Loading
Capture microphone and the other side of the call. Whisper transcribes on-device, diarization assigns each speaker a name, your local model writes the summary, and you can chat with the transcript like a document. Speaker accuracy improves over time - Typilot remembers voices across meetings.
Select any text in any app, press Polish - replaced in place with your tone. Or pause mid-sentence and let Typilot suggest the rest. Tab to accept.
Hold a key, talk. Filler words like "umm" and "like" are stripped, punctuation added, sentences capitalized. Speak the way you think - Typilot writes it down clean.
Write your own command - a name, a prefix, a system prompt, a shortcut, a model. As many as you want. Power-user muscle memory, not a fixed feature set.
Record once, walk away. Typilot transcribes in the background, then auto-tags each note by type - meeting, action, idea - and groups them by date and project. Searchable in seconds.
Every model runs locally via Ollama. No telemetry, no analytics, no API keys, no account required to type. Works in airplane mode, on a train, in a vault.
A full chat surface for the times the inline-typing model is not the right shape. Conversations grouped by command. Switch commands mid-thread. Markdown rendering, code blocks, copyable replies.
Full-screen or cursor-region capture. Built-in OCR reads text out of the image and feeds it to your prompt. Prefix any command with img: to attach screen context.
See what you actually do with AI. Total commands, weekly trend, model breakdown, favorite shortcuts. Every prompt and response stored locally - search, filter, copy, replay.
Tray icon with quick toggles. Start at login. Minimize to tray. Single-instance launch. Platform-native audio capture - Core Audio on macOS, WinAPI on Windows, PulseAudio/ALSA on Linux.
Llama, Mistral, Phi, Qwen, your own fine-tune - pull from Ollama, point at a custom endpoint, or both. Different model per command if you want.
Try it once. Decide later.