Find out what your codebase actually does.
An AI agent workforce reads your entire codebase and writes a complete, fact-based book about your app: what every part actually does, what works, what doesn't, and where the code has drifted from your vision. Built for vibe coders and founders who shipped fast and need to know what they actually have.
Works with: Websites, web apps, iPhone and Mac apps, backends and APIs, browser extensions. JavaScript, TypeScript, Swift, Python, Go, Ruby, Kotlin, Java, Rust.
What you get
Technical Edition
Every screen, route, and function mapped with evidence and a health badge. File and line references throughout, so any finding can be checked against the real code in seconds.
Owner's Edition
Your app explained screen by screen, in plain English. No file paths, no jargon. Just what it does, what it doesn't yet, and an honest bottom line.
Both editions are delivered as branded PDFs via a private link only you receive.
How it works
Get your free Table of Contents
Send a public URL, a zip upload, or a read-only token scoped to one repo. We auto-generate your route-and-screen inventory and confirm your tier, no payment yet.
The agents map every road
A tiered team of AI agents walks every screen, route, and function in your codebase and assesses what each one really does, with file-and-line evidence.
A completeness pass
A second pass hunts for anything the first sweep missed. Anything it cannot fully verify is named in the report, never silently skipped.
RT Labs reviews and delivers
We check the findings, then send your Manuscript to a private link only you receive. 3-5 business days (24-48h on the Due Diligence Edition).
Real results
Three production apps scanned so far. Every headline finding below is backed by a file-and-line citation you can check against the actual code yourself.
28,000 to 77,000 lines of production code each, every finding checked against the source.
Code that looked finished but silently did nothing in production.
A required link that 404s, caught before it cost a rejected submission.
Two users could claim the same task at once. Nothing in the code stopped it.
How we verify
Your codebase is read by a workforce of AI agents working in stages, then checked by a person before anything ships. No single model's guess becomes a finding — every claim is traced back to your actual code.
First we enumerate every file, screen, route, function, and service in your codebase, each with its file-and-line location. This is the same map the free Table of Contents is built from.
For each part, an agent states what it is supposed to do, drawing on your README, spec, and the code around it. This is the yardstick the next stage measures against.
A separate agent checks whether each part actually does that, and labels it works, partial, stub, broken, or dead. Every label carries file-and-line evidence you can open and check yourself.
An independent pass compares what was reviewed against the full file inventory and dispatches gap-scouts to close any holes. If parts could not be verified in depth, we say so plainly.
Before delivery, a person reads both editions in full and spot-verifies every headline finding against your real code. Nothing goes out with RT's name on it until that gate is cleared.
Samples
A real client Manuscript is private and never published. The two below are RT's own apps, shared on purpose so you can see the real thing before you buy.
Technical Edition sample available on request. Ask for one.
Pricing
See your app's inventory before you pay anything
Send your repo link and we auto-generate a route-and-screen inventory of your app, no payment, no verification. It is also how your tier gets sized: no self-reporting, no guessing, no LOC math on your end.
Up to 10,000 lines
- Technical Edition PDF
- Owner's Edition PDF
- Delivered via private link
- 3-5 business days
By purchasing you agree to the Terms.
Up to 50,000 lines
- Technical Edition PDF
- Owner's Edition PDF
- Delivered via private link
- 3-5 business days
By purchasing you agree to the Terms.
Up to 150,000 lines
- Technical Edition PDF
- Owner's Edition PDF
- Delivered via private link
- 3-5 business days
By purchasing you agree to the Terms.
For buying, selling, a contractor dispute, or getting investor-ready. Same scan, priority queue, faster turnaround.
Not sure how many lines your app is? The free Table of Contents scan sizes it for you automatically, before anything runs and before any payment. You never pay the wrong price and you never count a line yourself.
Repo access & confidentiality
Public repo URL
Already on GitHub, GitLab, or similar and public? Just send the link.
Zip upload
Not public, or want more control? Upload a zip and you decide exactly what's in it.
Read-only GitHub token
A fine-grained token scoped to one repo, read-only, revocable any time you want.
Your code is read, never changed. Your Manuscript is private and never published. The samples above are RT's own apps, shared on purpose.
Security & data handling
Private workstation
Your code is handled on a private RT Labs workstation, not a shared cloud service, and it is never committed to any repository of ours.
No secrets needed
Nothing in the scan requires secrets. We recommend you strip your .env before sending access, and the scan runs the same either way.
No training on your code
Your codebase is read to write your Manuscript. Under our AI provider's commercial terms it is never used to train any model.
Deleted after delivery
Our working copy of your code is deleted once your Manuscript ships, with a confirmation email so you know it happened.
Who processes your code
The scan itself is performed by AI models from Anthropic (Claude) under their commercial data terms, which state that Anthropic does not train on business inputs sent through their API. Two other services support delivery: Supabase stores your finished report PDFs behind private, expiring links, and Resend sends your delivery and deletion-confirmation emails. That is the full list of third parties your code or report touches.
Strip your secrets before you send
The scan never needs your secrets. If you send a zip, this one-liner leaves your .env files and secret folders out, so raw keys never leave your machine:
zip -r myapp.zip . -x '*.env' '*.env.*' '*/node_modules/*' '*/.git/*' '*/secrets/*'Want a mutual NDA before you send anything? Ask and we'll send one.
The guarantee
Before delivery
Full refund, any time, no questions asked, right up until your Manuscript ships.
14 days after delivery
Point out any finding that's materially wrong and we send you a corrected report or a full refund.
Who's behind this
Manuscripts is built and run by Regie Durana at Round Table LLC. Round Table ships AI-built products in public, and this scanning engine was built to audit Round Table's own apps first: the Quipp AI and Hinahon samples above are the real output, run against real production codebases, before it was ever offered to anyone else.
Questions, answered
Is this a subscription?
No. One payment buys one Manuscript: a complete reading of your code as it stands that day, like a photograph. If you change a lot and want a fresh one later, that is a new scan. A track-changes version that only re-reads what changed is in the works, and Manuscript owners will hear about it first.
Why is pricing based on lines of code?
Because that is literally the work. The agents read every line, so a bigger codebase means more reading, more verification, and a longer book. Each tier covers the full reading, RT Labs review of every headline finding, and both editions. The free Table of Contents scan measures your lines automatically, so you never count them yourself.
What's your refund and accuracy guarantee?
Full refund any time before delivery, no questions asked. After delivery you get a 14-day accuracy window: point out any materially wrong finding and we send a corrected report or refund you in full.
How do you handle my code and data?
Your code is handled on a private RT Labs workstation, not a shared cloud service, and it is never committed to any repository of ours. Nothing in the scan requires secrets, so we recommend you strip your .env before sending access. The scan is performed by Anthropic's Claude models under their commercial terms, which don't train on the code we send. Your copy is deleted after delivery, with a confirmation email when it's done. A mutual NDA is available on request.
What kinds of projects can it scan?
Websites, web apps, mobile apps, Mac apps, backends, APIs, browser extensions. If it has a codebase, it can be read. It has already produced Manuscripts for a website with a browser extension, a mobile app with a web backend, and a native iPhone app.
Can you scan just one part of my app?
Yes. The scan covers exactly the code you hand over, so you can scope it to a single feature, module, or folder, and your tier is based on the lines in scope.
What happens after I pay?
By the time you pay, your free Table of Contents has already confirmed your tier and repo access. Payment starts the full scan: we run it, review the findings, and send both PDFs to a private link within 3-5 business days. Report links stay valid for 1 year.
Will you change or run my code?
No. Your code is read, never changed and never executed. The scan works from the code itself, and your copy is deleted after the run.
Get started
See what your codebase actually does
Pick a tier above, or email us your repo details and we will get your Manuscript started.
dev@wsbroundtable.com