Sweet runs autonomously in your terminal — reading your codebase, making a plan, editing files, running commands, and delivering working code. Here's what it's built to handle.
Give Sweet a goal. It makes technical decisions, plans the work, and implements it without constant guidance.
Instantly understand and navigate any codebase — structure, patterns, and relevant files — before touching anything.
Spawn multiple agents to work simultaneously on different parts of the same task instead of running everything sequentially.
Break up god classes, migrate schemas, or modernize legacy modules — with checkpoints and tests at each step.
Point Sweet at a failing test or error log. It traces the issue through your codebase and writes the fix with a regression test.
Set Sweet to work for a defined time window. It queues and executes tasks in sequence, then surfaces a summary when done.
sweet workfor 2h "clear the backlog"See your token usage and cost for every task. Open source models. No hidden fees. No markup behind subscriptions.
Your code runs on your machine. We don't store it, we don't train on it. Nothing leaves your environment except the prompt.
No subscription. Open source models. Your first task runs in under 2 minutes.