AVIDS2/memorix

Memorix – A shared, local‑first memory layer for AI coding agents

What it is – Memorix is a Node‑JS library / CLI that creates a searchable, version‑controlled memory store inside the Git repository you are working on. The store lives in a local SQLite database (with optional full‑text and semantic indexes) and is exposed to any AI‑powered coding assistant that understands the Model‑Context‑Protocol (MCP) or can load a plugin/hook.

Why it matters – When you switch between agents (Claude Code, GitHub Copilot CLI, Cursor, etc.) or between IDEs and terminal sessions, each tool normally starts with a blank chat history. Memorix lets the agents share the same project‑wide facts, design rationales, code snapshots and commit‑derived knowledge, so you no longer have to re‑explain the same context.

Key capabilities

Feature	What you get
Memory Autopilot	A bounded “workset” JSON that bundles start files, current memory, and verification steps.
Observation Memory	Searchable facts, fixes, gotchas and session summaries scoped to the current Git repo.
Git Memory	Automatic ingestion of commits into searchable engineering facts (what changed, why it matters).
Reasoning Memory	Stores design decisions, alternatives and trade‑offs for future reference.
Code Memory / CodeGraph	Versioned code snapshots, symbol‑level indexing for TypeScript/JavaScript, and freshness checks.
Knowledge Workspace	Review‑gated Markdown pages and canonical workflows that never get overwritten silently.
Orchestration	Commands (memorix orchestrate, memorix lock, memorix team) to coordinate multi‑agent tasks, hand‑offs and verification loops.
Dashboard & HTTP API	A local web UI and optional HTTP endpoint for browsing memory, diagnostics and cleanup.
Bundled terminal agent – memcode	A ready‑to‑use CLI agent that reads/writes the same Memorix memory, useful for quick prototyping.
Sync	Optional privacy‑filtered sync to GitHub JSONL, S3, Postgres, etc., or multi‑device sharing via MCP.

How it works – When you run memorix init a ~/.memorix/config.toml (global) and/or <git‑root>/memorix.toml (project) are created. memorix setup --agent <name> then installs the appropriate plugin, hook or MCP configuration for the chosen agent. The core service (memorix serve) runs a stdio‑based MCP bridge; a separate memorix serve‑http can expose the same bridge over HTTP for IDEs or other clients.

Typical workflow

1. Install – npm install -g memorix.
2. Initialize – memorix init --global (creates defaults).
3. Connect an agent – e.g. memorix setup --agent claude --global.
4. Start coding – The agent can now query memorix memory or memorix codegraph to retrieve prior facts, and you can add new observations with memorix reasoning or memorix ingest commit.
5. Orchestrate – For larger tasks, run memorix orchestrate to hand off work between agents, lock files, and enforce verification steps.

Supported agents – The README lists dozens, including Claude Code, OpenAI Codex, GitHub Copilot CLI, Cursor, Gemini CLI, Pi, Kiro, Antigravity, DeepSeek Harness, OpenClaw, Hermes Agent, and any client that implements MCP.

Installation prerequisites – Node ≥ 22.18, Git (to locate the repository root), and optional SQLite FTS5/LanceDB for semantic indexing.

Where to learn more – The repo links to a website, a Chinese README, detailed integration docs (docs/INTEGRATIONS.md), and an active work tracker (ACTIVE_WORK.md).

---

All information above is taken directly from the project's README; no external assumptions have been added.