Jarvis Memory

A memory graph for me and my AI agents, in one SQLite file on my own machine.

Everything my agents should know about my work (projects, preferences, people, how I write) stored as linked memories in a single SQLite database that never leaves my computer.

What it does

• A graph view in the browser to browse, link and edit memories
• An MCP server, so any agent that speaks MCP reads and writes the same memory
• Recall that combines keyword search with meaning, using embeddings computed locally
• Files as memories: PDFs, Office documents and images are read, summarised and searchable by passage
• A scan that registers my projects and where their configuration lives, without ever storing a secret

How it is built

Node.js with the built-in SQLite module, so there is nothing to install. Keyword search is SQLite FTS5; meaning comes from a local embedding model through Ollama, and recall falls back to keywords when that is not running.

Secrets get particular care. Values from environment files are never stored in the database and are not reachable by agents at all. An optional vault keeps an encrypted copy, tied to my Windows login, and anything that looks like a secret in a document is redacted before it is saved.