Autonomous CLI agent

Give it a goal.
Come back to shipped code.

Sweet runs autonomous agent loops — it plans, writes, tests, and fixes its own work, running for as long as the task takes. No babysitting. No dead ends.

~/project — sweet
npm install -g @sweet-cli/sweet
How the loop works ↓
US only

The Agent Loop
Plan → Execute → Verify → Recover → Repeat.

Sweet doesn't suggest a line and stop. It holds the entire goal, breaks it into steps, and runs the full loop — catching failures, diagnosing them, and retrying — until the work is done.

01 · PLAN

Reads & plans

Explores your repo, then writes a concrete step-by-step plan before touching anything.

02 · ACT

Executes

Edits files, runs commands, calls tools — many steps, one coherent thread.

03 · VERIFY

Tests its work

Runs your test suite, reads the output, and checks results against the goal.

04 · ADAPT

Recovers & continues

When something fails, it diagnoses and retries — no babysitting, no dead ends.


The /workfor command
Set a timer. Walk away.

/workfor puts the agent in a continuous work loop — it plans, executes, verifies, recovers from failures, and self-nudges for the full duration you set. It won't stop or idle until the timer expires. Think of it as a stop hook that never stops.

How it works

  • Type /workfor 3h "build the payments integration"
  • The agent enters a loop: plan → execute → verify → recover → repeat
  • It self-nudges — no idle time, no waiting for you to say "continue"
  • Each step is verified. Failures are diagnosed and retried automatically
  • When the timer runs out, it reports everything it shipped
~/project — sweet
$ sweet Loading project… > /workfor 2h "refactor auth module to use JWT" ⏱ /workfor — 2h 00m Working in a loop until the timer expires. ◆ 1/6 Audited auth flow, 4 endpoints ◆ 2/6 Wrote jwt.ts (86 lines) ◆ 3/6 Patched 4 route files ◆ 4/6 npm test — 1 failure ↳ token expired in test, refreshing… ◆ 5/6 Updated middleware chain ◆ 6/6 npm test — 22 passing 6 steps · 1 recovery · 1h 14m ✓ Done — ready to review

Features
Built for the terminal.
Designed for autonomy.
TERMINAL-NATIVE

One command from any repo

No browser, no dashboard. sweet "build me a REST API" and it goes to work right where your code lives.

AGENT LOOP

Autonomy measured in hours

Not autocompletes. Not chat messages. Sweet loops: plan → execute → verify → recover, over and over, for as long as you need.

SELF-VERIFYING

Tests before it calls it done

Runs your test suite, reads failures, and fixes them. The task isn't done until the suite is green.

SELF-RECOVERING

Fails, diagnoses, retries

Flaky test? Wrong assumption? The agent catches it, figures out what went wrong, and keeps going.

ANY LANGUAGE

Node, Python, Go, Rust, Rails…

Reads your conventions and follows them. No language-specific configuration needed.

RUNS LOCALLY

Your machine, your code

Runs in your terminal. No repo storage on our servers. No training on your source.


Pricing
Start free, scale up.

Full agent, all languages. Model access through our relay — no API keys to manage.

Free forever
Free
~50K tokens / day
$0/mo

Full agent. No credit card. Install and go.

  • Full agent, all languages
  • ~50K tokens daily
  • /workfor (up to 30 min)
  • No API keys needed
Get Started Free
US only
Pro
Pro
~500K tokens / day
$20/mo

More power for serious daily use.

  • Everything in Free
  • ~500K tokens daily
  • /workfor (up to 4 hours)
  • Priority relay access
Get Pro
Most popular
Max
Unlimited tokens
$200/mo

Unlimited. For teams and power users.

  • Everything in Pro
  • Unlimited tokens
  • Unlimited /workfor
  • Parallel agent swarms
Get Max

Get started

Your terminal.
Your agent.
Autonomous loops.

One install. Every repo. No API keys needed. Start free, upgrade anytime.

npm install -g @sweet-cli/sweet