🧠
Prism MCP
Mind Palace for AI Agents
A Model Context Protocol server that gives Claude, Cursor, and other AI tools long-term memory that survives across sessions — with semantic search, cognitive routing, a visual dashboard, and the prism-coder LLM fleet for offline tool-calling.
Install
# npx (no install)
npx prism-mcp-server
# or install globally
npm install -g prism-mcp-server && prism-coder# LLM fleet (offline tool-calling — no API key needed)
ollama pull dcostenco/prism-coder:1b7 # 2.2 GB · any machine
ollama pull dcostenco/prism-coder:8b # 4.7 GB · Mac M1+ / iPhone 8GB
ollama pull dcostenco/prism-coder:14b # 8.4 GB · Mac M2+ / iPad Pro
ollama pull dcostenco/prism-coder:32b # 16 GB · Mac M2 Ultra+What it does
💾
Persistent memory
Every session feeds the Mind Palace. Next session your AI loads the right context automatically.
🔍
Semantic search
Vector + keyword + graph traversal over your full history. Ask questions, get citations.
🧬
Cognitive routing
Episodic, semantic, and procedural memory stored separately. The router picks where to save and retrieve.
🔄
Drift detection
Detects when your agent drifts from original goals mid-session and self-corrects automatically.
🛡
Local-first
Free tier runs entirely on your machine — SQLite, local embeddings, no API keys, no cloud.
🤖
100% BFCL accuracy
prism-coder models trained specifically for function-calling — 1.7B through 32B, all at 100% routing accuracy.
💻
Synalux Coder — Web IDE
Use Prism Coder directly in your browser. Local-first IDE with the prism-coder agent built in — connects to your GitHub repos, Synalux Mail, Drive, and Source. No install required.
Open Web IDE →