Every question you send to Multisis goes through the same two-stage process. This post walks through it in order, starting with how you set up the council.
Setting up your seats
Open Model settings and you will see four seats: three Thinkers and one Judge. For each seat you choose:
- A provider. Pick from the free providers in the list, or point a seat at any OpenAI-compatible endpoint with the custom option and its base URL.
- A model. Choose from the suggestions, or press Load live models to fetch the provider's current list and pick from it.
- An API key. Each seat has its own key field, and you can reveal or hide what you type.
- A specialty (Thinkers only). Give each Thinker a field such as reasoning, coding or research, so it answers with the right expertise.
To run the app you need a key for the Judge and at least the minimum number of Thinkers. The settings panel shows your progress, for example "Judge ✓, Thinkers 2/3", until you are ready to save.
A few conveniences are built in:
- Use Thinker 1's provider & key for all fills every seat from the first one, which is handy if you only have one key.
- Mixing providers spreads the free-tier rate limits, so one provider's quota is less likely to slow the whole council down.
- A seat warns you if a key doesn't start with the prefix its provider normally uses, which catches most paste mistakes.
- Some providers can't read images. If a seat uses one, the panel tells you that seat answers from text only.
Stage one: three thinkers
Your prompt, along with your recent conversation, is sent to your Thinkers at the same time. Each one answers through its own specialty, so the drafts overlap as little as possible.
Because they run in parallel, you wait for the slowest thinker rather than all three added together.
Stage two: the judge
Your Judge model reads the thinkers' drafts and does the editing work:
- Resolves points where the drafts disagree
- Removes repetition
- Organizes the result into one answer
- Streams it back to you as it writes
Staying in control
Nothing is hidden. Open any turn to see what each thinker wrote and which model played which role. You can also stop generation at any point, and you can reopen the settings whenever you want to swap a model or change a specialty.
When something fails
Providers go down and requests time out. Multisis is built to degrade instead of break:
| Situation | What you get |
|---|---|
| One thinker fails | The judge works from the remaining drafts |
| Several thinkers fail | The judge answers from whatever is left |
| The judge fails | The most complete thinker draft, plus a note on what happened |
Your keys
You bring your own provider keys, so you always decide which models fill each role. Keys are encrypted and saved to your account rather than to your browser, and they are sent only to the provider you chose for that seat.
For your security, saved keys are never shown again. Leave a key field blank to keep the saved one, or paste a new key to replace it. If you want them gone, Clear saved keys removes every saved key from your account.
Want more detail? Read why use your own API keys, whether your keys are safe, and our recommended combo of free models. See the privacy policy for how keys and chats are handled.