Add CSV export to the reports page and push a branch for review.
agent/csv-exportReady for review
Delegate real work to agents.
Agents with their own computers, for your whole team. Any model from any lab, clear controls, built-in security, and no per-seat fees.
We're onboarding teams one at a time. Create your account and we'll reach out to get you set up.
Product demo: a teammate asks Donvi to add CSV export to the reports page. The agent starts its own computer, clones the repository, runs the tests, checks the new button in a browser, and pushes a branch. Pushes to main are blocked, so the branch waits for review. The session ends ready for review, with the model and cost shown.
- Own computer for every agent
- Models from OpenAI, Anthropic, and more
- Code reaches main only through your review
- One switch stops every agent
- Agent computers run in the EU
Hand it off. Get it back done.
Find out why the signup page takes four seconds to load, and fix it.
agent/signup-speed4.1s → 0.9s
Upgrade us to the latest React and fix whatever breaks.
agent/react-upgrade214 tests passing
Write tests for the payments module and fix any bugs they turn up.
38 tests added2 bugs fixed
Open staging in a browser, click through checkout, and report anything broken.
Report3 issues, with screenshots
Scheduled
Every night at 2, run the full test suite on main and tell me what broke.
Nightly at 02:00Last run: all green
Review yesterday's commits and flag anything risky.
2 commits flaggedWith reasons
Agents that do the work
Every agent gets its own computer.
Donvi agents don't stop at answering questions. Each one gets its own cloud computer with a browser, a terminal, and files, and uses them to finish the job. Watch it work live, or close your laptop and come back to the result.
Works while you're away.
Tasks run in the cloud and keep going with your laptop closed.
Runs on a schedule.
Set up recurring work once: nightly test runs, weekly research, monthly reports.
Available anywhere.
Start a task at your desk and check on it from your phone.
We handle the hard parts.
Machines, model updates, and tuning agents for speed and reliability are on us. Your team gets the results.
For engineering teams
Clear the backlog your team never gets to.
Bug fixes, upgrades, tests, features. Hand them to agents and get each one back as a branch, ready for review. Your developers spend their time on the work only they can do.
Illustration: before leaving at 18:00, the team hands three tasks to agents. Overnight each agent pushes its own branch and marks it ready for review. In the morning, people on the team review and merge them into main. Main only changes when a person merges.
A real development environment. Agents clone your repository, install dependencies, run the tests, and check the result in a browser, all on their own computer.
Branches, not surprises. Every change arrives on its own branch. Nothing reaches main until someone on your team merges it.
Overnight work. Start a few tasks before you leave and review the branches in the morning.
In control, by design
Agents only do what you allow.
Your agents never touch your systems directly. Everything they do outside their own computer passes through Donvi's gate, which checks it against the permissions you've set. If you haven't allowed it, it doesn't happen.
Illustration: an agent's requests pass through Donvi's gate. Cloning acme/dashboard and pushing the branch agent/csv-export are allowed, and the gate adds the GitHub key only after the check. A push to main and a clone of a repository that isn't allowed are blocked. Every request is written to the session record.
You decide what's allowed. Choose which repositories agents can see and which they can push to.
A person merges to main. Agents push to their own branches. Pushes to main are always blocked, so code only reaches main when someone on your team merges it.
Keys stay behind the gate. Agents never see your passwords, tokens, or keys. The gate uses them on the agent's behalf, within your rules.
Every push is checked before it leaves. If a push can't be checked, it's rejected.
Every session on record. See who started each task and every step the agent took: the commands it ran, the tools it used, and what it pushed.
Your data stays your team's. Sessions, files, and connections are private to your team, and every session gets its own computer that's never shared.
The kill switch
One switch. Every agent stops.
If anything ever looks wrong, your team's owner flips one switch. Every running agent stops on the spot. Scheduled tasks stop. No new work can start and no model can be called until the owner turns it back on. It's a hard stop, not a polite request.
AGENTSONOFF
RUNNINGSTOPPED
OWNER ONLYLift the cover to reach the switch.
Stops every running agent immediately.
Stops all scheduled tasks.
Blocks new sessions and model requests.
Only the owner holds the switch, so it can't be flipped by accident, and only the owner can turn agents back on.
The best models, no lock-in
The right model for every job. Not just one lab's.
AI labs' team plans lock you into their own models. Donvi gives your team the best models from OpenAI and Anthropic, plus fast, low-cost open models, and lets you pick one for each task. Use the most capable model for hard problems and the cheapest for routine work.
New models as they launch. When a better model ships, your team gets it. No new contract, no migration.
Switch any time, even mid-conversation.
Nothing to migrate. Your skills, history, and scheduled tasks stay put when you change models.
Costs you can see and control
Pay for work, not seats.
No per-seat licenses and no monthly subscription. Your team shares a prepaid balance and pays only for the work agents actually do. Bring in the whole team at no extra cost.
Team size
3 people
No surprise bills. You set the budget up front. When it runs out, new work stops until you top up.
See where the money goes. Daily spend and cost per model, so you know what each kind of work costs.
Cheaper by design. Put routine work on low-cost models. Donvi routes and caches requests to keep costs down.
Built for teams
Teach it once. Your whole team benefits.
Skills capture how your team works: your coding conventions, your release checklist, your report format. Share a skill with the team and every agent follows it, for everyone. Keep a skill private when it's just for you.
release-checklist
- 01Run the full test suite on main
- 02Update CHANGELOG.md
- 03Tag the release and push the tag
- 04Post the notes in #releases
- MLYourelease-checklistchecklistfollows this skill
- AKAlexrelease-checklistchecklistdoes not use this skill
- PRPriyarelease-checklistchecklistdoes not use this skill
- SOSamrelease-checklistchecklistdoes not use this skill
How it works
01
Connect GitHub.
Choose which repositories agents can see and which they can push to.
02
Hand off the task.
Describe it in plain language, or schedule it to run on its own.
03
Review the result.
A branch, a report, a chart. Every step is on record, and nothing reaches main without you.
Coming soon
- Coming soon
Developer API. Start and manage agents from your own code and systems.
- Coming soon
Claude Code and Codex. Developers keep using the tools they know on their own machines. Every call routes through Donvi, so it counts toward the same budget and shows up in the same usage view.
- Coming soon
EU-only inference and zero data retention. Keep model requests inside the EU, with no data retained by model providers. Donvi's servers and your agents' computers already run in the EU.
Simple pricing. No seats.
Models and computers at cost, plus a flat 10% platform fee. One prepaid balance covers it all: model usage and the time agents spend on their computers. No per-seat fees, no monthly subscription, and you can see exactly where the money goes.
Model usageat costComputer timeat costPlatform fee10%
- Model usageat cost
- Computer timeat cost
- Platform fee10%
Early access. We're onboarding our first teams personally. Create your account and we'll be in touch.
Get early access
or email Jonas, Donvi's founder, at jonas@neuston.io
Frequently asked questions
Those plans tie you to one AI lab, and most charge per seat. Donvi gives you the best models from several labs in one place and charges only for the work done. Every agent gets its own cloud computer, and your whole team's access, budget, and records are managed in one place.
The best models from OpenAI and Anthropic, plus fast, low-cost open models. New models are added as they launch. Pick a different one for each task, or switch mid-conversation.
Agents work on their own cloud computers, not on your machines. They only reach the repositories you allow, they can never push to main, and every push is checked first. If something looks wrong, one switch stops every agent.
No. Credentials stay behind Donvi's gate. When an agent works with GitHub, the gate makes the request on its behalf, so the credential never reaches the agent's computer or conversation.
Models and computers at cost, plus a flat 10% platform fee. No seats and no subscription. You buy credits up front and spend them as agents work.
New work stops until you add credits. There's no bill at the end of the month, because you pay up front.
It stops every running agent, stops scheduled tasks, and blocks new sessions and model requests until your team's owner turns it back on.
Donvi's servers, your team's data, and your agents' computers run in the EU. Model requests go to the providers that serve each model, and some of them are outside the EU. EU-only inference and zero data retention are coming soon.
Yes. Donvi works in any browser. Add it to your home screen to get an app icon and a notification when an agent finishes.
No. We run the computers, the models, and the connections. Your team signs in and gets to work.
Yes. Email Jonas, Donvi's founder, at jonas@neuston.io.
Give your team agents that do real work.
Donvi is in early access. Create your account and we'll reach out to set up your team.
Prefer to talk first? Email Jonas, Donvi's founder, at jonas@neuston.io.