Beta Version 0.7.0 · experimental · macOS & Windows
Your models.
Your machine.
Chat, agents, cronjobs, documents, pictures and video in one window — running on hardware you own.
0.7.0 is an experimental beta · macOS signed & notarized · free for personal use · commercial licenses for companies
In one window
Everything happens here.
The chat in the middle, the panels beside it, the servers behind it. Nothing opens a second window, and nothing leaves your machine.
Fourteen events in one file. A hook can watch a tool, log it or stop it before it runs.
A real browser in the right column. The model reads the page while you watch.
Every working folder gets git: five tools, a history, and every repository of the machine as a 3D tree.
/goal keeps the model working round after round; its task list shows the plan and its agents.
Web App
Exe AI on your phone.
Open Exe AI in the phone’s browser and put it on the home screen. The models keep running on your computer; the phone shows the same chats, agents and cronjobs.

Images and Video
Pictures and video. Drawn here.
Steer every picture and every video from the chat or in the studio: the same models, the same values, nothing held back.
One click on a card downloads the model with every part it needs, and it starts with the maker’s own values.
The maker’s prompt upsampler rewrites your prompt before the picture is drawn. Optional, downloaded with the workflow.
The model picks the image model, the format, the LoRAs and the hires fix for every picture, and makes videos with length, references and sound.
Our own build of stable-diffusion.cpp with its own Metal kernels.
Every face in a picture is found and redrawn with its own LoRA and words.
Watch every picture form: on the canvas, on the chat tile and over Exe Link.
Servers
Six servers. Six local ports.
Each server has a page of its own, a log of its own and an honest memory reading before it starts. Any one of them can stand in for another: the agent server can clean up dictation while its own load is low.
- Model server llama.cpp · context, GPU layers, KV cache
- Agent server a model of its own for tasks
- Embedding server finds the passages that match a question
- Speech server dictation
- Clean-up server cleans up dictation, or stands in as a spare
- Image server stable-diffusion.cpp

Appearance
Six appearances, and one of your own.
Exe ships in warm grey; five more change the whole window while it runs, or you build a gradient of your own.
Privacy
Built for privacy.
Not a setting — the architecture.
- Binds to 127.0.0.1 chats, settings and models stay in local files
- Keys are never shown again a local file with tight permissions
- Memory stays home it reaches a cloud model only if you say so
- Tool output is data, not orders a web page cannot re-instruct your assistant
This is Exe AI.
Beta Version 0.7.0 · experimental · PolyForm Noncommercial 1.0.0 · free for personal use · commercial licenses for companies