AUTONOMOUS CODING AGENT · CLI + WEB APP

Hand off the backlog. Get back pull requests.

Sweet reads your repo, makes the change, runs your tests, and hands you a diff to review. Use it from your terminal, or from the web app, where every session runs in its own isolated microVM.

Start free → Team pilot
Free plan, no cardUsage hard-capped, no surprise billsCancel in one click
Prefer the terminal? $ npm install -g @sweet-cli/sweet
REAL OUTPUT
Built by Sweet. Try them yourself.
Each of these was built by the agent from a one-line prompt, then published as is. Open them, use them, and read how each one was made.
Diff viewer comparing two React components side by side

A visual diff viewer

Paste two versions of a file and see character-level changes. Runs entirely in the browser.

Open the app →
 3D double-slit experiment simulation with controls

A 3D physics simulation

An interactive double-slit experiment in Three.js, with wave and particle modes.

Open the app →
 GitHub Wrapped start screen asking for a username

GitHub Wrapped

Enter a GitHub username and get a slide-by-slide story of a year in code.

Open the app →
How they were built: diff viewer · double-slit · GitHub Wrapped
WHERE IT RUNS
Two ways to run it. Same agent.
CLI

In your terminal

  • Runs locally on your machine, in your repo
  • Uses your git, test runner, and package manager
  • Works alongside any editor
npm install -g @sweet-cli/sweet
Web app

In your browser

  • Each session gets its own Firecracker microVM (4 vCPU, 8GB)
  • Unprivileged and egress-restricted, never shared between users
  • Keeps working with /workfor after you close the tab
Open the web app →
Either way, the task context is sent to our model provider for inference (DeepSeek V4 Flash, hosted in the US). The Trust Center lists every sub-processor.
WHY SWEET
Built to finish tasks, not suggest lines.
✓

Checks its own work

Runs your tests and reads the failures before it calls a task done.

✓

Recovers on its own

Flaky test or wrong assumption? It diagnoses, retries, and keeps going.

✓

Parallel subagents

Splits bigger jobs across subagents that coordinate with each other.

✓

Any language

Node, Python, Go, Rust, Rails, plus scripts and data work.

✓

Fits your tools

Plain git and your existing CI. No index to build, no new workflow to learn.

✓

Priced to run all day

An efficient open-weight model keeps long autonomous runs affordable.

GO HANDS-OFF
Set a horizon. Walk away.
Type /workfor 3h "build the payments integration" and Sweet keeps working (plan, change, test, fix) until the time is up, then reports everything it did. In the web app it keeps going after you close the tab.
SHIPPING WEEKLY
Latest updates.
Aug 27, 2026
Web app

Bigger sessions, real dev work.

Every session now runs on 4 vCPUs and 8GB of RAM — enough headroom for a full build, not just a script.
Aug 24, 2026
Web app

Download a whole directory, not just one file.

The files menu now zips and downloads an entire folder in one click.
Aug 21, 2026
Web app

A quieter chat view.

Tool calls collapse to a single line by default, and the reasoning trace lives behind its own dropdown — expand only what you want to inspect.
Aug 10, 2026
CLI + Web app

/workfor — set a horizon and walk away.

Give it a time budget and a goal. Sweet loops on its own — plan, execute, verify, recover — until the timer runs out.
PRICING
Start free. Upgrade when it's earning its keep.
No API keys to manage. Every plan has a hard usage cap, so your bill is always the price on this page.
No credit card
Free
~27M tokens / month
$0/forever

Ship real tasks today. ~27M tokens per month. No card required.

  • Full agent, all languages
  • ~27M tokens per month
  • Resets each month
  • /workfor autonomous runs
  • No API keys needed
Start free
Most popular
Pro
~108M tokens / month
$20/mo

For developers who ship with Sweet daily. Unlimited /workfor loops and priority throughput.

  • Everything in Free
  • ~108M tokens monthly
  • Unlimited /workfor loops
  • Priority throughput
  • Cancel in one click
Get Pro
Power
Max
~1.09B tokens / month
$200/mo

For heavy users who run Sweet around the clock.

  • Everything in Pro
  • ~1.09B tokens monthly
  • 10× Pro capacity
  • Unlimited /workfor loops
  • Highest throughput priority
  • Cancel in one click
Get Max
FOR TEAMS

Put agents on your team's backlog.

One flat price for the whole team. Start with a 30-day pilot: we agree on a success metric, onboard you over a shared Slack channel, and the pilot fee counts toward your first two months.

Start a pilotSee the Team plan
  • Up to 50 concurrent agents
  • Unlimited seats
  • Admin and usage dashboard
  • Onboarding and a shared Slack channel
$1,000/mo · pilot $1,500 for 30 days
FAQ
Questions, answered.
How do I get started?
Sign in with Google, GitHub, or email (no password) on the web app and pick a plan — the free tier needs no card. Prefer the terminal? npm install -g @sweet-cli/sweet, then sweet login.
What kinds of tasks can Sweet handle?
Everything from routine bug fixes and test-writing to full features and multi-file refactors. It reads your repo and plans before it touches anything, and keeps going — verifying and recovering — for as long as the task takes.
How does Sweet work with my existing tools?
It runs in your terminal, or in an isolated container behind the web app, and uses the command line tools you already have — git, your test runner, your package manager. Any repo, any stack, language-agnostic.
Is my code private?
In the CLI, Sweet runs on your machine and your code stays there, apart from the context sent to the model for each step. In the web app, each session runs in its own isolated microVM that's never shared with other users. In both cases, task context goes to our US-hosted model provider for inference; the Trust Center lists every sub-processor.
What are the requirements to run Sweet?
The web app just needs a browser. The CLI installs via npm and runs on macOS, Linux, and Windows (via WSL).
How much does Sweet cost?
Free gives you about 27M tokens a month with no card. Pro is $20/mo (about 108M tokens) and Max is $200/mo (about 1.09B tokens). Teams use the Team plan at $1,000/mo for up to 50 agents and unlimited seats. Every plan is hard-capped, so you never pay more than the listed price.
Can I cancel anytime?
Yes. Plans are monthly and you can cancel in one click from the billing page. Payments are handled by Stripe; we never see your card details.
Do you have a team plan?
Yes — Team puts up to 50 concurrent autonomous agents on your backlog with unlimited seats, an admin + usage dashboard, and onboarding, for $1,000/mo (about 8% of one senior engineer). There’s also a 30-day paid pilot. See Team →
Where does my code run?
The CLI runs locally on your machine — git, tests, and your tools execute right there, so the agent runs locally on your machine. The web app runs each session in its own isolated Firecracker microVM on Sweet's infrastructure (unprivileged, egress-restricted, one per session). More on security →
GET THE TECHNICAL RUNDOWN
Getting started Commands reference Configuration API reference Workflows Troubleshooting

Give it a task from your backlog.

Free plan, no card. Start in the browser, or install the CLI.

Start free →
$ npm install -g @sweet-cli/sweet  &&  $ sweet start