Onboarding design partners

Some work is for agents.
Some is still for you.
Leroy runs both.

A Sentry error, a Linear ticket, a GitHub issue — or a whole project you scope out with Leroy in chat: scoped work goes to an AI engineer that plans the change, writes the code, reviews its own diff, and opens the PR. The work that needs you stays with you — and hands off the moment it’s ready.

Join the waitlist →Watch it work
app.useleroy.com — chatreal session
a customer says checkout hangs on Apple Pay — are we seeing that anywhere?
Yes — Sentry has a TypeError spike in the checkout flow, 142 events since 02:14, all Safari. It matches the report: the token verify call returns null and we don’t guard it. stack trace · checkout.ts:88
Want me to open a ticket and take it? Small fix — null guard plus a regression test. Est. $2–4.
PIT-751 created. Plan’s ready for you — I’ll start the moment you approve.
The ticket, from here
✓Ticket created from chat13:05
✓Plan approved — by you13:06
●Coding · sandboxed branchwatch live
4Self-review + checks
5PR opens — merges by your rules
1:12
// one continuous take: chat → ticket → plan → code → merged PR · no cuts, no script
Work flows in fromSentryLinearGitHubSlackMCPyour editor
// how it works

An AI engineer you manage,
not one you babysit.

If you’ve handed work to an agent and gotten back something broken, or just not what you asked for, here’s what’s different. Behind every ticket is a collection of agents and checks that plan the change, write it, review their own work, and fix what’s wrong before you ever see it — that’s why the code holds up. And it all runs at the autonomy you choose: Leroy pauses wherever you tell it to, at three points you can set to ask first — or none.

01

Take in the work

From Sentry, Linear, GitHub, Slack, chat, or your editor. Leroy dedupes, reads it, and decides whether it’s worth doing at all — noise never costs you anything.

picking up workask me firstrun on auto
02

Plan before code

A plan you can read in a minute — files, approach, risk, estimated cost. Nothing gets built that you haven’t seen coming.

starting to codeask me firstrun on auto
03

Code in a sandbox

On its own branch, in an isolated environment — your main branch is never touched. Watch every step live, and stop a run anytime: your note about what’s off becomes its direction for the retry.

// no pause here — this step is Leroy’s job
04

Check its own work

Before you ever see a PR, Leroy reviews its own diff, runs the checks, and fixes what it broke. You review work that already passed — not the agent’s first draft.

05

Ship the PR

The pull request opens — and merges however you’ve set it: always ask, or auto-merge when checks pass.

mergingask me firstrun on auto
Always asks

Some changes ask for a human no matter what mode you’re in: anything touching paths you’ve marked sensitive — auth/ billing/ migrations/ — and any run that hits its per-task spending cap.

You control how much autonomy Leroy has.

Three presets to start — then tune any step, per repo or per source.

Balanced — the default. Leroy picks up work and plans on its own; a human approves plans and merges. Flip any switch above to see it change.

There’s more underneath — spend caps, path rules, per-project policy.Learn more →
// the front door

Run the whole thing from a conversation.

No dashboards to build, no filters to save. Ask about your product, your code, or your work — and act on the answer in the same breath.

// is something broken?
users say search feels wrong since yesterday
Let me check Sentry and what shipped to search recently…
Yesterday’s ranking PR reweighted recency — that’s probably it. Two Sentry reports agree. Want a ticket to tune it back? create ticket
// where are we?
status on the checkout rebuild?
Let me pull the checkout rebuild project…
3 of 8 done, 2 running now. The Stripe retry PR is waiting on your review — it blocks the last three tickets. open PR #218
// what happened?
what shipped last week?
Let me search our open and shipped tickets…
11 PRs. 6 from the kiosk project, 4 bug fixes from Sentry, 1 dependency sweep. Total agent spend: $41. see the list

The UI is optional. Chat scopes projects, files tickets, answers for the whole system — and your editor can do the same over MCP.

// you + leroy

Work moves between you and Leroy. Both directions.

An agent can only run with work that’s been figured out — and not everything is. Leroy is built for the whole spectrum. Find your day below.

You build the hard 80%. Leroy finishes.

You’re pair-programming the tricky part in your editor — the architecture calls, the taste calls. Somewhere along the way, the rest becomes obvious: a clear-cut endpoint, the tests, the wiring. Hand exactly that across via MCP and start your next hard thing. Leroy ships its 20% while you’re still deep in yours.

YouLeroy
// you own the ambiguity · leroy owns the follow-through

This is the bet: the next generation of coding workflows won’t be humans or agents — it’ll be one system where work flows to whoever should own it, and the handoff costs nothing. Leroy is that system.

// pricing

Tasks, not tokens.

Every tier gets the full product. Per-task budget caps, top-up packs that never expire, no surprise bills.

Trial
See it ship one task.
$0one-time
5 tasks · 1 concurrent
no card required
Start free
Hobby
Keep your side project alive.
$99/mo
25 tasks/mo · 1 concurrent
all integrations · caps
Choose Hobby
Pro
Ship like a small team.
$399/mo
100 tasks/mo · 3 concurrent
priority routing · top-ups
Choose Pro →
Enterprise
Scale your org.
Custom
Negotiated volume
SSO · audit logs · SLA
Talk to us
// unlimited repos & teammates on every tier · annual = 2 months free
// closed beta

The first 50 get onboarded by hand.

Real time with every design partner, until Leroy fits the way you actually work.

// no spam · no drip · one email when it’s your turn