ai-memory hit #2 on GitHub trending today with 2,885★ — and it earned it. This is a Rust-based long-term memory layer for coding agents that solves the most annoying problem in the multi-agent workflow: handoff.
The pitch is direct: quit Claude Code mid-task, open OpenAI Codex in the same directory, continue without re-explaining the architecture, the failed approaches, or the open questions. It works by capturing sanitized lifecycle observations from every agent session, compiling them into summaries, and feeding the next agent a bounded handoff packet.
The interesting part is the storage layer. No vector database. No write_note ceremony. The "wiki" is plain markdown in a git repo — grep-able, openable in Obsidian, backed up with rsync. This is a deliberately contrarian take from someone who clearly built this for real use, not demo gravity.
The support matrix is what got my attention. ai-memory has lifecycle hooks and MCP integration for Claude Code, OpenAI Codex, Gemini CLI, Cursor, Grok Build CLI, Kimi Code, Kiro CLI, OpenCode, Devin CLI, and more. It also has experimental support for Hermes Agent via a community plugin. The v1.28.1 release dropped yesterday (Aug 18), which is likely what triggered the surge.
Signal worth tracking: The agent ecosystem is fragmenting fast — every vendor ships a coding CLI, and none of them share context. ai-memory is betting that the fix isn't a new platform but a thin, language-agnostic persistence layer that agents opt into. The "plain markdown, no vector DB" stance is a bet on durability over complexity. If this becomes the de facto bridge between agent ecosystems, it's a bigger deal than the star count suggests.
Verdict: Not a new category — but a signal that the agent interoperability problem is finally getting production-grade solutions. Worth running on Monday to see whether the handoff really works.