A general coding agent that plans, executes, verifies, and recovers — for as long as the task takes. Message it from your browser, or drop into the terminal — same agent, either way.
$ npm install -g @sweet-cli/sweet (or use the CLI)
Generous free plan · no card required
The interaction, from your side.
sweet
What is Sweet?
A general coding agent that runs a full loop — plan, execute, verify, recover — for as long as the task takes. Any language, any part of the stack.
Built to run for hours, not suggest a token.
✓
Agent loop
Plan → execute → verify → recover, over and over, until the work is done.
✓
Self-verifying
Runs your tests and reads the failures before it calls it done.
✓
Self-recovering
Flaky test? Wrong assumption? It diagnoses, retries, and keeps going.
✓
Parallel subagents
Splits work across subagents that coordinate with each other.
✓
Any language
Node, Python, Go, Rust, Rails — plus ML, scripts, and more.
✓
/workfor
Set a horizon and walk away — it keeps working until the timer expires.
Fast. Cheap. Capable.
The old rule says pick two. Sweet breaks it — open-source models make it cheap, the autonomous loop makes it fast, and it verifies its own work before it ships.
Fast
Runs for hours, with subagents working in parallel.
Cheap
Open-source models — more usage for your dollar.
Capable
Near-frontier intelligence at the best price-to-performance ratio available.
Free plan · built independently · DS v4 Flash hosted in the US
Set a horizon. Walk away.
`/workfor 3h "build the payments integration"` — Sweet enters a loop: plan, execute, verify, recover. It self-nudges so it never idles, and reports everything it shipped when the timer runs out.
Questions, answered.
What makes Sweet different?
Autonomy over long horizons, subagents that coordinate with each other, and a persistent workspace that remembers your project across sessions. Plus it's free to start and runs on open-source models.
Does it work with my stack?
Yes — any repo, any stack, language-agnostic. Use it from the browser or the terminal; either way it reads and edits your actual files.
Is my code private?
In the web app, your code runs in an isolated, single-tenant container for your session — never shared with other users. In the CLI, Sweet runs locally and your code never leaves your machine.
What's free?
A free plan with usage limits. Pro ($20/mo) and Max ($200/mo) unlock more usage and the faster model.
Where does it run?
Model inference runs on DS v4 Flash (DeepSeek), hosted in the United States. Your code runs locally via the CLI, or in an isolated cloud container via the web app.
Start free. Upgrade when you need more.
Every plan runs through our relay — no API keys to manage. Usage is capped to your plan, so there are never surprise bills.
No credit card
Free
~25M tokens / day
$0/forever
Ship real tasks today. 25M tokens daily, resets every 24 hours.