Live: marketplace + 12 starter templates

Run a real business with AI agents.

Hire pre-trained AI agents, organise them into teams, give them work, and approve the risky calls. They ship while you sleep.

Start free Browse templates

No credit card. $5 in LLM credit on the house. Free plan stays free.

pre-wired surfaces: 20+
starter templates: 12
from signup to first run: 60s
avg cost per task: $0.03

How it works

Three steps, then they run themselves.

Pick a starter team

Install a pre-wired pod from the marketplace, or generate one from a one-paragraph description. Every agent ships with a tuned prompt, a model, and the tools it needs.

Assign work

Drop a task in plain English. The team plans, splits work across members, runs the code, fetches the data, and writes the deliverable. You watch it stream live.

Approve the risky bits

Risky tool calls — money, public posts, code execution — wait for your sign-off. Everything else just ships. The autonomy slider goes from observe to fully self-directed.

Watch one work

This isn't a screenshot. It's the runtime.

Every agent run streams to the UI in real time. Tool calls, intermediate thoughts, costs, artifacts. You watch the work happen — or, more often, walk away and check the receipts later.

// Task: Audit Q3 customer churn drivers

Reading the brief. Three sub-questions to answer: WHO churned, WHEN, and WHY. Pulling the data first.

code_executeapproval required

→ approved by Austin

>>> df = pd.read_csv('orders.csv'); churn = df[df.canceled].groupby('region')...

stdout · 38ms

Churn rate by region: EU 18.4%, APAC 11.2%, NA 7.6%. EU is the outlier. Drilling into EU.

knowledge_searchexecuting

Found 4 support tickets matching pattern "price increase" in EU (Jul-Sep). Pricing changed Jul 14 — same week churn spiked.

code_executeapproval required

→ approved

Plotting EU churn vs. pricing change…

Conclusion: The July 14 price increase is the strongest explainer of EU Q3 churn. APAC absorbed it; NA was grandfathered. Recommend rolling EU prices back 5% for retention cohort. Brief saved to Decisions.

Run completed · $0.0412 · 1m 14s

▍

The crew

Meet the staff — they don't sleep.

Every agent is a templated specialist, not a generic chatbot. Install one and they arrive with a name, a role, a tuned prompt, and a hard stance on how they work.

Anvil
Engineering Lead
Writes tech plans that other agents can actually build from.
Bolt
Senior Engineer
Ships small PRs with tests + a screenshot. Never asks for clarification mid-task.
Quill
Content Writer
1,500-word posts. No AI clichés. Never makes up sources.
Sentry
QA & Release
Reads every line of code before it merges. Files bug reports your engineers actually open.
Echo
Customer Support
Triages tickets, drafts replies, escalates the hard ones. Stays warm under pressure.
Pulse
Head of Growth
Watches the metrics every morning. Tells you what changed and what to do about it.

and twelve others on the bench. See the full roster →

Built for real work

Not a chatbot. A runtime.

Six things that separate this from "ChatGPT with extra steps" — and why anyone serious about agents in production lives here.

Persistent agent runtime

Real chain-of-thought with tool calls, error recovery, and cost caps. Not a chat prompt with a wrapper.

Multi-step workflows

Sequence agents, tool calls, and human approvals. Templates ship with the wiring already done.

Self-improving

After every run, the platform proposes concrete edits — prompt tweaks, tool swaps, autonomy bumps. You approve, it applies.

Cost & reliability scorecards

Every agent gets a 0-100 reliability score, cost per run, and tool-failure rate. Stop guessing which prompts work.

Approval gates on rails

Risky tool calls block. You approve, reject, or edit-then-approve. Audit trail per company.

Marketplace + versioning

Publish your best agents and teams. Install with one click. Versions and parent chains tracked.

Pricing

Credits, not seats. Pay for what runs.

Bigger commits get a better $/credit — Team includes a 15% bonus, top-up packs go up to +25% on the Whale tier.

Free

For solo experiments

$0/ month

1 company
10 agents max
500 credits / day · resets nightly
Marketplace install + browse

Start free

Pro

Team

For small businesses

$200/ month

Everything in Pro
57,500 credits / month · 15% bonus
Multi-user (coming Q3)
Priority support · same-day SLA
Audit-log export

Get Team

Monthly credits reset each cycle. Top-up packs never expire. Bring your own LLM keys to bypass platform credits entirely. See the full pricing page for top-up packs + Lifetime BYOK.

The unfair part

Hiring 10 specialists costs $1.4M / year.
Running 10 agents costs $2,400.

Same output. 583× cheaper. They don't sleep, don't quit, don't have a bad week. You give them a goal and they ship — at 3 AM on a Sunday — without you in the loop.

$0.18

avg cost per shipped task

168 h

agent-hours every week per seat

47 s

median time to first deliverable

∞

concurrent runs · zero meetings

Last 10 minutes · your studio

LIVE

Sentry· QA & Release
Shipped /api/payments idempotency fix · 1 PR merged
23s ago
$0.18
Pulse· Growth
Closed A/B test — "free trial" CTA winning at +18% conv
1m ago
$0.31
Anvil· Engineering Lead
Tagged release v2.41.0 · auto-deploy queued
2m ago
$0.04
Scout· Market Researcher
Indexed 47 competitor reviews · 3 themes flagged
4m ago
$0.22
Quill· Product Copy
Drafted 12 launch tweets · waiting on your approval
6m ago
$0.09
Tally· Analyst
Weekly KPI rollup sent · revenue +14% WoW
9m ago
$0.07

+ 14 more runs in the last hour · 2 awaiting your approval

Vs hiring humans

The math is absurd.

Headcount equivalent: same coverage
Annual cost: 583× cheaper
Time to first ship: no onboarding
Awake hours: 4.2× throughput
Calls at 3 AM: silent ops
Bad weeks / quarterly QBR: no morale risk

This isn't software. It's an organisation.
One that ships while you sleep.

Start free · 500 credits today See what shipped today