AgentMemory is a TypeScript-based persistent memory system designed specifically for AI coding agents, built on the iii engine and positioned as the top solution based on real-world benchmarks. The project provides memory persistence across multiple popular coding agents including Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, Codex CLI, Hermes, OpenClaw, and any MCP client, eliminating the need for users to repeatedly explain context to their agents.
The repository demonstrates substantial real-world adoption and engagement. As of the most recent tracking period, the project maintains approximately 24,550 stars on GitHub. GitGenius activity tracking across 314 issues and pull requests shows a median response latency of 2.2 hours with a mean of 27.6 hours, indicating active maintenance and community engagement. The most frequently applied issue labels are medium priority (29 occurrences), enhancement requests (27), and high priority (17), reflecting ongoing development and feature expansion. The primary maintainer rohitg00 has logged 545 tracked events, with secondary contributors Tanmay-008 (20 events) and HaleTom (8 events) also actively involved in the project's development.
The system achieves measurable performance improvements in agent interactions. According to the README, AgentMemory delivers 95.2 percent retrieval accuracy at R@5, reduces token usage by 92 percent compared to baseline approaches, provides 53 MCP tools for integration, includes 12 automatic hooks for seamless agent integration, requires zero external databases for operation, and passes over 1,423 tests. The implementation extends Karpathy's LLM Wiki pattern with confidence scoring, lifecycle management, knowledge graphs, and hybrid search capabilities, as documented in a viral GitHub gist that has accumulated 1.3 thousand stars and 182 forks.
AgentMemory is distributed via npm as the @agentmemory/agentmemory package and supports installation through multiple pathways. The fastest installation method directs users to follow instructions at a dedicated INSTALL_FOR_AGENTS.md file that automates end-to-end setup and verification. Alternative installation options include npx execution without local installation, with version management handled through cache clearing if needed. The project maintains compatibility with existing iii engine installations by pinning to iii-engine v0.11.2 and refusing to attach to different versions due to protocol incompatibility.
The project's scope extends across multiple integration points. It functions as an MCP server, includes a real-time viewer for memory inspection, integrates with the iii Console for advanced management, and exposes both REST API and programmatic interfaces for agent integration. The system supports multiple languages in its documentation, with README translations available in Simplified Chinese, Traditional Chinese, Japanese, Korean, Spanish, Turkish, Russian, Hindi, Portuguese, French, and German, indicating international adoption and community contribution.
GitGenius classification data identifies AgentMemory within multiple overlapping domains including AI agents, memory management, knowledge storage, context retrieval, agent communication, long-term memory, information recall, natural language processing, multi-agent systems, and persistent storage. The repository maintains connections to related projects through shared contributors, linking to openclaw/openclaw, nousresearch/hermes-agent, and yeachan-heo/oh-my-claudecode, suggesting an ecosystem of complementary agent-focused tools and frameworks.
