Make a Mode

Research the engines, build the panel, test it in the Lab, list it and certify it. One path, in the browser or over MCP.

A Mode is a saved panel: which engines sit in which seats, how they argue, what a call costs. You build one, test it on questions with known answers, list it and certify it.

Each step works two ways. The Mode Builder does it in the browser. The 13 MCP tools at https://www.quorum.dog/mcp do it from an agent.

Saving a Mode needs an organization account. Publishing to the Marketplace uses the same organization.

1. Research

Decide what the Mode is for before you choose engines. One sentence is enough: "review commercial contracts for risky clauses", "check a data migration plan".

Then look at what you can seat.

In the Builder, ask the Mode Genie. The Research prompt asks which engines to consider for this Mode and why.
Over MCP, call list_engines. It returns the engines your plan or key can seat, each with its id, name, maker and minimum tier. The list stops at your entitlement.

2. Build

The Builder's Setup tab walks Brief, Panel, Staff, Depth, Price and Review.

Panel. Pick who sits in each of the 3 seats. If you hand-pick all 3 engines, they cannot all come from one maker. Choose a fixed pool, where you name the engines, or a dynamic pool, where Quorum's live ranking picks within your filters on each run.

Depth and critique. Set the round budget and the per-seat token ceiling. Set the critique policy to never, gated, or always. A gated policy runs critique on harder questions and skips it on easy ones.

Staff. Choose the engines that synthesise, judge and pre-score difficulty, or leave each on the default.

Review. Check the disclosure and save. A saved Mode is yours to run straight away. To share it with your organization, open Price in Setup and turn on Callable by API before you save. Your teammates can then call it by its ticker with your organization's API key or from a connected app.

The Auto seat

Put Auto in a seat and the Mode stores a rule where an engine would go. Auto seats need a fixed pool. An Auto seat resolves at call time to the best active engine inside the caller's tier ceiling, ranked by the AA score for the question's task type. With family diversity on, it skips engines from a maker another seat already holds.

Any one to 3 seats can be Auto. The AA score is the Artificial Analysis benchmark score. Each question is classed as one of general, coding, analysis, creative, explanation, strategy, math. Creative questions use the overall index.

Family diversity. Opt-in per Mode: off unless the maker ticks it, and only with at least one Auto seat. Engines you picked by hand count as taken.
An Auto seat takes no backup. If its engine fails, the call falls to the next-ranked engine inside the same ceiling.
If no engine is eligible for an Auto seat, the panel runs with fewer seats.
You set the staff roles yourself. A staff role cannot be Auto.

The Mode stores the rule, so the pick follows the ranking as engines change.

Build over MCP

An agent designs, saves and lists a Mode with these tools.

genie_propose_mode takes a plain-words description and returns a complete spec: name, ticker, seats and backups for every tier, judge, synthesis, depth, rounds, express lane, thinking level and price. Before it answers, the Genie runs the spec through the save's own checks and fixes what it can. If a refusal remains, the reply names it. The call saves nothing.
validate_mode checks any spec, yours or the Genie's, against the rules a save runs. It returns each refusal with its field and rule, plus the price floor and the recommended price. It is free and read-only.
save_mode saves the spec as a draft, which cannot run yet. Pass release: true to release it so it can run. An API key needs the modes:write scope. A connected app saves as you.
list_mode lists the released Mode in the Marketplace.

3. Test in the Lab

The Lab runs your Mode on questions with a known answer and grades each reply against the key. Your Mode's seat engines also answer the same questions alone, so the scorecard has a comparison.

In the Builder, open the Test tab. Choose the tiers, the question count per tier and whether to include punch-up, which fires a cheaper tier's Mode on the higher tier's questions. The Builder shows the price before you press Run.
Over MCP, call run_test with dry_run: true for the quote, then without it to launch. It returns a batch_id. Call it again with only the batch_id to poll until the scorecard is ready.
Call ask_genie for the next step. It reads the Mode's own Lab and certification results, names where coverage is too thin to judge (fewer than 5 answers), prices the next batch and says whether to certify. The limit is 30 asks per hour, 200 per day, per person (per org for an API key).
A quoted run starts when you call approve_quote with the quote_id from the Genie's reply. It launches once, at no more than the quoted price. If the price has risen, nothing launches and a new quote comes back.

A scorecard bar is drawn when the sample supports it. Smaller samples are listed with their count.

Who pays for a run

The person who launches it. With an API key, your organization pays. The quote is a maximum, and you are charged only for questions that produced an answer. Signed in through a connected app, your plan pays real cost on a Mode you or your organization owns.

4. List

In the Builder, press Publish, or call list_mode. Certification runs against a Marketplace listing, so list the Mode before you certify it.

A listing stays closed to others until the Mode is certified. A Mode that is already certified lists as open.

5. Certify

Certification runs a held-out question set against the Mode, graded the same way as the Lab. The certified card is pinned when the run is complete: every question answered and graded. An incomplete run pins nothing. An older batch does not replace the pinned one, so the card you have stays in place. Certifying a listed Mode approves its listing. An explicit rejection by Quorum still keeps it closed.

In the Builder, open the Certify tab.
Over MCP, call certify_mode, or ask the Genie for a quote with {"kind":"certify"} and approve it. The Mode's owner, or its owning organization, can certify it.

The certified card stays pinned until a newer certification replaces it. Editing the Mode leaves it in place. The card shows the date it was certified, and Lab runs never overwrite it.

Who can use a Mode

Who	Needs
You	A saved Mode. Run it in Chat, or call its ticker from a connected app. With your organization's API key, add `:LAB` to the ticker to run the Mode as last saved, or turn on Callable by API to call the ticker itself. A Mode saved over MCP or with an API key must be released first.
Your team	Callable by API turned on when you save. Teammates then call the ticker with your organization's API key or from a connected app.
Invited users	A grant from Quorum.
Everyone else	Anyone else can use a Mode once it is listed and certified.

Every Mode has a display name and a ticker in the form COMPANY:MODE. The Builder and the Marketplace show the name with the ticker. Receipts and the API show the ticker, and it is what you pass as model.

The paper

This guide is also a paper: Make a Mode (PDF).

More papers are on the whitepaper shelf.