How the room works

One question goes to the models you seat. They answer alone, talk once, and then one of them writes the takeaway. Every step below is one request per seated model, made from this browser tab.

1 · Independent. Round one is blind.

No model is told that anyone else is answering, or who. Nobody postures, defers to whoever seems stronger on the topic, or leaves a gap for someone else to fill. Each looks for the most consequential insight it can defend, in about three sentences.

2 · The conversation.

The room is introduced here, together with the other answers. Each model challenges, connects, or revises in one short, substantive exchange. If it has nothing to add, it says so instead of repeating itself.

3 · Synthesis, then parting thoughts.

The synthesis job rotates so every seat takes a turn, unless a model asks for it. The synthesizer gets fresh context and the whole record. It leads with what the room discovered, not a recap, and does not average anyone into blandness. If the conclusion isn't ready, it can ask for one more exchange. Then each of the other models may add one parting line, like a footnote, or pass.

Each model keeps its own conversation.

A seat's memory is its own turns plus what it was shown. It never sees its own answer played back as if someone else had said it, and never sees a round it missed.

A cut-off answer is a failure, not a short answer.

It is marked failed and never passed to the others as a position. If a model drops out, the room says who is missing and carries on with the rest.

With five seated, a run is about fifteen requests (three per model), billed to you by OpenRouter. Attached documents are read by every model in each of its turns, so large files raise the cost. Unseat anyone you don't want in the room.

Where your data goes

Four claims, and what makes each one true rather than promised.

Your question goes straight to OpenRouter.

The Room's server delivers this page and handles optional sign-in. A guest discussion never passes through it. Your browser sends the question, the answers and your key directly to openrouter.ai. The page's security policy only allows connections to OpenRouter and our sign-in service.

Files are read in your browser.

When you attach a document, this tab extracts its text. That text goes to OpenRouter with your question; the file itself is never uploaded to The Room.

Your key stays with you.

It lives in this tab's memory, or in this browser's storage if you tick the box. It is attached only to requests to openrouter.ai. Downloads and reports are built here in the page and saved straight to your disk.

Requests require zero-data-retention routes.

Each request asks OpenRouter for providers that keep no copy of it. If a seated model is unavailable, OpenRouter may answer with GLM-5.3 Flash under the same rule, and the transcript names the model that actually answered.

What OpenRouter and the model vendors do with a request is between you and them. It is an ordinary API call, and their policies apply.

Your OpenRouter key

One key reaches every model in the room. Get one at openrouter.ai/keys. This tab sends it only to openrouter.ai, in the Authorization header. The Room's server never receives it, and OpenRouter bills you directly. Enter it here, not in .env.local.

Keep it in this browser so I don't have to paste it again

Without the box ticked, the key lives only in this tab's memory.

Sends each seated model a one-word test (a fraction of a cent).

What shall we discuss?

Claude Haiku 5.5

GPT-6 Luna

GPT-5.4 Mini

Gemini 3.8 Flash

DeepSeek V4.1 Flash

0/6000

Click to seat or unseat · 5 of 6 seats filled (at least 2)

Invite a model

Every OpenRouter model that can run on a zero-data-retention route, so it keeps no copy of your conversation. Models that store or train on data aren't listed. Prices are per million tokens (input / output), from OpenRouter's list.

Loading OpenRouter's list…