✈ Launch Safely Under the hood Pricing Join the waitlist
Under the hood

Simple on the surface.
Rigorous underneath.

The calm screens sit on real engineering. This page shows the machinery: who reads your code, who checks the checkers, and where it all runs.

19
Checks
across 7 categories
11
Zero-LLM checks
pure code, finish in seconds
25
Tool calls
max per AI investigation
307
Automated tests
on the app itself
×2
Verified
every fix, before it counts
The pipeline

Nothing turns green on an agent’s say-so.

Every project moves through the same gauntlet. The gate in the middle is the point.

↩ Fails anything at the gate? The fix goes back, and the whole checkup runs again.

“A checklist that turns green on an agent’s say-so is theater. Verification is the product.”

The review board

Many specialists, one accountable verdict.

A checkup runs the way a real engineering org would run it. No black box: every finding shows what was read, who decided, and why.

Deterministic scanners run alongside the models, and neither side gets the last word without proof: a model can only overturn a scanner’s finding by citing the code that disproves it.

The audit

All 19 checks, two kinds of rigor.

11 checks are deterministic scanners: secret-pattern detection, dependency audits, config analysis. They run in seconds, before any model weighs in. The other 8 are AI investigations that read your actual code, up to 25 tool calls each.

Code deterministic, no LLM involved AI agentic investigation of your code

Security

7 checks
A secret key is sitting in your code where anyone could find itCode
Your private settings file is saved into your code historyCode
Some building blocks your app uses have known security problemsCode
Anyone on the internet may be able to download your customer dataAI
The app trusts whatever a customer’s device tells itAI
Someone could flood your sign-up form with thousands of fake accountsAI
Sign-in and passwords may not be handled safelyAI

Your data

2 checks
If your database is lost, there may be no way backAI
Customer personal details may be scattered where they could leakAI

Reliability

3 checks
When something breaks, customers see a blank pageCode
Your app keeps no diary, so problems can’t be traced afterwardsCode
There’s no way to ask your app “are you okay?”Code

Operations

3 checks
Changes to your app aren’t being saved in a historyCode
Your app’s parts list isn’t pinned downCode
You test changes on the same app your customers are usingAI

Cost safety

1 check
Nothing limits how much your app can spend on AIAI

Legal

1 check
Your app collects personal details but has no privacy policyCode

Testing

2 checks
Nothing automatically checks that your key flows still workCode
Your app can’t prove it builds cleanly from scratchCode
The verify gate

Fixed means proven fixed.

After agents fix something, the gate doesn’t ask them how it went.

Anything that ships under your name, like a privacy policy or customer-facing copy, stops and waits for your explicit approval. One click reverts everything.

The run environment

A locked room, destroyed after use.

Every checkup spins up an ephemeral, isolated sandbox. Your code runs and is read inside a locked container that is destroyed when the run ends.

The sandbox
a fresh Cloudflare Sandbox, one per checkup
Your coderead and run only inside the container
The agent fleetdurable, stateful orchestration on the Agents SDK
Deterministic scannerssecret patterns, dependency audits, config analysis
Destroyed on finishnothing outlives the run
all model traffic
AI Gateway
every model call routes through it
Spend capsper-run limits enforced at the platform layer
Full loggingevery request recorded
Prompt cachingre-checks are fast and cheap

Runs stream live. You watch the agents read your actual files, not a spinner.

Trust, but verify. Twice.

Join the waitlist