Data basis: the happydog-intj/ai-xiaohongshu-daily bot issue #575 (GitHub Trending AI daily, 2026-10-19, 01:00 slot), compared with the same slot the previous day (#572, 10-18) to measure day-over-day growth. The primary metric is stars today. Six of the seven listed repos were excluded for having 40k+ total stars (ai-job-search 52.1k, claude-plugins-official 42.4k, opencode 216.2k, ruflo 76.4k, orca 96.2k, CLI-Anything 54.4k). Trendshift and direct github.com figures were not used due to caching issues.
akitaonrails/ai-memory — +513 ⭐ today
https://github.com/akitaonrails/ai-memory
Total stars: 10,303 · Day over day: +208 → +513 (about 2.5x) · Language: Rust · License: MIT
ai-memory is a long-term memory system for AI coding agents: you can stop work in Claude Code and pick it up in Codex or another agent on the same project without re-explaining the context. Lifecycle hooks capture prompts, tool calls and session boundaries as privacy-sanitized observations, and at session end these are consolidated into human-readable markdown wiki pages. A single Rust binary runs an MCP/HTTP server and owns one data directory containing wiki/ (git-versioned markdown source of truth), raw/ (sanitized transcript segments) and db/ (SQLite indexes with FTS5, entities and embeddings, rebuildable from the markdown). The next session, on any agent or machine, gets a bounded brief and can search memory by full text, entities, links and optionally vectors, while cross-agent handoffs are typed and claimed exactly once. It supports a broad set of tools including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Kimi Code and Kiro CLI, and its rise reflects growing demand for shared memory across vendors as developers juggle multiple coding agents.
Key features:
- Cross-agent and cross-machine handoffs, with memory hosted on a server you run
- Team mode with shared per-project knowledge, multi-user auth, per-person attribution and an audit log
- Plain markdown storage you can edit by hand or open in Obsidian
- Silent capture with privacy sanitization and per-repo capture rules
- Zero-LLM default mode (no API key needed), with optional LLM and embedding providers for better summaries and semantic search
- Managed workstreams (
ai-memory run/continue) plus an optional read-only web UI and JSON API (--enable-web)
Practical takeaway: Teams using both Claude Code and Codex can host project memory on an internal server and share it, cutting the cost of re-explaining context when switching agents or handing work between engineers.
Tags: #AgentMemory #ClaudeCode #MCP #Rust #CodingAgents #KnowledgeBase
Excluded for reference: MadsLorentzen/ai-job-search (+2,213), anomalyco/opencode (+1,417), anthropics/claude-plugins-official (+1,187), stablyai/orca (+1,146), HKUDS/CLI-Anything (+582) and ruvnet/ruflo (+502) all grew strongly but already exceed 40k total stars.
Leave a Reply