Prism Mcp

Gives Claude persistent memory across sessions through a local SQLite store with vector search and cognitive routing. Exposes tools for semantic search over conversation history, session state management, and proactive drift detection that alerts when your agent veers off task. Ships with the prism-coder LLM fleet (1.7B to 32B parameters) for offline tool routing via Ollama, plus optional Brave Search and Gemini integration. Includes a knowledge ingestion pipeline that indexes your codebase into the graph via MCP tools, GitHub webhooks, or REST API. The local tier runs entirely on device with no API keys. Reach for this when you want your AI to remember decisions across weeks without re-explaining context, or when you need tool calling that works offline at zero cost per query.