The horizontal model for structured deliberation.
Submit an artifact or a question. H4 runs independent deliberations in parallel and returns one written report in minutes. The report is the deliverable; there is no chat.
H4 is not one system with one voice. It is a combinatorial parallel architecture: several independent reasoners take the same question, work it alone, then press on each other's disagreements. One voice writes the result.
Submit
You send an artifact or a question, and choose the lens. Nothing is trained on.
Independent positions
Several reasoners read it in parallel, each alone, so no single view dominates.
Cross-examination
They see each other's positions and press on the disagreements.
Structured report
One voice writes the result in minutes. You can leave and come back to it.
A deliberation takes one to three minutes. The report is the deliverable; you will not sit and watch a spinner.
One model, four lenses. Each tool takes a different kind of call; all of them return the same thing, a structured report you can defend.
RJE EU-resident
Reasons a difficult ethical question from distinct lenses grounded in European values: human dignity, fundamental rights, and the deliberative tradition. Returns a decision with the vote and the dissent on record.
Tetrad
Four independent experts assume your claim, design, or decision is false and try to prove it. Returns which claims survive and which break, ranked blocker, major, minor, each with the attack and a fix.
Gresol
Put code, a product, a page, or a contract in the crucible. A red-team panel attacks it the way a real adversary would and shows you exactly how it breaks, ranked by severity. Good-faith adversarial analysis, not an authorized penetration test.
The Chamber
Not a model, not a framework: an ACI. For the questions one mind cannot settle. Independent minds reason your question together, each keeping its own character, and return where they converge, the strongest objection, and a considered view.
Not sure which fits? Most calls have several dimensions. Start with The Chamber, or describe your case and we will route it.
| Tool | The question it answers | Typical input | The report emphasises |
|---|---|---|---|
| RJE | Is this the right thing to do? | A dilemma, its context and stakeholders | Decision, vote, dissent, reasoning trail |
| Tetrad | Is this claim true? | A claim, a design, a decision | What survives falsification, what breaks, with fixes |
| Gresol | How is this attacked? | Code, a product, a page, or a contract | Severity-ranked attacks, and the worst case |
| The Chamber | How should we read this? | An open question | Consensus, the objection, a considered view |
Payments API, POST /v1/refund
Cross-account crediting (IDOR)
The endpoint takes accountId from the request body and never checks the caller owns the order. Post a victim's orderId with your own accountId: the victim's charge is refunded and the credit lands on you. Fix: derive the account from the authenticated session; ignore any body accountId, and reject orders the caller does not own.
Replay with no idempotency (double refund)
No idempotency key, no "already refunded" flag, no row lock. Fire twenty concurrent requests for one order and each reads it as unrefunded, issuing twenty Stripe refunds and twenty credits (TOCTOU). Fix: compare-and-set the order status in one transaction (UPDATE ... WHERE status='paid') and send a Stripe idempotency key.
Dual payout: card refund and store credit
One reversal both refunds the original card via Stripe and adds equal store credit, a systematic double-dip that chains with the replay flaw. Fix: one payout path per policy, card refund or credit, never both, recorded in a ledger tied to the net captured amount.
Not used for training
Your submissions are used to produce your report and nothing else. They are not used to train any model.
Where it runs
RJE is processed and hosted in the EU. The other three tools run on a global panel; you are told before you submit, and you decide.
Retention you control
A report has a result link you can return to. You can delete a report and its content at any time.
Don't paste secrets you can't share
Treat a submission like sending a document to an outside reviewer. For code and red-team runs, leave out live credentials.
One subscription covers all four tools. A run costs one credit; The Chamber, a deeper deliberation, costs two. Large inputs use proportionally more. Start free, no card.
HQ is sold for professional use: solo founders, teams, and organisations.
- RJE, Tetrad, Gresol
- Up to 8k tokens / run
- On-screen report
- 20 credits / month
- All four tools
- Up to 30k tokens / run
- Email support
- 100 credits / month
- Up to 60k tokens / run
- 5 runs in parallel
- Priority support
- SSO / SAML, SLA
- EU-dedicated or on-prem
- Custom knowledge base
Credits from a subscription reset each month and do not roll over; purchased top-up credits never expire. RJE is processed and hosted in the EU; the other three run on a global panel. Nothing you submit is used for training. You can delete a job and its content at any time.