Sweet is your team of AI agents in the cloud.
Give each agent a job. Each one works in its own sandbox, keeps its own notes, and asks you in the thread when it needs a decision. Pay only while they work.
An example team. Click the agents, try the Logs tab, and answer Website's question.
Built to keep working
Most coding agents work while you sit and watch. Sweet agents run in the cloud and keep going for hours without losing the thread.
Runs around the clock
Agents live in the cloud, not on your laptop. Close the tab or shut the lid, and they carry on for an hour, eight hours, or until you stop them.
Stays on track over long runs
Work is organised into separate conversations, and long histories are condensed into short, faithful summaries, so an agent hours into a job still knows what it's doing and why.
Remembers your project
Each agent keeps a brief of what you're building and notes on what it's done, in its own workspace, so tomorrow it picks up where it left off.
A team, not a session
Give each agent its own job and its own cloud computer. Up to five work at once, and they only stop to ask you when there's a real decision.
Manage every agent from one screen
Team shows what each agent is doing, which ones are waiting on you, and every conversation across them, newest first. Answer the one that needs you and get back to your day.
Team
What people hand off to their agents
“Upgrade our dependencies and fix whatever breaks.”
Bumps the versions, runs the test suite, fixes the failures, and opens a branch for review.
“This test fails one run in ten. Find out why.”
Reruns it until it reproduces, tracks down the cause, fixes it, and explains what was wrong.
“Add a waitlist form to the landing page.”
Builds the form, stores sign-ups, and adds the tests for both.
“Write tests for the payments module.”
Reads the code, covers the edge cases, and flags anything that looks like a real bug.
“Move this service to the new framework.”
Ports it piece by piece, runs the tests as it goes, and reports what changed.
“Go through this week's bug reports and fix the easy ones.”
Sorts them, fixes what it can on separate branches, and asks you about the rest.
Pay as you go
One balance covers model use and sandbox time, charged while your agents work. Idle agents sleep and cost nothing. You get a heads-up before you run low, and topping up takes one click from the chat.
Questions
Can it work on my existing code?
Yes. Give your agent a link to your repository and it clones it into its own sandbox. It has git and the GitHub command-line tool, so it can work on a branch and push it when you give it access.
Is my code safe?
Each agent works in its own isolated cloud sandbox, separate from other agents and other customers. It only reaches what you give it. See Security and our privacy policy.
What happens when my credit runs out?
Your agents pause and ask you to top up, right in the chat. There's no card on file until you add one and no subscription, so you're never charged by surprise.
Can I stop an agent partway through?
Any time. Press Esc to stop what it's doing, or turn off its working time so it only acts when you message it.
Do I need to be a developer?
It helps to know what you want built, but you talk to agents in plain English. They explain their plan before they start and ask you when there's a real decision to make.
How is this different from running an agent on my laptop?
Sweet agents run in the cloud, so they keep working with your laptop closed, several can work at once, and you can check in from your phone. You can still connect the command-line tool on your own computer if you like.
Give your first agent a job.
Four quick questions about what you're building, then it gets to work.