Give dassi a task and it does the work, in the web apps and accounts you're already signed into. Teach it a job you repeat and it keeps doing it — on your schedule, stopping for the approvals you set. Teach it a job once. It keeps doing it — you approve the big calls.
Walk it through the job once — in plain English.
Which sites, what schedule, what needs your OK.
It works on schedule while your browser is open. You handle only the exceptions.
Swipe →
dassi isn't a genie that already knows your job. It's closer to a new hire who's good at the web and needs one walkthrough of how you do things.
Describe the job in plain English, or walk dassi through it once while it watches. It reads the pages you're logged into, so there's no context to paste in and no integration to wire up.
Choose the sites it uses, what changes each time, when it runs, where the result goes, and which actions it has to ask you about first.
Trigger it yourself or let the schedule fire. dassi tells you when it's done — or when it hits something only you can decide. You handle the exceptions, not the job.
Not a gallery of everything possible. These are the shapes that keep coming back: something to reconcile, something to check, something to file — every week, inside apps only you can log into.
Claude, ChatGPT, and Gemini are excellent assistants, and each can drive a browser now. dassi does the job in the browser you already use, on the accounts you're already signed into, with the model you choose — and it still owns the job next Friday.
When dassi gets something wrong — the wrong supplier, the wrong column, the wrong tone — tell it. The correction goes into a memory you and dassi can both read and edit, so it carries into the next run instead of dying in the chat.
Most of what a browser agent spends goes on two things: re-reading pages it has already read, and working out how a site is put together every single time. dassi is built to stop paying for both.
Within a run, dassi keeps the job's context byte-stable, so every step after the first reads it from your model provider's cache — up to 90% off input tokens on Claude. You pay for that context once per run, not on every step.
As dassi works a site, it writes itself a small JavaScript library for that site. From then on it calls a function instead of re-deriving selectors from screenshots and HTML — the cheap part of the stack doing what the expensive part used to.
And dassi keeps tuning a job each time it runs, so work you repeat gets cheaper the longer you keep it.
Use Claude, GPT, Gemini, or another model inside dassi, with your own key or included credits. The model is the engine; the workflow is the thing you keep.
dassi runs in the Chrome, Edge, or Brave you already use, on the sites and accounts you're already signed into. There's nothing to integrate, no API to request, and no new browser to adopt. If you can do the job in a tab, dassi can be taught to do it.
Bring Claude, GPT, Gemini, Grok, DeepSeek, Kimi, or anything on OpenRouter — your own key or dassi's included credits, switchable anytime.
A workflow that runs without you only works if you trust where it runs and where it stops.
Learn more in our privacy policy.
Get updates on new features, integrations, and surprises. No spam, unsubscribe anytime.