Skip to content
One review queue for every AI coding agent

Fire off tasks. Close the laptop. Merge from anywhere.

Connect a GitHub repo, describe a task, and pick an agent. It runs in the cloud on your own key, opens a real pull request, and lands in one queue with a summary, risk flags and what it cost. No VPS to babysit, no machine left running.

Claude CodeOpenAI CodexGitHub CopilotKimiGLMGitHub
CopilotNeeds you
Move sessions to httpOnly cookies
touches-authwaiting 2m
Claude CodePR open
Add idempotency keys to the charge endpoint
touches-moneyCI passing#218
CodexRunning
Fix mobile nav overflow on iOS Safari
working…$0.02
Your queue, floating calmly. Ducks in a row, at last.
Beneath the surface

All the churn happens down here. You never touch it.

Every task runs in a fresh cloud sandbox on the vendor's infrastructure. AgentDuck orchestrates the machinery and only surfaces what needs a human, so you can shut the VPS down: there is nothing left for it to do.

Cloud sandboxes

A throwaway environment per task. Your repo is cloned in, the work happens, the sandbox is torn down. Your machine is never involved.

Branch and PR machinery

Agents commit to their own branches and open real pull requests. Nothing merges without you.

Risk analysis

Every diff is scanned and flagged before you see it: touches auth, changes money, no tests, big diff.

Cost metering

Every task priced to the cent on your own keys, totalled per agent and per project, checked against your weekly budget.

Works withClaude CodeOpenAI CodexGitHub CopilotKimiGLMGitHub
How it works

From idea to pull request in three steps.

1

Connect a repo

One click through the GitHub app. No tokens to copy, no keys to paste. Pick the repos AgentDuck may open pull requests on.

✓ payments-api connected
2

Describe the work

Write what you want done, pick an agent and a model, hit send. Then close the laptop. Seriously, it's fine.

"add rate limiting to the OTP endpoint"
3

Merge from the queue

The PR lands in your queue summarized and risk flagged, ready to merge or send back with a comment.

✓ PR #218 · summary ready · $0.04
Supervision, not vibes

Fire agents. Keep the calm.

Autonomy is only useful if you can trust what comes back. AgentDuck turns a pile of agent output into decisions you can make in seconds.

Review queue

Read the diff in one sentence

Every PR arrives with a plain-English summary and automatic risk flags: touches auth, changes money, no tests, big diff. You know where to look before you open a single file.

Move sessions to httpOnly cookiestouches-auth
Add idempotency keys to the charge endpointtouches-money
Generate monthly statement PDFsCI ✓
Claude$1.12
Codex$0.63
Copilot$0.39
Total$2.14
Cost control

Know the bill, per agent

Every task priced to the cent, totalled per agent and per project so you see which one earns its keep. Set a weekly budget and get warned before you blow past it.

Context packs

Your standards, in every task

Write your architecture, conventions and rules once. The pack rides along with every dispatch, so agents build like they've read your codebase. No re-explaining per task.

## standards
Next.js App Router, plain SQL, no ORMs
## rules
never touch migrations without a task
always add a test for money paths
🧠

Pick the brain per task

Opus for the gnarly refactor, Haiku for the copy tweak. The cost tracking follows your choice.

↩️

Send it back with a comment

Type what to change and the same agent updates the same PR. No starting over.

📱

The queue fits in your pocket

Install it like an app, get pinged when an agent needs you, merge from the couch.

Pricing

Priced like a tool, not a tax.

You're already spending $20 to $100 a month on the agents themselves. AgentDuck is the thin layer that makes that spend pay off.

Basic
$19/mo
For a solo dev putting agents on a couple of repos.
  • 3 projects · 75 tasks / month
  • Claude Code + Codex
  • Review queue, summaries & risk flags
  • Cost tracking & budgets
Start 3-day free trial
● Most popular
Growth
$39/mo
For heavy daily use across your whole portfolio.
  • 10 projects · 300 tasks / month
  • All 5 agents (adds Copilot, Kimi & GLM)
  • Context packs
  • Everything in Basic
Start 3-day free trial
Pro
$79/mo
For agent-first development with no ceilings.
  • Unlimited projects & tasks
  • All 5 agents
  • Context packs
  • Priority support
Start 3-day free trial
Questions, answered

The things you'd ask before trusting us.

Do I need my own API keys?+
You connect your own agent accounts, and they don't have to be expensive metered API keys. Connect your GitHub Copilot subscription, or a GLM or Kimi coding plan (flat-rate, from a few dollars a month) and pay nothing extra per token. We never mark up tokens.
Where does my code actually run?+
In a fresh, throwaway sandbox per task on the vendor's cloud. Your repo is cloned in, the work happens, the PR is pushed to GitHub, and the sandbox is torn down. Your machine is never involved.
Can the agents break my main branch?+
No. Agents only ever open pull requests on their own branches. Nothing merges without you. Your CI runs on every agent PR like on any other.
What happens when the trial ends?+
After 3 days your plan starts and your card is charged. We email you before that happens, and canceling is two clicks in the billing portal. Cancel during the trial and you pay nothing.
Ready when you are

Stop babysitting agents. Get your ducks in a row.

3 days free · Your own keys · Cancel in two clicks