Start free
Live demo · click Save below

Point at the bug.
Get the Fix.

Drop one snippet into your deployed app. Click the broken Save button below — watch the autopilot ship the fix.

acme.com/settings/profile
run #1042 · live
AAcmeSettings › Profile

Profile

How your name appears across Acme.

Sam Sutton
sam@acme.comverified
change
↑The dev shipped Save with bg-muted instead of the primary style. Click anyway — the autopilot will catch it.
acme/web›runs›#1042
idle
TimelineCaptureVerify logPull request
Waiting for capture from /settings/profile
run #1042 · acme/web · default-branch
  1. Capture receivedselector + DOM + computed styles + runtime errors · 248ms
  2. Cloning into microVMacme/web @ default-branch HEAD · 3.1GB · 4.2s
  3. Agent investigatingcomponents/SettingsForm.tsx:42 · claude-sonnet-4.6 · 38s
  4. Fix proposed1 file · +1 −1 · 4m 12s
  5. Draft PR openedacme/web · #1042 · awaiting your review
PR #1042 · draft · ready for review
fix(settings): use primary button style for Save
↳ components/SettingsForm.tsx · +1 −1 · ready in 4m 12s
View pull request
No credit cardGlobal kill switchRead-only agent toolsDraft PRs only
Building, not reporting?Use the free dev tool with claude →
The loop · four primitives

A pipeline, not a chatbot.

Each stage has a hard contract. The agent only edits inside the microVM. Nothing reaches your default branch without your review and merge — that's the safeguard against cost blow-outs and agent mistakes. The human review step is the feature.

  1. 01

    Capture

    selector · DOM · runtime · source

    Origin-pinned, quota'd, rate-limited. Text fields are token-scrubbed before persistence; selector, DOM, computed styles, runtime errors, and source line travel with the report.

  2. 02

    Sandbox

    ephemeral microVM · read-only tools

    Your repo, cloned per-run into a Vercel Sandbox microVM. Read-only agent tools — no shell, no sockets, no code execution in your repo. Destroyed at end-of-run.

  3. 03

    Verify

    lint · typecheck · tests · build

    Lint, type-check, tests, build, preview deploy — all inside the microVM. The PR doesn't open until they pass. Failed runs route to needs_human and cost you nothing.

  4. 04

    Ship

    draft PR · your review · your merge

    A draft or ready PR on the default branch base — your review and merge, always. Per-project + global kill switches, both <1s.

The difference

Bug reports used to be ambiguous.
Now they merge.

Bug tools file dumb tickets. Coding agents work greenfield. InSitue carries selector + DOM + styles + runtime + source from your live app straight into an agent-actionable, verified change.

JIRA-era bug report

  • "Save button looks broken" — no repro steps.
  • Engineer spends 40 minutes trying to reproduce.
  • Closed as 'can't reproduce' or pushed to next sprint.
  • Customer churns three weeks later.
ACME-4217 · 14 days open
Save button looks broken??
the save button on the profile page looks dead. greyed out? it does save when I click it though. on chrome. please fix
needs reprostale

InSitue PR

  • Captured element + computed styles + runtime errors.
  • Agent resolves source line, fixes, and runs your build.
  • PR opens in 3–8 minutes — with the repro attached.
  • You review, click merge, customer sees the fix today.
acme/web · #1042 · draft
fix(settings): use primary button style for Save
typecheck ✓build ✓screenshot + repro
↳ 1 file changed · +1 −1 · ready in 4m 12s
Hardening · explicit

Safety isn't a slide.
It's the architecture.

Every safeguard is on by default, visible in the dashboard, and reversible.

Manual review by default

Every fix lands as a draft PR for you to review. Nothing reaches your default branch without your merge — the safeguard against cost blow-outs and agent mistakes.

Ephemeral microVMs

Every run is a Vercel Sandbox microVM destroyed at end-of-run. Your source never persists in our infrastructure.

Read-only agent tools

No shell. No sockets. The agent can't escape the repo. It proposes diffs; it doesn't run code in your repo.

Two kill switches

Per-project pause + global stop-the-world. Both <1s. Every PR is reversible with one click.

Least-privilege scopes

GitHub: contents + pull-requests. Vercel: Projects: Read. Selected repos only.

Rate-limited capture

Capture keys are Origin-pinned, quota'd per project, rate-limited, deduplicated. Cost-DoS is bounded by design.

I kept watching engineers spend Monday mornings reproducing bugs that QA filed on Friday. InSitue is what I'd build for myself — the report, the repo, and the fix in one loop.
Rod Leviton
Founder · InSitue
Usage-honest pricing

You only pay for fixes that ship.

Reports that route to needs_human or that you reject on review are never billed.

Free
$0/mo
Trial — 3 resolved PRs, 10 captures, 1 project. Demo the cloud autopilot.
View pricing
Solo
$12/mo
12 PRs · 200 captures · 2 projects · $2.50/PR overage. The no-brainer expense.
View pricing
Propopular
$39/mo
35 PRs · 500 captures · 5 projects · $1.50/PR overage.
View pricing
Team
$149/mo
150 PRs · 3,000 captures · 15 projects · unlimited seats · $1.00/PR overage.
View pricing
FeatureFreeSoloProPOPULARTeam
Monthly base$0$12$39$149
Resolved PRs included31235150
Bug captures / mo102005003,000
Per-PR overage—$2.50$1.50$1.00
Projects12515
Team seats1 (you)1 (you)1 (you)Unlimited
Max bill (spend cap)hard-stop$50$250$1,000
SupportEmail · best effortEmail · 48h SLAEmail · 24h SLAPriority · 4h SLA
Frequently asked

The questions every team asks first.

Does InSitue merge code automatically?
Never. Every InSitue change opens as a pull request you review and merge yourself. Safe tier opens a draft PR (CI may not run on draft in some repos); Standard opens a ready-for-review PRwith CI + preview deploy. Either way, the human merge is the feature — we don't auto-ship to your default branch.
What happens if the agent can't produce a fix?
The run is marked needs_human with the full agent transcript attached. No PR opens, no meter ticks, no charge.
Do you store our source code?
No. Your repo is cloned into an ephemeral Vercel Sandbox microVM that is destroyed at the end of every run. We persist the agent transcript and the proposed diff for the run — not your source.
Which AI model do you use?
Anthropic's Claude family for the autopilot. Per-run model selection by task type; you can pin a model per project if compliance requires it. See security posture for sub-processor detail.
How does billing work?
A small monthly base plus a metered price per resolved PR. Runs that route to needs_human or that you reject on review are never billed. Spend caps are enforced per project; flip them anytime.
Can I self-host?
Not on launch. On the roadmap for Scale / Enterprise — likely tenant-isolated dedicated region first, then true self-host. Email us if you have a hard requirement.
GDPR / DPA?
Yes. Our DPA lists every sub-processor, customer-source residency policy (ephemeral, per-run, never persisted), and data transfer scope.
How is this different from Sentry, LogRocket, or Cursor?
Sentry tells you something broke. LogRocket replays the session. Cursor helps you write greenfield code. InSitue is the loop that connects your live app's bug report — selector, DOM, styles, runtime, source — directly into an agent-verified pull request.

Ship fixes while you sleep.

3 resolved PRs free. No card. Kill switch from setup minute zero.

Start freeRead the docs