TRUSTED BY FOLKS AT
1.
Download app, sign in, and minimi icon appears in your tray
2.
Copy the MCP link and paste it in your LLM's custom connector
3.
Start a new chat with your LLM and it will use minimi to answer
Let Cotton plug your memory into any LLM you like and it works.
ChatGPT
Claude
Your memory lives on your Mac, where it's safest
50% more accurate than previous SOTA*
LIGHT (BEAM)
36%
Minimi
54%
*Measured on BEAM, ICLR 2026 benchmark for long-term memory
Same permissions as WisprFlow and Granola
Stored on your Mac
Your memories are stored as embeddings in a local vector database on your Mac. Nothing is uploaded to the cloud.
Embeddings via Gemini
Your memories are turned into embeddings using Gemini's API, on a paid plan. Google doesn't train on or retain this data.
Encrypted in transit
Your memory is decrypted only for the moment it's processed by the LLM, then re-encrypted on the way back to your device.
You can monitor your device's API calls yourself. They go to our backend, which relays to Gemini and to Deepgram. SOC2 is on our roadmap.
Skip the extra prompting and get straight to the answer.
