Local-first. Open source. No cloud.

Say what you need.
Your computer does the rest.

IntentOS turns one plain sentence into real actions in the browser, using a language model that runs on your own machine. No API keys. No clicking through apps.

Try the console Install it
The IntentOS symbol, a faceted crystal in neon blue and purple

Intent console

Checking for an engine

Nothing runs until you approve it.

One sentence, three stations.

  1. The brain understands

    A language model in Ollama reads your sentence and returns a structured JSON plan. Invalid output goes back to the model for correction.

  2. The gatekeeper checks

    Code rates the risk from the actions themselves. If something essential is missing, the engine asks you instead of guessing.

  3. The hands execute

    Playwright runs the steps in a real browser, stops at the first failure, and hands you the result.

Autonomy ends where your approval begins.

You approve every plan
In the console, nothing runs until you have seen the steps and pressed approve.
The model can't downgrade risk
It can raise the risk level the code calculated. It can never lower it.
Your logins stay out of it
The browser runs in a clean, temporary profile, without your passwords or cookies.
Your sentences stay with you
The model runs on your machine. Sites only see what the plan you approved sends them.

Running in three commands.

  1. Pull a model, once Ollama is installed

    ollama pull llama3

  2. Install the dependencies from the project folder

    python -m pip install -r requirements.txt

  3. Start it, then open localhost:8765

    python -m intentos.web