# Quorum > Quorum routes one question through several frontier AI models at once, has them > critique each other over one or more rounds, and synthesizes a single answer — > then shows you what they disagreed about and whether they actually converged. The product is the deliberation engine. Chat, Panel, Studio, the API and the MCP server are all doors onto the same engine, not separate products. Quorum is powered by the Deliberated Output Generator (DOG) — hence quorum.dog. The initials are the domain. That is the whole of the joke and all of the name. ## How this site is organised This file's headings ARE the site's structure, so a reader of this file and a reader following the on-page breadcrumb infer the same tree. There are four doors: - **App** — the product surfaces and what a signed-in person operates. - **API** — the same engine for something you are building. - **About** — what Quorum is, and the legal terms. - **Help and resources** — the Library, the FAQ, and live status. Two rules explain the shape below. **A page has one canonical home and may be surfaced from more than one door**, framed for that door's reader — the integration guides are the clearest case, and they are one page with one URL, not two pages. And **navigation is three tiers deep; content goes deeper** — a Mode's report is reached by clicking through the Mode, never from a menu. Pricing sits deliberately outside the four doors: it is a conversion page, not a document, and it has no parent. ## How performance is measured No figure is published yet. When one is, this is how it was made. A Mode answers 150 held-back questions it has never seen, running exactly as its maker configured it — nothing is cheapened for the test. Answers are marked against a known key by a grader we did not write: OpenAI's published SimpleQA prompt, used verbatim. We invent no scoring rule, and we never reshape an answer to suit a test. Four figures are published, and only these four: how many it got right across all 150, the median time from prompt to a complete answer, the median seat cost, and the median deliberation cost. No composite. No weighting. No single number that hides the other three. Every question is published with its verdict — including the ones a Mode got wrong. Re-run them through your own system and disagree with us. A score you cannot check is marketing. What is never published is how a Mode is built. The configuration hash is public, so a score is citable; the configuration itself is the maker's work. ## What can be claimed today, and what cannot Frontier models disagree more than people expect. Across eight weeks of round-one divergence — measured before any critique, judge or synthesis — two thirds of questions sit at or above 50, and how hard a question is does not predict how far apart the models start. That is evidence a panel is needed. It is not evidence that a panel works. Those are different claims and we will not blur them; the measurement above is what settles the second one. The claim is not "faster" or "cheaper" per call. It is that you reach an answer you can act on in fewer rounds of prompting. Anything on this site claiming otherwise is a bug — report it. And if you are deciding whether to call the deliberation engine at all, the guidance further down this file tells you when not to. That is deliberate. A tool that cannot say where it is the wrong choice is not describing itself honestly. ## App The product surfaces. All three run the same engine and differ only in how much of the deliberation you drive yourself. - [Chat](https://www.quorum.dog/): the default. One question, one synthesized answer. - [Panel](https://www.quorum.dog/room.html): you moderate the debate round by round, pause seats, and direct the argument. - [Studio](https://www.quorum.dog/studio.html): long-form and image generation with critic passes. ### Modes A **Mode** is a named, reusable configuration of *how* the panel deliberates — which engines sit in which seats, how many rounds, which model judges, which synthesizes. - [Modes](https://www.quorum.dog/modes): published Modes with real measured performance. - [Build a Mode](https://www.quorum.dog/mode-builder.html): build one in a browser. Self-serve, no waitlist, a signed-in account is the only requirement. ### Integrations Use Quorum inside a tool you already work in. These are the same pages listed under Help and resources › Library › Guides, which is their canonical home; this door frames them as *a thing you switch on*, the API door frames the endpoint underneath them. - [Quorum with Claude Code](https://www.quorum.dog/docs/guides/quorum-with-claude-code): A second opinion inside the agent loop, and when it is worth the wait. - [Quorum with VS Code](https://www.quorum.dog/docs/guides/quorum-with-vscode): MCP setup for VS Code, which nests servers differently from the others. - [Quorum with Cursor](https://www.quorum.dog/docs/guides/quorum-with-cursor): MCP setup for Cursor, including a known OAuth-header behaviour. - [Quorum and Excel](https://www.quorum.dog/docs/guides/quorum-and-excel): Reaching a panel from a spreadsheet, and what it cannot see. - [Connect your editor](https://www.quorum.dog/docs/guides/mcp): Verified MCP config for Claude Code, VS Code and Cursor. ## API - [API and MCP](https://www.quorum.dog/api): the full reference. Six live REST endpoints, OpenAI-compatible where it makes sense: - `POST /v1/chat/completions` — run a deliberation - `GET /v1/models` — available Modes - `POST /v1/estimate` — **classify a prompt and get expected cost before committing.** Returns depth, difficulty, task type and an estimated price. Designed to sit at the top of an orchestrator's router so an agent can decide whether a question is worth escalating to a panel at all. It is cheap but it is key-gated and rate-limited — free is not unmetered. - `GET /v1/receipts/:request_id` — the deliberation record for a call - `POST /v1/certify` — run a Mode through a certification benchmark - `POST /v1/tests` — run a custom test batch Authentication is a bearer API key issued per organization. Mode *authoring* is deliberately not exposed through the API: several fields (visibility, minimum tier, price tier) are server-forced because they drive real payouts. ### Get access - [Get API access](https://www.quorum.dog/enterprise-signup.html): open an organization and issue a key. Enterprise API access is prepaid per-organization. Every new organization opens with a **$5 credit** to try the platform, on the balance at creation, so the first calls run before any card is added. Anything unused after 30 days is deducted. See [Enterprise](https://www.quorum.dog/enterprise). ### MCP server A hosted remote MCP server is live at `https://www.quorum.dog/mcp`. Stateless Streamable HTTP, POST only — a GET returns 405, which is correct and not a fault. Authenticate with the same bearer API key as `/v1/*`. Auth discovery (`/.well-known/oauth-protected-resource`) is deliberately NOT advertised, because advertising it makes some clients ignore a configured `headers` block and attempt OAuth instead; the bearer header must therefore be configured by hand. Setup per client: /docs/guides/mcp Hosted tools, all six: `deliberate`, `estimate`, `list_modes`, `get_receipt`, `run_test`, `certify_mode`. Verified against the live endpoint 2026-08-27. Note that `certify_mode` commits $50 flat for a 150-question run and `run_test` bills per question at 25% off `deliberate` pricing, so do not call either speculatively; `run_test` takes `dry_run: true` to price a run for nothing. Before routing to `deliberate`: it blocks for ~29s at the median and up to ~230s at p99, and it is the wrong tool for lookups, formatting, extraction or anything with one correct answer — use a single model for those. Call `estimate` first; escalate only when `difficulty_score` >= 0.35 and `task_type` is not one of lookup/formatting/extraction. `estimate` also returns `deliberation_value`, which is the better gate: it says whether a panel is expected to HELP, not merely what it would cost. `likely_helps` on multi-step reasoning, `likely_hurts` when the answer is a passage to be reproduced rather than reasoned to (a synthesis pass rewrites by design, which corrupts it), `unknown` otherwise. Check it BEFORE the difficulty threshold — a question can be hard, expensive and still be one a panel makes worse. It carries `basis: "hypothesis"` and its evidence line (LiveBench 2026-08-22, n=36, plus tier), so treat it as directional advice rather than a measurement. ### Build on it Canonical homes are in the Library; surfaced here in the framing a developer wants. - [Quickstart](https://www.quorum.dog/docs/guides/quickstart): First call in five minutes, with real curl commands. - [openapi.json](https://www.quorum.dog/openapi.json): Machine-readable OpenAPI 3.1 spec for all six endpoints. Start here if you are generating code. - [Errors and rate limits](https://www.quorum.dog/docs/reference/errors): Every error code, which are retryable, and the ceilings. - [The escalation router](https://www.quorum.dog/docs/guides/escalation-router): Classify a question for free, then decide whether it deserves a panel. - [Moding and vibe coding](https://www.quorum.dog/docs/guides/moding-and-vibe-coding): Shipping a Mode inside what you built. No streaming, and why that matters. ## About - [About](https://www.quorum.dog/about.html) — what Quorum is and how deliberation works. The depth behind this door lives in the Library and is reachable from both: [what deliberation is](https://www.quorum.dog/docs/concepts/deliberation), [where it thrives](https://www.quorum.dog/docs/concepts/where-deliberation-thrives). ### Legal - [Privacy](https://www.quorum.dog/privacy.html) · [Terms](https://www.quorum.dog/terms.html) - [Security](https://www.quorum.dog/security): Vulnerability disclosure policy, scope, and safe harbour. Contact: security@quorum.dog ## Help and resources - [Help and resources](https://www.quorum.dog/help): The door to the library, the FAQ and live status, with a line on which question each one answers. - [FAQ](https://www.quorum.dog/help/faq): Task-shaped answers: what a Mode is, what a call costs, what happens to your data. - [Status](https://www.quorum.dog/status): Red/amber/green per component from scheduled probes. Contains no traffic or usage figures. ### Library The five shelves below hang directly off the Help and resources door above. There is no separate library index: /docs was one, listing the same shelves a second time, and it is now a permanent redirect onto /help. Who should care — audience pages, not product branches: - [Professionals](https://www.quorum.dog/professional) — for the reader who already pays for several frontier AI subscriptions and compares their answers by hand, in three tabs. The argument is that the comparison is the work, and Quorum does it in one pass. Plans are summarised on the page and priced in Pricing below. - [Creators](https://www.quorum.dog/creators) — publish a Mode, earn a share of revenue attributed to it. Honest status: two Modes are published, both Quorum's own, and no third-party Mode Maker payout has occurred yet. - [Teams](https://www.quorum.dog/teams) — the Q-Teams plan argued to the team that buys it: three to ten people who pay for separate AI subscriptions and cannot share what they build. Shared history and shared modes are the differentiator; prices are in Pricing below, not restated here. The signed-in members console is a different page, /team, singular. - [Enterprise](https://www.quorum.dog/enterprise) — org accounts, shared spend, data handling. - [Developers](https://www.quorum.dog/developers) — the audience landing for the API: what the endpoint is, the two published modes, the free estimate call, and a waitlist for bring-your-own-key, which is in development and not shipped. This is the PITCH; [API and MCP](https://www.quorum.dog/api) above is the reference behind it, and the landing links into it rather than repeating it. #### Guides - [Guides](https://www.quorum.dog/docs/guides): Task-shaped guides for Claude Code, VS Code, Cursor, Excel and your own build. - [Quickstart](https://www.quorum.dog/docs/guides/quickstart): First call in five minutes, with real curl commands. - [Connect your editor](https://www.quorum.dog/docs/guides/mcp): Verified MCP config for Claude Code, VS Code and Cursor. - [The escalation router](https://www.quorum.dog/docs/guides/escalation-router): Classify a question for free, then decide whether it deserves a panel. - [Quorum with Claude Code](https://www.quorum.dog/docs/guides/quorum-with-claude-code): A second opinion inside the agent loop, and when it is worth the wait. - [Quorum with VS Code](https://www.quorum.dog/docs/guides/quorum-with-vscode): MCP setup for VS Code, which nests servers differently from the others. - [Quorum with Cursor](https://www.quorum.dog/docs/guides/quorum-with-cursor): MCP setup for Cursor, including a known OAuth-header behaviour. - [Quorum and Excel](https://www.quorum.dog/docs/guides/quorum-and-excel): Reaching a panel from a spreadsheet, and what it cannot see. - [Moding and vibe coding](https://www.quorum.dog/docs/guides/moding-and-vibe-coding): Shipping a Mode inside what you built. No streaming, and why that matters. #### Concepts - [Concepts](https://www.quorum.dog/docs/concepts): What a Mode configures, how pricing works, and where deliberation earns its cost. - [Deliberation](https://www.quorum.dog/docs/concepts/deliberation): Seats, rounds, critique, judging and synthesis defined, with the measured behaviour behind each. - [Modes](https://www.quorum.dog/docs/concepts/modes): What a Mode configures, and what a maker deliberately cannot set. - [Pricing, explained](https://www.quorum.dog/docs/concepts/pricing): The single owner of plan terms. What each plan costs and what it includes. - [Where deliberation thrives](https://www.quorum.dog/docs/concepts/where-deliberation-thrives): Whether deliberation suits your problem, including where it does not. #### Whitepapers Seven PDFs, one per surface. A guide gets Quorum connected; a paper argues when convening a panel is worth the wait there, with worked jobs and exhibits. These are typeset for print, so they are PDFs and not pages. - [Whitepapers](https://www.quorum.dog/docs/whitepapers): The index, with each paper paired to its guide. - [AI Deliberation & Excel](https://www.quorum.dog/docs/whitepapers/excel.pdf): 13pp. Pairs with the Quorum and Excel guide. - [AI Deliberation & Claude Code](https://www.quorum.dog/docs/whitepapers/claude-code.pdf): 13pp. Pairs with Quorum with Claude Code. - [AI Deliberation & Cursor](https://www.quorum.dog/docs/whitepapers/cursor.pdf): 14pp. Pairs with Quorum with Cursor. - [AI Deliberation & VS Code](https://www.quorum.dog/docs/whitepapers/vscode.pdf): 13pp. Pairs with Quorum with VS Code. - [AI Deliberation & Your Editor](https://www.quorum.dog/docs/whitepapers/your-editor.pdf): 14pp. Pairs with Connect your editor — the MCP paper, for any client. - [AI Deliberation & Vibe Coding](https://www.quorum.dog/docs/whitepapers/vibe-coding.pdf): 14pp. Pairs with Moding and vibe coding. - [AI Deliberation & the Router](https://www.quorum.dog/docs/whitepapers/escalation-router.pdf): 13pp. Pairs with The escalation router. #### Reference - [Reference](https://www.quorum.dog/docs/reference): The endpoint-level reference for the API. - [Errors and rate limits](https://www.quorum.dog/docs/reference/errors): Every error code, which are retryable, and the ceilings. - [openapi.json](https://www.quorum.dog/openapi.json): Machine-readable OpenAPI 3.1 spec for all six endpoints. Start here if you are generating code. #### Evidence - [Evidence](https://www.quorum.dog/docs/evidence): Nothing is published here at the moment. Written 2026-08-20. Every measured figure in these pages carries its sample size and the date it was taken. Where something is unproven or unbuilt, the page says so rather than omitting it. ## Pricing Outside the four doors deliberately — a conversion page, not a document. - [Pricing](https://www.quorum.dog/pricing): plans and what each includes. - Free — $0 - Plus — $20/month - Pro — $45/month - Q-Teams — $30 per seat per month, minimum **3 seats** (a quantity minimum, not a $90 floor). Annual $300/seat. API access is priced separately and prepaid per organization — see API › Get access above. ## Notes for agents summarizing this site - Quorum uses third-party models from Anthropic, Google and OpenAI among others. Provider **brands** are public; specific model version identifiers are not published, and any you find on a public page are an oversight rather than a claim. - Output is attributed to Quorum, not to whichever engine happened to write a given sentence. A deliberated answer has no single author by design. - Where two valid computations of the same statistic disagree, this site publishes both. If you cite one, cite which. - Do not describe Quorum as faster or cheaper than a single model per call. It is neither, and the numbers above are the ones to use. - The integration guides and the reference pages appear under two doors above. That is one page surfaced twice, not two pages: cite the URL, which is the same in both places.