A lean, local-first interface for capturing thoughts, recalling anything, and managing todos. All AI runs on your machine via Ollama. No cloud, no data harvesting.
Choose your platform, grab the latest release, then connect Min to your local Ollama runtime. Min checks the engine on first launch, reconnects automatically, and guides you to Settings when models are missing.
Recommended for everyday use. Includes the desktop installer and MSI package for standard deployment.
↓ Open latest releaseAvailable for Apple Silicon and Intel builds from the latest release page.
↓ View macOS buildsGrab the latest AppImage or DEB package and keep everything local on your own machine.
↓ View Linux buildsPick the latest desktop build for your OS from GitHub Releases.
Install and start Ollama locally. Min does not bundle the engine yet; it verifies that your local runtime is reachable.
Open Min, download a chat model and an embedding model from Settings, and let Min recommend the best fit for your hardware.
Global shortcut pops up a 416px chat window anywhere. Thoughts are stored in SQLite with auto-generated semantic tags.
Ctrl+Shift+SpaceAsk in plain language. Min embeds your query, searches your local vectors, and answers using only your own stored data.
RAG_PIPELINE: ACTIVECreate, list, complete, update and delete tasks with natural language. No forms, no checkboxes, just tell Min what to do.
TYPE: TODOWhen you save something, Min finds related notes via embedding similarity (threshold 0.7) and surfaces connections you missed.
COSINE_SIM: > 0.70Type /digest for an AI-generated summary of today's thoughts, meetings, notes and todos — grouped by themes.
CMD: /DIGESTCtrl+Shift+T opens a minimal 480×72px floating input. Save a thought and it auto-closes. Zero interruption to your flow.
Ctrl+Shift+TMin now recommends models dynamically inside Settings based on available RAM and local acceleration. The site shows the profiles Min can suggest, not a single hardcoded winner.
Downloads happen inside Min, where you can track progress, cancel pulls, and switch chat or embedding models safely.
Every message is classified into one of 5 intents and routed to a specialized agent. Each agent uses a dedicated skill prompt loaded from skills/*.md.
Splits input into storable items (thought, todo, note, meeting) with auto-generated tags. Preserves exact user wording.
Answers only from your stored data via RAG. Never uses training knowledge. Embeds query, searches vectors, augments prompt.
Resolves delete/update by matching natural language to stored document IDs. Asks for clarification when ambiguous.
Stores user behavior preferences and instructions that shape future responses and agent behavior.
Handles greetings, help, and casual chat. Explains Min's capabilities without executing storage actions.
Most AI tools pipe your thoughts through remote servers. Min is different. All inference runs locally via Ollama. All data lives in a single SQLite database on your machine. We have no servers, no telemetry, no analytics. Your intellectual property stays yours.
Free, open source, forever local.
Available for Windows, macOS, and Linux.