A detailed explanation on how to get started and how our system works.
Getting Started with Multisis
Multisis lets you ask one question and get one considered answer built from several AI models. Instead of trusting a single model, your question goes to three specialist "thinkers," each taking a different angle. A separate "judge" model then reads all three write-ups and merges them into a single, organized response.
This guide walks you through everything from creating your account to reading your usage dashboard.
Contents
- How Multisis works
- Before you begin
- Create your account
- Add your API keys
- Choose your models
- Ask your first question
- Attach images and files
- Read and control the answer
- Manage your conversations
- Track usage on your dashboard
- What happens when something fails
- Tips for better answers
- Frequently asked questions
- Next steps
1. How Multisis works
Every question you ask goes through three stages:
| Stage | What happens |
|---|---|
| You ask | You type a question and can attach images or files. Your recent conversation is passed along as context, so follow-up questions make sense. |
| Three thinkers respond | Three specialist models work on the same question at the same time, each from a different angle (for example career, learning curve and use case). |
| A judge writes the answer | A separate model reads every write-up and merges them into one organized response. |
Why this helps: different angles are compared and combined, so gaps or blind spots in a single model's answer are more likely to be caught. It is not a guarantee of a perfect answer, and for very simple questions one model may be enough, but for decisions, comparisons and open-ended problems a panel often gives a fuller picture.
2. Before you begin
You will need:
- A Multisis account. You can sign in with Google or with your email.
- At least one API key from a supported provider, such as OpenRouter, OpenAI or GitHub. Multisis uses your keys, so you pay your providers directly and choose which models run.
- About two minutes for setup.
Tip: If you are not sure where to start, a single OpenRouter key gives you access to many models, including free and low-cost ones, through one account.
3. Create your account
- Go to the Multisis home page and select Get started (or Create your account).
- Choose to sign in with Google or enter your email.
- Complete any verification step shown on screen.
Already have an account? Select Log in instead.
Because your keys are tied to your account, you can sign in from any device later and keep working without re-entering them.
4. Add your API keys
An API key is a private code that lets Multisis call a model provider on your behalf. You set this up once.
- Open the keys or settings area after signing in.
- Pick a provider (OpenRouter, OpenAI, GitHub and others).
- Paste your key and save it.
- Repeat for any other providers you want to use.
Where do I get a key? Create one in your provider's own dashboard, then copy it into Multisis. Each provider has its own sign-up and billing rules, so check their pricing pages.
Keep your keys safe:
- Never share a key publicly or paste it into screenshots, chats or code repositories.
- If you think a key has leaked, delete it in your provider's dashboard and add a new one.
- You can use separate keys for separate purposes so that one can be revoked without affecting the rest.
5. Choose your models
Multisis gives you control over which model does which job. You can pick the provider and model behind:
- Each of the three thinkers, and
- The judge.
Ways to set it up:
- Mix providers for variety. Different model families tend to reason differently, which is the whole point of having a panel.
- Use a stronger model for the judge. The judge reads everything and writes the final answer, so a capable model here often improves the result.
- Start cheap. Free or low-cost models for the thinkers are a good way to try Multisis without a large bill.
You can change these choices whenever you like.
6. Ask your first question
- Open a new conversation.
- Type your question in the message box.
- Send it.
Behind the scenes, the three thinkers start working at the same time and the judge assembles the final response.
Good first questions are ones with more than one reasonable angle, for example:
- "Should I learn Rust or Go first as a backend developer?"
- "What are the trade-offs between renting and buying a home in my situation?"
- "Help me plan a launch for a small side project."
Follow-ups work naturally. Your recent conversation is passed along as context, so you can say "make that shorter" or "what about the second option?" without repeating yourself.
7. Attach images and files
You can attach images and files to a question. Every thinker receives the same material, so all three angles are based on the same source.
Useful for:
- Asking about a screenshot, chart or diagram.
- Getting feedback on a document or draft.
- Comparing information across files.
Note: Which file types and sizes work depends on the models you choose. If a model cannot read a certain type of attachment, try a model that supports it.
8. Read and control the answer
- Live streaming. The final answer appears as it is written, so you can start reading right away.
- Stop at any point. If the answer is going in the wrong direction, stop it and rephrase your question. You do not have to wait for it to finish.
- Organized output. The judge merges the thinkers' write-ups into one structured response rather than three separate ones.
9. Manage your conversations
Chats are saved to your account. From your conversation list you can:
- Reopen a conversation and continue where you left off.
- Rename it so it is easy to find later.
- Delete it when you no longer need it.
10. Track usage on your dashboard
Every call is logged so you can see where your tokens go, which models are fast, and when a provider is close to its limit. The dashboard is available after you sign in.
| Section | What it shows |
|---|---|
| Tokens over time | Usage per hour for the last 24 hours, or per day for 7 and 30 days. |
| Cost saved | A comparison between what you spent and what the same tokens would have cost on a reference model you choose. |
| Latency per model | p50 and p95 response times, time to first token and tokens per second for each model. |
| Rate-limit gauges | Remaining requests and tokens from each provider's latest response, plus a count of 429 (too many requests) errors. |
A note on "cost saved": this is an estimate. It assumes the same tokens would have run on the reference model you pick, using list prices. Some token counts may be estimated by the provider, so treat the figure as a guide, not an invoice. Your provider's billing page is the source of truth for what you were actually charged.
11. What happens when something fails
Models and providers sometimes error out, hit rate limits or time out. Multisis is built so one failure does not ruin your whole answer:
- If a thinker fails, you still get a result built from the thinkers that finished.
- If the judge fails, you get a result from the work that did finish.
- Failed calls are recorded on your dashboard so you can spot patterns.
If you see repeated failures:
- Check the rate-limit gauges on your dashboard for 429 errors.
- Confirm your key is still valid and has credit or quota.
- Try a different model or provider for the affected role.
12. Tips for better answers
- Be specific. Include your goal, constraints and context. "I am a backend developer who wants to get hired in six months" beats "which language is better?"
- Ask for trade-offs. Panels shine on questions where reasonable people disagree.
- Give the judge a strong model. It writes what you read.
- Mix model families. Variety in the thinkers is what catches blind spots.
- Use follow-ups. Narrow down the answer instead of starting over.
- Stop early when needed. If the first lines show a misunderstanding, stop and rephrase.
- Verify important facts. Combining models reduces some errors but does not remove them. Check anything high-stakes against a reliable source.
13. Frequently asked questions
Do I need my own API keys? Yes. Multisis uses your keys (OpenRouter, OpenAI, GitHub and others), so you pay your providers directly and choose which models run.
What happens if one model fails? You still get an answer built from the work that finished. The failed call is also recorded on your dashboard.
Is the cost-saved figure exact? No, it is an estimate based on list prices and the reference model you choose. Some token counts may be estimated by the provider.
Can I stop an answer midway? Yes. The final answer streams as it is written, and you can stop it at any point.
Can I use free models? Yes, if your provider offers them. The Multisis blog has notes on getting useful answers from free and low-cost models.
Are my keys available on other devices? Your keys are tied to your account, so you can sign in from any device without entering them again.
14. Next steps
- Ask a real question you are currently weighing and compare the final answer with what a single model tells you.
- Experiment with models by swapping one thinker at a time to see how the result changes.
- Check your dashboard after a few conversations to understand your usage and costs.
- Read the blog for notes on multi-model reasoning, free models and how Multisis assembles a final answer.
Ready? Log in now and start playing.