m3 Memory
Local-first memory layer that stores facts and history for AI assistants to recall.
Tested on Linux
Use it
Linux x86_64
You need first
- Runtimes:
Python 3.12+
Install m3-memory via uv tool
uv tool install m3-memory
Check that it works
m3 --versionThe output should contain “m3-memory 2026.10.25.0”.
First thing to try
m3 doctor 2>&1Run m3 doctor first to check that the setup works.
What it does
- Remember projects and preferences across sessionsTested
How to use it
- Save a memory with
m3 memory memory_write --type decision --title "<title>" --content "<your text>". - Look for the "Created:" line with the new memory's ID.
- Save a memory with
- Find memories without using the exact wordingTested
How to use it
- Ask in your own words with
m3 memory memory_search --query "<your question>" --k 1. - Read the matching memory in the "content" field of the result.
- Ask in your own words with
- Share one memory across AI assistantsNot tested
- Save and recall conversation historyNot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 2 of 4 features work
- Works offline
macOS and Windows: the project says it runs there; pickfit hasn’t tried it.
Good for
- remember facts from earlier sessions
- recall past conversations
- share memory across AI assistants
- find memories by meaning
- share notes and handoffs between agents
An alternative to Mem0, Letta and Zep.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
Its main function works
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 2026.10.25.0
Differences the test machine hit
- README: local BGE-M3 embedder provides semantic search
In the test: m3 works via keyword fallback with EmbedFallbackError/EmbedPrimaryError warnings; no embedder setup needed for basic write/search - README: m3 doctor checks that everything is working
In the test: m3 doctor exits 1 reporting warnings (no embedder, no agents, no LLM) but store and CLI remain usable - README: chat-log capture keeps conversations and finds them again
In the test: conversation capture/recall uses m3 conversations conversation_start/conversation_append/conversation_search; m3 chatlog only has init/status/doctor/hook-path