Follows your cursor
His eyes track your mouse as it moves around the screen.
Mellow is a desktop companion for Windows. Ask him a question out loud and he answers by voice, shows you where to click, writes into any text field and takes notes in your meetings.
Free and open source. Windows 10 and 11.
Bring your own key, or run it all locally.
- Claude Code
- Codex
- Ollama
- OpenRouter
- NVIDIA NIM
- Gemini
Hold a shortcut and ask.
Press and hold Ctrl + Shift + Space, speak, and release. Mellow answers out loud. He can also open apps and websites, play media and adjust the volume of a single app.
Try it here: press Ctrl + Shift + Space, or tap the keys.
- “Open YouTube and play Faded”
- “Where's the export button?”
- “Reduce the volume by 50%”
He behaves like a real pet.
Between tasks, Mellow lives on your desktop. He reacts to your cursor, enjoys attention and has a few habits of his own.
Enjoys being petted
Move your cursor over his head and he responds with hearts.
Hunt mode
Move the cursor quickly and he chases it across the screen.
Dislikes being shaken
Drag him around too roughly and he lets you know.
Peek mode
When you need the space, he tucks into the edge of the screen and peeks out.
Stretches and yawns
Now and then he takes a moment to stretch and yawn.
Ask where something is, and he points to it.
When you cannot find a setting or a button, ask Mellow. He looks at the window you are using, moves his bone pointer to the control and explains what it does.
- Where do I click to stop the recording?
Hold Ctrl + Shift + Space and ask your question in your own words.
- Looking at your screen...
Mellow takes a single look at the active window, and only because you asked.
- That is the Stop Recording button.
The pointer stays on the control until he finishes explaining.
- Where do I click to stop the recording?
Hold Ctrl + Shift + Space and ask your question in your own words.
- Looking at your screen...
Mellow takes a single look at the active window, and only because you asked.
- That is the Stop Recording button.
The pointer stays on the control until he finishes explaining.
Pointing is a guide. Very small icons or crowded layouts can reduce its accuracy.
Write anywhere with your voice.
Screen-aware writing turns your voice into text in any application. Place your cursor in a text field, hold Ctrl + Shift + Space and say what you want written. Mellow types it where your cursor is. It works in documents, email, chat apps and terminals, and you can turn it on in Settings under Screen-aware writing.
Dictate
Speak the words you want and they are typed exactly as you said them.
Describe
Say what the text should say, such as "a reply saying I'll be twenty minutes late", and Mellow writes it for you. When the request depends on what is on screen, like the email you are answering, he reads it first.
Revise
Not quite right? Ask again before you edit the text and Mellow rewrites what he wrote instead of starting over.
You say: “Reply saying I'll be twenty minutes late”
Mellow is writing...
Focus sessions and reminders.
Mellow includes a Pomodoro timer and reminders, so the tools that keep your day on track sit on your desktop instead of in another app. Both work in every engine mode, including Just the pet.
Pomodoro timer
Work in focused rounds with short breaks in between. Set the focus length from 5 to 60 minutes, the break from 1 to 30 minutes, and up to 8 rounds. The default is four 25 minute rounds with 5 minute breaks. The countdown floats above Mellow, and he lets you know when it is time to take a break or get back to work.
Reminders
Set a reminder for a specific time, once or every day, such as drinking water, stretching or returning a call. When it is due, Mellow says it out loud and shows it in a speech bubble. Reminders are saved, so they still arrive after a restart.
Start either one from Mellow's menu, or just ask: "Start a focus timer" or "Remind me at 3 to call Sam".
Meeting transcripts and notes.
Start a transcription and Mellow puts on his glasses and writes while you take part in the conversation. When the meeting ends, you get the full transcript and a structured set of notes.
Both sides of the call
Mellow transcribes your microphone and the meeting audio together, either on your computer or with your speech provider.
Speaker labels
Your side is labelled You. Remote voices can be separated into Speaker 1, Speaker 2 and so on, and renamed afterwards. This feature is experimental and off by default.
Structured notes
The engine you chose writes an overview, key topics, decisions, action items and open questions.
Export
Save the transcript or the notes as Markdown, plain text or JSON, or copy them to the clipboard.
He remembers what matters to you.
With memory turned on, Mellow keeps a short summary of what you have told him, such as your name, your work or your preferences, and uses it to give more relevant answers.
Tell him directly
Say "Remember that I'm allergic to peanuts" and it is used from the next answer onwards.
Or let him learn
After a conversation ends, Mellow saves the lasting details you mentioned along the way.
Read and edit it
The full summary is in Settings. You can change or delete any line, or turn memory off entirely.
Got it. No peanuts, Priya.
Memory summary
Updated just nowWork and study
You are writing a thesis on multimodal agents, and your defense is in May.
Food and health
You are vegan and allergic to peanuts.
How you like answers
Short answers with the steps first.
Just told Mellow: I'm allergic to peanutsAlready used in answers, added to the summary shortly.
Make him yours.
Choose a coat that suits your wallpaper. Pick one of the presets or set the fur, patch and blush colours yourself. The shading adjusts automatically so his face stays clear.
Choose how he thinks.
Mellow can run entirely on your computer, use a cloud provider, use an AI subscription you already have, or run with no AI at all. You choose during setup and can change it at any time in Settings.
Everything on your computer
Answers come from any model you run with Ollama. Speech recognition uses Parakeet or Whisper and the voice uses Kokoro. No API key is needed and nothing leaves your machine.
The local speech models are about 1 GB in total and are downloaded once.
Private by design.
Screen access only on request
Mellow looks at your screen only when a question needs it. Nothing is recorded in the background, and screen reading can be turned off.
Audio stays in memory
Meetings keep transcripts and notes, never recordings. Speaker labelling runs on your computer and forgets voices when the meeting ends.
Keys stay on your computer
Provider keys are stored in Mellow's settings on your machine and are never displayed in full.
Open source
The complete source code is on GitHub under the Apache-2.0 license, so you can read it, build it and modify it.
Questions, answered.
Give him a corner of your screen.
Free and open source, for Windows 10 and 11.