👋 Hey MemPalace community!

We've been building MemPalace-AGI — a community integration project that pairs MemPalace's spatial memory architecture with ASTRA-dev's autonomous scientific discovery engine. The idea: what if an AI research system could remember everything using MemPalace's palace/wings/rooms/drawers model, and use that structured memory to drive better hypothesis generation and testing?

We've put together a live dashboard showing the current state of the project, and we'd really appreciate honest feedback from the community before we go further.

🔗 Dashboard

MemPalace-AGI Dashboard

What the dashboard covers

• System Architecture — how MemPalace and ASTRA-dev connect (the bridge components)
• Discovery Cycle Results — real benchmark data from two autonomous research cycles (55 discoveries across 5 domains, 100% search relevance, 187 knowledge graph entities)
• Knowledge Graph — interactive visualization of causal relationships discovered by the system
• Phase Timeline — 14 phases of development so far
• Upstream Engagement — PRs and contributions back to this repo (docs: Add STAN extension to Community Plugins #319, fix: mitigate shell injection risk in hooks (#110) #320, and reviews on feat: optional remote ChromaDB support via HttpClient #294, fix: Ignore wait_for_previous argument to support Gemini MCP clients #322)
• System Health — test suite (86/86 passing), component status, blockers
• Technical Specs & Data Sources — the full stack

What we built (integration components)

What we'd love feedback on

1. Does this integration make sense? Is pairing spatial memory with autonomous research a natural fit for MemPalace, or are we stretching the architecture in ways that don't feel right?
2. Dashboard clarity — Is the dashboard understandable? Too much? Too little? What's confusing or missing?
3. Architecture concerns — We used a composition pattern (wrapping DiscoveryMemory with PalaceDiscoveryMemory) rather than forking either project. Does this approach seem sound?
4. Memory mapping — We map research domains → wings, hypotheses → rooms, evidence → drawers. Does this mapping feel natural for the palace metaphor?
5. What would you want to see next? Some directions we're considering:
   • Provenance tracking in the knowledge graph (agent_id, evidence chains)
   • Domain-adapted embeddings (SciBERT) for better scientific query matching
   • A2A protocol support for agent-to-agent discovery
   • Scaling tests to 200+ discoveries

Known issues we're already tracking

• ⚠️ ASTRA-dev has no license — legal blocker for redistribution (tracked separately)
• ⚠️ ChromaDB version pin at 0.6.3 (MemPalace uses 0.6.x while stable is 1.5.5)
• PR fix: mitigate shell injection risk in hooks (#110) #320 (shell injection fix) and docs: Add STAN extension to Community Plugins #319 (STAN extension) are awaiting review

Context

This is a community-driven project — not affiliated with or endorsed by the MemPalace maintainers. We're sharing it here because MemPalace is the foundational memory layer and we want to make sure we're building something the community actually finds useful.

All integration code lives at the composition boundary — we haven't forked or modified MemPalace core. The STAN extension was extracted as a separate repo specifically to keep things clean.

Any and all feedback — positive, critical, or "you're doing this completely wrong" — is welcome. Thanks! 🙏