Memorix

Memorix keeps project memory for AI coding agents across chats and switches between tools. It's for developers who want decisions, fixes and unfinished work to remain available to the next session.

• Shared project memory: Claude Code, Codex, Cursor, Gemini CLI and other compatible MCP clients can use the same local pool of searchable facts, session summaries and design rationale.
• Task context: Retrieve a bounded selection of relevant facts, verification results and next steps instead of replaying whole conversations.
• CodeGraph: Connect memory to current files, symbols and code relationships, with freshness checks and impact context; TypeScript and JavaScript get detailed symbol coverage, while other languages have lighter support.
• Reviewed knowledge: Preserve evidence-backed claims and reusable project workflows through review, alongside engineering facts drawn from commits.
• Inspection and agent access: Inspect memory sources, retention and health in a local dashboard, or use the memory through MCP, the CLI and the bundled memcode terminal coding agent.

Memorix stores memory locally in SQLite and scopes search to the Git project by default. It's open source under Apache 2.0. Local full-text retrieval doesn't require an LLM key; model providers are optional. Hooks can capture useful session events. Media capture is opt-in, and explicitly imported assets stay in its local data directory. Image and video generation through MCP are disabled by default and depend on external providers.