AVIDS2/memorix

What it solves

Memorix provides a shared, local-first memory layer for AI coding agents. It prevents the "forgetting" problem where agents lose context between different chat sessions, IDE switches, or when switching between different AI tools (e.g., moving from Claude Code to Cursor). By storing project memory outside of a single chat window, it ensures that design decisions, fixes, and project-specific knowledge persist across the entire development lifecycle.

How it works

Memorix acts as a project-scoped memory pool that integrates with agents via the Model Context Protocol (MCP), plugins, hooks, and project rules. It uses SQLite as the primary local store and Orama for search. It can turn Git commits into searchable engineering facts and store reasoning (the "why" behind decisions) separately from the code itself. It is local-first, meaning it works with full-text retrieval even without LLM embedding keys.

Who it’s for

Developers who use multiple AI coding agents or frequently start new sessions and find themselves repeatedly explaining the same project architecture, constraints, and previous decisions to their AI assistants.

Highlights

• Cross-Agent Compatibility: Works with a wide array of agents including Claude Code, Cursor, Windsurf, GitHub Copilot CLI, and others via MCP.
• Git Memory: Automatically converts Git commits into searchable project facts.
• Reasoning Memory: Dedicated storage for design rationale, trade-offs, and risks to prevent architectural drift.
• Orchestration: Tools for coordinating multi-agent work, including task planning, file locks, and handoffs.
• Local-First: Uses SQLite and Orama for local storage and search, ensuring privacy and performance.