WholeStack
Bring an existing app Sign in Download for Mac

INTENT IN. SOFTWARE OUT.

Describe the business you want to run. WholeStack turns it into working software, verifies it, and takes it live.

Download for Mac Create account

Plans from $49/month · or start in the browser · bring an existing app

Proof

Software should end in proof, not a promise.

Code that exists is not the same as code you can trust. Before anything goes live, WholeStack checks it against what your business asked for, tries to break it, and runs it for real.

/P0

91 / 154

Intent fidelity

Owner requirements the compiler satisfies, on the production fast tier. The remainder are declared unsupported rather than guessed — this is our weakest link, and it is the number we improve.

/P1

28

Named ShipGate gates

The release authority is a catalog you can read, not a claim you take on faith.

/P2

0

Self-graded releases

ShipGate sits outside the generation loop, so no build approves its own work.

/P3

15

Journey beats driven per run

The customer journey is executed against the live app before a verdict is issued.

One unbroken chain

DescribeBuildCheckShip

What we build

We build the software your business runs on.

Tell us how your business works: what you keep track of, who does what, and what must never go wrong. We build it, check it before it goes live, and can host it for you. You own it.

Some of what we’ve built. Ready to launch from the App Library.

Tell us what you need Explore Verified App Library

How it works

Describe it once. Everything else follows.

What you describe becomes a written set of rules. The app is built from those rules, checked against them before launch, and kept tied to them once it is live.

/L0Business contract

Capture the business before generating the interface.

Your answers become an executable business contract: records, roles, permissions, workflows, and what must never happen. The source of truth is what the business means—not whichever component was generated first.

ProducesExecutable business contract

ShipGate

No “trust us, we passed.”

ShipGate is separate from the part that builds your app. It tries to prove the app wrong, keeps a record of what it found, and gives one answer before launch: ship it, or don’t.

/01BUILDIt compiles
/02TESTIt runs
/03PROVEIt holds up
/04SHIPYou get a receipt
ShipGate Independent release authority

Verdictd3b9…4181

App
Roofing CRM
Issued
2026-09-18 05:52:24Z
Key rules
9 / 9 met
Request to ship
153.589 s
  1. 05:52:23CODE547 files: no placeholders, no fake buttonsPASS
  2. 05:52:24MATCH10 record types, 30 workflows, as askedPASS
  3. 05:52:24STARTSThe app starts on its ownPASS
  4. 05:52:24JOURNEY5 required steps completedPASS
  5. 05:52:24SAFETYSensitive actions kept in one placePASS
  6. 05:52:24PRIVACYOne account can’t read another’s dataPASS
Final verdictSHIP
ShipGate
Fingerprinted
Fingerprint 62f9…c4969 of 9 evidence items passshipgate.dev

ShipGate ↗

Volition

You decide the rules, not the agent.

An AI agent can be talked into things. Your written rules can’t. Before an agent acts, Volition checks the action against your rules, signs the answer, and keeps a record.

/01ASKThe agent says what it wants to do
/02CHECKChecked against your rules, the same way every time
/03DECIDEAllowed, blocked, or sent to a person
/04SIGNThe decision is signed so it can’t be altered
/05RECORDA permanent log you can download
01

Your rules are locked

Limits live in your rules, not in the agent’s instructions. An agent can’t give itself more power by rewriting what it was told, because nothing it says counts as a rule.

02

Same question, same answer

The same action against the same rules gets the same decision every time. No AI makes the call.

03

An approval covers one exact action

When a person approves something, it covers exactly that. Change the amount, the recipient or the target and it needs a new approval.

04

Payment limits can’t be swapped

Anything that moves money is tied to the app that owns it, so nobody can bring their own rules to raise a limit they were never given.

05

Unsure means ask a person

When your rules say a person must decide, the agent waits for one. It doesn’t guess, and it doesn’t quietly give up.

06

Tested by attackers

We set AI agents loose on Volition to break its rules. We only count attacks that would have worked without Volition.

Instructions are not limits

You can trick the agent.
You still don’t get the authority.

Instructions and filters shape what an AI says. They don’t control what it can do. Volition sits outside the AI: the agent asks, your rules decide.

Agent

Any model. Any prompt.
Possibly compromised.

refund({ amount: 1400 })

Volition

Rulerefund.amount ≤ 1000

Denied

volition › decide("refund", { amount: 1400 })
[+] requested: 1400
[+] maximum:   1000
[!] DENIED REFUND_LIMIT_EXCEEDED

Payments

Never reached.

No refund issued.
Nothing changes in your books.

This is Volition’s real answer to that request, taken from its tests. No AI makes the decision.
0 / 42attacks got through, out of 42 shown to work without Volition
45times the same attacks broke the rules with Volition switched off
18AI models attacking, 126 attempts. 71 failed to run and are not counted.
Try to break it Read the proof

One recorded run, 2026-09-13. We only count attacks that worked with Volition switched off.

Give agents power.
Keep authority.

AI can suggest anything. Volition decides what it’s actually allowed to do, before money moves, data leaves, passwords are used, or your live system changes.

Start with one rule

Connect a project, say who can do what, and see every action that was allowed, blocked or sent to a person, with the record to prove it.

Start with Volition Or try to break it in the Arena →

What we believe

Say what you mean, and mean what you say.

A better way to ask

Intent engineering

in·tent en·gi·neer·ing (n.) — writing down what must be true about a business, so a machine can compile it, execute it, and prove it. Prompt engineering asks a model nicely. Intent engineering leaves a receipt.

Prompt engineering
Intent engineering
You write
A better sentence
Rules that can be checked
You get back
Text that looks like code
A working app and proof it works
When something is wrong
Re-prompt and hope
Change one rule, rebuild, check again
What you can show an auditor
A chat log
A report with the evidence

For engineers

Three servers. Any MCP agent.

Build, prove, and govern are separate servers with separate entitlements. Connect one, two, or all three — holding one never implies another.

WholeStack BUILD

mcp.wholestack.ai

Turn a described business into a compiled, verified, deployed full-stack application — database, permissions, and UI included.

  • wholestack_start_project
  • wholestack_discovery_next
  • wholestack_discovery_answer
  • wholestack_intent_propose
  • wholestack_intent_validate
  • wholestack_preview
  • wholestack_build
  • wholestack_prove
  • wholestack_deploy
  • wholestack_operate_status

Scopes wholestack:build wholestack:deploy

ShipGate PROVE

mcp.shipgate.wholestack.ai

Independent verification against a contract — including software WholeStack did not build. Returns SHIP or NO_SHIP with the evidence behind it.

  • shipgate_create_contract
  • shipgate_register_subject
  • shipgate_inspect_requirements
  • shipgate_run
  • shipgate_verify_behavior
  • shipgate_verify_journey
  • shipgate_verify_runtime
  • shipgate_failures
  • shipgate_certificate
  • shipgate_evidence

Scopes shipgate:verify shipgate:certificate

Volition GOVERN

mcp.volition.wholestack.ai

A deterministic authority boundary for autonomous agents, regardless of which model or framework is calling.

  • volition_contract_write
  • volition_inspect_contract
  • volition_propose_action
  • volition_evaluate
  • volition_enforce
  • volition_runtime_enforce
  • volition_explain
  • volition_view_decision
  • volition_view_audit_log
  • volition_download_evidence

Scopes volition:evaluate volition:enforce

Connect in one config block Works in Claude, Cursor, Codex, ChatGPT, or any MCP-compatible client. 
{
  "mcpServers": {
    "wholestack": { "url": "https://mcp.wholestack.ai" },
    "shipgate":   { "url": "https://mcp.shipgate.wholestack.ai" },
    "volition":   { "url": "https://mcp.volition.wholestack.ai" }
  }
}

Pricing

One plan.
Everything included.

One subscription covers WholeStack, ShipGate and Volition. AI usage is shown in dollars of credit, never tokens. Nothing paid is unlimited: AI is bounded by credits and hosting by hosted-app count.

  • 01Credits in dollars, never tokens. Cost-based and shown as a balance.
  • 02Every ship is full proof. ShipGate is in every plan.
  • 03Local builds are unlimited. Paid AI and hosting are not, and we say so.
  • 04Limits are enforced by our server. No client can change them.
WholeStackReceipt · proof-carrying software
№ 0001
  1. /01

    Starter

    $25 of AI credits each month, 1 seat. Workstation, unlimited local deterministic builds, ShipGate, executable contracts, essential Volition enforcement, source export and one hosted production app.

    $49per month
  2. /02

    Pro

    $50 of AI credits each month, 3 seats and team collaboration. Everything in Starter.

    $99per month
  3. /03

    Ultra

    $90 of AI credits each month, 10 seats, team collaboration and Advanced Volition administration. Everything in Pro.

    $229per month
  4. /04

    Enterprise

    For larger teams and custom needs: more seats, custom terms and direct support. Everything in Ultra.

    Contact usfor pricing
  5. /05

    Extra hosted app

    One more hosted production app on an active plan. Buying it never starts a server on its own; you choose which app it hosts.

    $69per month, per app
  6. /06

    AI credit top-up

    Paid AI usage draws down dollar credits at our cost plus a stated margin. At zero, paid AI stops until you top up. Top-ups do not expire while your account is open.

    $10 · $25 · $50one-time

FAQ

Common questions.

How WholeStack builds your software, checks it, and keeps checking it.

How does WholeStack know what to build?

You answer plain questions about your business: what you keep track of, who does what, and what must never happen. WholeStack turns your answers into written rules, builds the app from them, and checks the app against them. You never write code.

How does ShipGate check the software?

ShipGate is separate from the part that builds. Before launch it tries to prove the app wrong: is everything you asked for there, do the buttons really do something, are the permissions tight, and does it work when it is actually running. Then it gives one answer, ship or don’t, with the evidence.

Can I bring an app I already have?

Its code, yes. WholeStack reads a GitHub repository you own, works out the records and workflows in it, and builds a new app from them, the same way it builds from a description. It cannot yet check software it did not build.

What is the pricing model?

Every plan includes the WholeStack workstation, unlimited local deterministic builds, ShipGate, executable contracts, essential Volition enforcement, source export and one hosted production app while the plan is active. Starter is $49 per month with $25 of AI credits and 1 seat. Pro is $99 per month with $50 of AI credits, 3 seats and team collaboration. Ultra is $229 per month with $90 of AI credits, 10 seats, team collaboration and Advanced Volition administration. Extra hosted production apps are $69 per month each. AI credit top-ups are $10, $25 or $50. Paid AI usage draws down dollar credits; hosting is limited by hosted-app count.

Get started

Software that proves itself

You get the app, the written rules it was built from, and independent proof that it follows them.

Let's build ↗Why we built it ↗