Skip to content
Mnemo the clownfish scholar

Remember everything.
Find anything. Mnemo — AI Memory System for Coding Agents. Four-Tier Architecture. Local-First. Private.

Mnemo captures every coding session, builds a living reef of knowledge, and lets you search it all. Your agent starts every session already knowing you.

$ mnemo --version
coming soon

Everything stays.
Nothing rots. Auto-Capture Sessions, Four-Tier Memory, Project Scoping, Local-First Privacy.

Mnemo captures every coding session automatically — conversations, decisions, discoveries. Four tiers of memory, from instant pattern matching to a living knowledge graph.

Auto-capture

Sessions recorded automatically. You code, it remembers.

Four depths

From instant pattern matching to a living knowledge graph. Each depth serves a different kind of memory. Always growing.

Project-scoped

Each project gets its own memory. Hierarchical scoping for monorepos. Git-boundary aware.

Local & private

All data on your machine. No cloud. No accounts. Your reef, your machine.

Ask in plain language.
Get the answer. Unified Search, Adaptive Ranking, Code Intelligence, Memory Collections.

Every memory is searchable. Ask how you implemented auth, what broke last week, or why a decision was made. Mnemo finds it across all four tiers.

Search everything at once

One command. All four depths. Conversations, decisions, code patterns. `mnemo recall` finds it.

Learns what matters

mnemo learns which memories matter to you. Every recall makes the next one better.

engram

Our code intelligence engine. Impact analysis, safe rename, symbol context. Your codebase has a memory too.

Memory pearls

Group related memories into collections. mnemo suggests what belongs together. Gather the pearls.

Your agent starts
every session smarter. Session Injection, Knowledge Extraction, 48 CLI Commands, 17 MCP Tools.

Mnemo injects relevant context at session start — past decisions, active problems, project patterns. Your agent begins with knowledge, not amnesia.

Session injection

Relevant memories surface automatically at session start. No manual lookup needed.

Knowledge extraction

Learnings, decisions, solutions surfaced from every session. Organized by how you think, not how databases work.

Dozens of commands

Everything you need. Recall, impact analysis, knowledge extraction, reef tending. One tool, full control.

Direct agent access

Claude Code and any MCP-compatible agent. Memory, search, and code intelligence right where your agent works.

From install to living memory
in four steps. Install, Capture, Recall, Grow — Four-Step AI Memory Setup.

One command to set up. Then your reef grows with every session.

1

Install

One command sets up everything. Hooks deploy automatically.

2

Capture

Code with your agent. Every session is recorded — conversations, decisions, discoveries.

3

Recall

Ask in plain language. Mnemo searches all four tiers and surfaces the answer.

4

Grow

Your reef gets smarter with every session. Knowledge compounds. Context deepens.

Fills the gap between you and your agent. Use Cases: Debugging, Onboarding, Refactoring, Knowledge Retrieval.

Here is what having memory actually looks like.

You are debugging a regression from three weeks ago.

Your agent says it has no context on that. mnemo does. Every session, every decision, every file change from that window is searchable. Find what broke and why.

A new team member needs to understand the auth system.

The decisions are in your reef. Why JWT. Why the middleware is ordered that way. Knowledge that used to live in someone's head, now preserved and searchable.

You are about to refactor a core module.

Before you touch anything, engram shows you every caller. Every dependency. Everything that breaks. Impact before action.

Your agent solved this last month. You cannot remember how.

mnemo can. One recall and it surfaces. Your past self already did the work. Stop re-solving problems you already solved.

Common questions about mnemo and how it works.

Everything you need to know about Mnemo and how your reef works.

What is mnemo?
mnemo is an archival intelligence companion. It captures every session, indexes every conversation, understands your codebase, and surfaces what matters. So your agent starts every session already knowing you. Four depths of memory. Local-first. Private.
How does the four-tier architecture work?
mnemo searches across four depths of memory. At the surface, instant pattern matching finds relevant files before you ask. Below that, deep semantic search finds meaning across every conversation. Then curated knowledge surfaces learnings, decisions, and patterns. At the deepest level, a living knowledge graph maps relationships across your entire body of work. Each depth serves a different kind of recall.
Which AI tools does mnemo support?
mnemo is CLI-first. It works with every AI coding agent. Deep integration with Claude Code and OpenCode. MCP server for direct agent access. Same power from the terminal, no agent required.
Is my data private?
Yes. Everything stays on your machine. SQLite for storage, Ollama for embeddings, optional cloud LLM for extraction only. No accounts. No cloud sync. Your reef, your machine.
Is mnemo free and open source?
Yes. Apache 2.0 licensed. Run it, modify it, extend it. No usage limits. No cloud service. It runs locally on your machine.
How is mnemo different from a vector database?
A vector database is one layer. mnemo adds instant pattern matching, curated knowledge extraction, a knowledge graph that connects entities across sessions, and adaptive ranking that learns from your usage. It is a memory system, not a search index.
Do I need to change how my agent works?
No. After mnemo init, hooks deploy automatically. Your agent works normally. mnemo captures in the background and injects context at session start. Zero friction.
Mnemo

The reef is growing. Mnemo — Open Source AI Agent Memory System. Coming soon.

Every session you don't capture is a memory lost to the deep. Be first to start your reef.

$ mnemo --version
coming soon