AI coding agents.
10x cheaper.

Sweet runs parallel coding agents on open source models. Same tasks as Claude Code and Codex. A fraction of the cost. You always know what model you're using and what you're paying.

Start for $5 Open Web App

sweet — building auth system

$ sweet start "build a JWT auth system with login, register, and protected routes"

⧗ planning task...
↳ spawning 4 agents in parallel

agent-1 building auth middleware and JWT utils agent-2 creating user model and database schema agent-3 writing login and register API routes agent-4 generating test suite for all endpoints
agent-1 ✓ complete — auth/middleware.ts, auth/jwt.ts agent-3 ✓ complete — routes/login.ts, routes/register.ts agent-2 ✓ complete — models/user.ts, migrations/001_users.sql agent-4 ✓ complete — tests/auth.test.ts (12 tests passing) ✓ task complete — 8 files, 847 lines, 12 tests passing time: 6m 12s · tokens: 284k · cost: $0.41

4 agents built a full auth system in parallel for $0.41

Real cost comparison

Same task. Different price.

	Sweet	Claude Code	Cursor
Auth system with tests	$0.41	$4.80	$3.20 + sub
REST API (CRUD + validation)	$0.28	$3.50	$2.10 + sub
React dashboard component	$0.52	$5.20	$4.00 + sub
Bug fix + regression tests	$0.15	$1.80	$1.40 + sub
Monthly bill (daily use)	$18–35	$150–400	$40 + overages

Costs based on real usage with DeepSeek v3.2 through Fireworks AI. Claude Code costs based on Anthropic API pricing. Cursor costs include $20/mo Pro subscription plus token usage.

Get started

Running in 2 minutes.

Open the web app or install the CLI

Use Sweet instantly in your browser at sweetcli.com/web or install with npm i -g @sweet-cli/sweet

Add balance

Start with $5. No subscription. No commitment. Your balance works across web app and CLI.

Describe your task

Tell Sweet what to build. It plans the work, spawns parallel agents, and delivers working code with tests.

Under the hood

Open source. Transparent pricing.

Sweet runs on open source models and tells you exactly what you're paying for. No proprietary models disguised as in-house technology. No markup hidden behind subscriptions.

DeepSeek v3.2

671B parameter MoE model. 37B active per token. Post-trained on custom engineering data.

Parallel agents

Multiple agents work simultaneously on different parts of your task instead of running sequentially.

Fireworks AI inference

Open source models served at $0.28/M tokens versus $3-15/M for proprietary alternatives.

Transparent billing

See your token usage and cost for every task. No hidden fees. No surprise overages.

Pricing

Pay for what you use.

No subscriptions. Prepaid balance that works across web app and CLI. Top up anytime.

Try it out

Good for 10-15 coding tasks. Enough to know if Sweet fits your workflow.

Start with $5

$10

Start building for $5.

No subscription. No lock-in. Open source models. Transparent pricing. Your first task runs in under 2 minutes.

Get Started Read the Docs

AI coding agents.10x cheaper.