatic Capture: Features 12 auto-hooks to capture information with zero manual effort.
• Token Efficiency: Significantly reduces token usage compared to pasting full context or LLM-summarization.
• Real-time Viewer: Includes a built-in viewer on port 3113 to monitor memory in real-time.