Just shipped v3.5 of the Memory Sidecar: a production-grade persistent memory for AI agents. The architecture stays battle-tested (3-tier: Hot→Warm→Cold + knowledge graph + Focused Dossiers), but the scope just expanded dramatically.
What's new:
• Multi-agent shared runtime — Hermes, Claude Code, Cursor, Codex all share one memory pool. Deploy once, every agent benefits.
• Bilingual installer — The installer itself auto-detects zh/en. Chinese prompts throughout.
• Install modes 1/2/3 with automatic fallback — Pick your depth, the installer adapts.
• KMM integration — Official tie-in with the Knowledge-and-Memory-Management pipeline. Web → video → documents → all flow into the same memory graph.
6 embedding models to choose from (96MB–2GB), Focused Dossiers for priority topics, and 11 production scripts. All MIT licensed, 152 stars.
→ github.com/mage0535/hermes-memory-installer
Top comments (0)