minimi
A quiet memory that watches what you're doing on your Mac so you don't have to explain it to Claude every time — your open tabs, documents, and calls become context Claude already knows, without ever leaving your computer.
🔗 Visit minimi
Description
Every time you start a new conversation with an AI assistant, you usually have to re-explain what you're working on — paste in the doc, describe the meeting, remind it what project this is about. minimi's idea is to remove that re-explaining entirely: it quietly watches what's on your screen and stores it as memory, so Claude already has the context of what you've been reading, discussing, and working on.
minimi is a macOS app that passively captures open browser tabs, documents, calls, and Slack messages, storing everything as vector embeddings in a local, on-device database — nothing is uploaded to a cloud server — and feeds that context to Claude through a Model Context Protocol (MCP) connector. It uses Google's Gemini API specifically to generate the embeddings (a paid API call, though Google states it doesn't train on or retain that data) and Deepgram for meeting transcription, reporting 54% accuracy on the BEAM long-term-memory benchmark — a figure worth treating cautiously since there's no context given for how difficult that specific benchmark is. It's built by Shram Intelligence, Inc., published in July 2026, and pricing isn't published on the homepage, currently appearing to be a free Mac app.
💬 Our review
The short version: for a heavy Claude user on a Mac who's tired of manually re-briefing context at the start of every conversation, minimi's on-device ambient memory is solving a genuinely annoying friction point — and keeping the raw capture local (only embeddings, generated via Gemini, leave the device) is a meaningfully more privacy-conscious design than a cloud-first competitor would likely choose.
There isn't a well-established direct competitor doing exactly this (ambient, passive, on-device memory specifically for Claude) yet, since most "AI memory" products either work within a single chat app or require manual note-taking; minimi's bet is that automatic background capture beats manual context-feeding. The 54% BEAM benchmark figure is presented without enough surrounding context (dataset difficulty, comparison baseline) to know if that's a strong or mediocre result, and as a very new company (published July 2026) with no disclosed funding, it's worth treating as an early, promising tool rather than a proven one.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Application Mac gratuite au moment de la rédaction ; aucune tarification publiée
Pros
Capture passive automatique — plus besoin de rebriefer Claude manuellement
Stockage 100% local en embeddings, rien n'est envoyé au cloud (sauf appel Gemini pour générer les embeddings)
Intégration native via MCP directement dans Claude
Gratuit au moment de la rédaction
Cons
Score de 54% sur le benchmark BEAM sans contexte de difficulté pour l'interpréter
Entreprise très jeune (juillet 2026), aucun financement divulgué
Réservé à macOS uniquement
Dépend d'un appel API Gemini payant pour générer les embeddings