Form AB-1  ·  The Approval Layer for AI-Written Software

AI writes code fast.
AIOS Build makes sure it's the right code —
and keeps the receipts.

Think of it as a building inspector for AI-written software. Every request gets checked, risk-scored, and approved by a human before the AI touches your code. Every decision is recorded, forever.

See the proof, side by side Try the live pipeline
7Checkpoints per request
8Safety rules, always on
470Automated tests passing
25Approved directives in the rulebook
100%Decisions recorded & explained
SECTION 01 — WHY THIS EXISTS

The bug nobody approved

A true story from thousands of teams. Maybe yours.

Friday afternoon, someone asks an AI assistant to "fix the login page." The AI fixes it. It also — helpfully — rewrites how user sessions work. Nobody asked for that. Nobody noticed it either, because it was buried in 900 lines of confident-looking code.

Monday morning, customers can't log in. The team scrambles. And then the hard questions start — the ones every company using AI to write code will eventually face.

AIOS Build exists so that every one of those questions has an answer. Before the code is written, not after it breaks.

Incident review — Monday, 9:04 AM
Who approved the session change?NOBODY
What exactly did the AI change?UNKNOWN
Was it risk-checked first?NO
Where's the record of the decision?NO RECORD
Could it happen again tomorrow?YES
SECTION 02 — SAME REQUEST, TWO OUTCOMES

The proof, side by side

One plain-English request. Lane A goes straight to the AI. Lane B goes through AIOS Build first. Watch what changes.

Incoming request — submitted by product team Add a "forgot password" feature to our app.
Lane A — Straight to the AINo checks

The AI writes the feature. It also makes some decisions on its own:

  • Quietly rewires the email system to send reset links its own way
  • Changes the structure of the user database "while it's in there"
  • Picks its own security rules for reset tokens — no one reviews them
  • Delivers 900 lines across 14 files, all mixed together

It might be fine. It might not. You can't tell — and there's no record to check.

CHECKED: nothing · APPROVED BY: no one · RECORD: none
Lane B — Through AIOS BuildCommand Pack AB-2041

The request passes 7 checkpoints. The AI receives a Command Pack — exact, fenced instructions:

  • Build: a reset form, a secure token, an email with the reset link
  • Use: the team's approved security pattern for reset tokens
  • Do not touch the email system's configuration
  • Do not change the user database structure

Risk score 38/100 — the security rule triggered, so a human reviewed and approved it before any code was written.

CHECKED: 8 rules · APPROVED BY: a human, by name · RECORD: permanent audit entry

Same AI. Same request. The difference is the paperwork — and the paperwork is the product. A Command Pack tells the AI what to build and what it must never touch.

SECTION 03 — THE SEVEN CHECKPOINTS

How a request earns approval

Like airport security for code requests: seven stations, in order, every time. Takes seconds, not meetings.

  1. Read the request carefully / Request Input

    Cleans up the request and flags anything vague. Vague in means vague out — so vagueness gets caught here first.

  2. Figure out what's really being asked / Intent Detection

    Two independent checkers classify the request. If they disagree by too much, that's a red flag — a human gets pulled in.

  3. Map what it touches / Requirement Analysis

    Lists exactly which parts of your app this request would affect. Touching more than expected? Flagged.

  4. Match it to your rulebook / Directive Selection

    Finds the approved way your team already does this kind of work. The AI follows your standards — not its mood.

  5. Run the 8 safety rules / Risk Assessment

    Scores the risk from 0 to 100. If any rule trips, everything stops and a human decides. No exceptions, no bypass button.

  6. Write the Command Pack / Generation

    Produces the fenced instruction set: build exactly this, never touch that. Works with any AI assistant — it's plain, structured text.

  7. Show the work / Explainability

    Writes a plain-English explanation of every decision made at every checkpoint, and files it in a permanent record.

SECTION 04 — THE TRIPWIRES

The 8 escalation rules

"Escalation" means one thing: stop and ask a human. These eight rules run on every single request. Any one of them can stop the line.

RULE 01 HIGH

Too big

The request touches more than two parts of the app. Big changes need human eyes.

RULE 02 CRITICAL

Touches deployment

It changes how software ships to real users. That's never automatic.

RULE 03 CRITICAL

Uninvited changes

The AI would change things nobody asked for. This is the #1 silent killer — and it's caught here.

RULE 04 CRITICAL

Security & privacy

Anything near passwords, logins, or private data stops for expert review. Always.

RULE 05 HIGH

Low confidence

The system is less than 70% sure what you meant. Guessing is not allowed.

RULE 06 HIGH

The checkers disagree

The two independent classifiers came to different conclusions. Disagreement is a warning sign, not a coin flip.

RULE 07 MEDIUM

Never seen before

The request matches nothing in your rulebook. New territory gets a careful look.

RULE 08 HIGH

Missing information

Details are unclear or missing. AIOS asks first — it never fills the gaps with guesses.

"Confidence is not permission." No matter how sure the AI sounds, permission comes from humans. That single sentence is the whole philosophy.

SECTION 05 — WHO IT'S FOR

Two kinds of winners

The same governance layer pays off in two very different ways.

TRACK A

For engineering leaders

Someday soon, your board, your auditor, or your biggest customer will ask: "Who approved the AI's code?" You'll have an answer.

  • A permanent, tamper-resistant record of every AI request, decision, and approval — with a name attached
  • Five roles and thirteen permissions, so the right people approve the right things
  • Risk scoring and mandatory human review for anything touching security, deployment, or data
  • Works with any AI assistant your team uses — Claude, Copilot, Cursor, ChatGPT. Switch tools, keep governance
  • Runs on your own machines. Your requests never leave your building
TRACK B

For solo builders & small teams

Better instructions in, better code out. A Command Pack is the best prompt you've ever written — generated automatically, every time.

  • Stops scope creep cold: the AI gets fences, not just wishes
  • Your Project Constitution remembers your standards — framework, style, testing rules — so you never re-type them
  • Rescuing an old spaghetti project? The Renovation planner turns it into five governed phases — map it, protect it with tests, then fix it — one Command Pack per phase
  • Catches vague requests before they turn into three days of rework
  • Every decision explained in plain English, so you learn why, not just what
  • Free to start. One project, the full pipeline, no credit card
SECTION 06 — THE AIOS FAMILY

Build is the beginning

One governance idea, four products. Two are working today.

Working today

AIOS Build™

The approval layer. Plain-English request in, human-approved Command Pack out, permanent record kept. Everything on this page.

Working today

AIOS Relay™

Two AIs, one referee. A supervisor AI plans the work, Claude executes it — and the Relay sits between them, blocking anything that exceeds its written authorization. Already governing 34 real projects.

Roadmap

AIOS Code™

Governed code generation. The Command Pack doesn't just instruct the AI — it becomes the contract the generated code is checked against.

Roadmap

AIOS Standard™

The certification. "AIOS Standard Certified" tells customers and partners: this team governs its AI. Like a safety inspection sticker for engineering organizations.

SECTION 07 — PRICING

Start free. Govern everything.

Pay for seats and control, not for requests. Governance shouldn't have a meter on it.

FOR SOLO BUILDERS

Builder

$0
forever · one project
  • Full 7-checkpoint pipeline
  • All 8 escalation rules + 25-directive rulebook
  • Command Packs for any AI assistant
  • Project Constitutions + guided project setup
  • Renovation planner for legacy codebases
  • Plain-English explanations
Start free
FOR TEAMS

Team

$49
per person / month
  • Everything in Builder, unlimited projects
  • Roles & approval workflows (5 roles, 13 permissions)
  • Permanent audit log with exports
  • Request history & ownership tracking
  • AIOS Relay multi-project governance
  • Priority support
Talk to us
FOR ORGANIZATIONS

Enterprise

Custom
annual agreement
  • Everything in Team
  • Self-hosted on your infrastructure
  • Custom escalation rules & directives
  • Compliance-ready audit reports
  • SSO & provisioning
  • AIOS Standard certification prep
Request a demo
SECTION 08 — QUESTIONS

Fair questions, straight answers

Does AIOS Build write the code?

No — and that's the point. Your AI assistant writes the code. AIOS Build writes the instructions and the fences the AI must work inside, then keeps the record of who approved what. Inspector, not builder.

Won't the checkpoints slow us down?

The pipeline runs in seconds. What's genuinely slower is the deliberate part: when a risky request stops for human approval. That pause is measured in minutes — and it routinely replaces the days you'd spend untangling a change nobody reviewed. Slow is smooth, smooth is fast.

Which AI tools does it work with?

Any of them. A Command Pack is plain, structured text — paste it into Claude, Copilot, Cursor, ChatGPT, or whatever comes next. Your governance survives every tool switch.

What if the AI ignores the instructions?

The fences make violations visible: the Command Pack states exactly what must not change, so a reviewer can check the work against it line by line. And with AIOS Relay, a policy firewall sits between the AI and your files, blocking unauthorized actions before they happen.

Where does my data go?

Nowhere. AIOS Build runs on your own machines, with zero third-party packages in its runtime. Your requests, your rules, and your audit log stay inside your walls.

What does "escalation" actually mean?

Stop and ask a human. That's it. Eight rules watch every request, and any one of them can pull the cord. No AI confidence score can override a human decision — confidence is not permission.

FORM AB-1 · FINAL SECTION

Your AI is already writing code.
Who's approving it?

Put an inspector at the gate. Start free with one project, and see your first stamped Command Pack in minutes.

Start free Request a demo