Memorix

What Memorix does

Memorix keeps searchable project memory under the Git project instead of inside one chat window, so a new session, IDE or agent can pick up what the last one learned. SQLite is the canonical store. Small projects use an in-process Orama path, while larger ones use a persistent SQLite FTS5 candidate index and optionally a local LanceDB semantic index. LLM-backed formation and embedding remain optional.

Capabilities include Memory Autopilot task worksets, observation memory for facts and gotchas, curated long-term memory, code state and code memory, Git Memory that turns commits into searchable facts, and Reasoning Memory that records why choices were made. A memorix orchestrate command coordinates handoffs, locks and review loops for parallel agent work. The README says it is listed in the official MCP Registry and names Claude Code, Codex, Cursor, OpenClaw, Hermes Agent and others as supported.

Key features

• Project-scoped shared memory kept under the Git project
• Git Memory that turns commits into searchable engineering facts
• Reasoning Memory that stores decisions with alternatives and trade-offs
• SQLite canonical store with an optional local LanceDB semantic index
• memorix orchestrate for task context, handoffs, locks and review loops
• Access through MCP, hooks, CLI, SDK and a bundled terminal agent

When to use it

• Keeping project context when switching between Hermes Agent, Claude Code and Cursor
• Recalling why an architecture decision was made without digging through old chats
• Coordinating several agents on parallel work in one repository

Who it is for: Developers who use several coding agents on the same project and are tired of re-explaining it each session.

How it fits with Hermes Agent

Hermes Agent is one of the many agents Memorix lists as compatible. It is a general memory layer reached through MCP and is not built only for Hermes Agent.

FAQ