Documentation & Help Center

Clearafi asks the same question to several leading AI models at once, compares their answers, and shows you where they agree and where they disagree. Instead of trusting one model's word, you see the consensus and the conflicts.

Any single AI model can state something wrong with complete confidence. When several independent models are asked the same question, the points they all agree on are far more likely to be reliable, and the points they disagree on are exactly the ones worth checking yourself.

A percentage from 0 to 100 showing how strongly the models agreed on the core claims in their answers. It is calculated from the individual claims, not from an overall impression: each claim is marked as agreed, partially agreed, or disagreed, and the score follows from that.

After the models answer, an independent judging step breaks the topic into core claims. "Agreed" means every model supported that claim. "Partial" means they broadly aligned but differed in emphasis or detail. "Disagreed" means they genuinely conflicted. The counts appear next to your score.

A single combined answer written from all the models' responses. It takes the strongest points from each, states shared conclusions confidently, and presents genuine disagreements openly rather than hiding them.

No. It means the models agreed. Models can share the same blind spot or the same outdated training data, so agreement raises confidence without guaranteeing accuracy. Treat a high score as a reason to worry less, not as proof, and use the listed sources for anything important.

No. You can ask a question without signing up. An anonymous conversation is limited to two turns — your opening question and one follow-up — after which you will be asked to create a free account to continue.

Running several AI models on one question has a real cost. The anonymous allowance lets you see exactly what Clearafi produces before deciding, and a free account lifts the per-conversation limit immediately.

No. The conversation you started anonymously is attached to your new account when you sign up, so you can carry on from where you were.

Ask as you would ask a knowledgeable person: full sentences, and any context that matters — your location, your timeframe, your constraints. Comparison questions ("X versus Y") and decision questions work especially well, because that is where models most often differ.

Choose Sign up and either register with your email address or continue with Google, GitHub or Apple. There is no separate approval step — your account is ready immediately.

Email, Google, GitHub and Apple.

Yes. You confirm you are 18 or older when you create your account.

Account emails, such as billing receipts and security notices, are sent to the address you register, so use one you can actually receive mail at.

Sign in the same way you signed up. If you registered through Google, GitHub or Apple, use that same provider each time — it identifies your account.

Yes, Clearafi delivers mail to Apple's private relay addresses. Note that Hide My Email creates a distinct address, so signing in later with your real Apple address may look like a different account.

Confirm you are using the same method you registered with, that you are on the right address, and that you have no ad or privacy blocker interfering with the sign-in window. If you still cannot get in, email help@clearafi.ai from — or quoting — the address you registered.

The registered address identifies your account and is not changed from the app. Email help@clearafi.ai if you need it moved.

The interface is available in English, Spanish, French, German, Portuguese (Brazil), Italian, Japanese, Hebrew and Hindi.

Yes. You can ask in any language the models understand, which is far more than the nine the interface is translated into. Your answer comes back in the language you asked in.

From the language selector in the app. Your choice is remembered.

Yes. The PDF report is produced in your selected language, including right-to-left layout for Hebrew.

That should not happen — an answer is meant to mirror the language of your question, even when the sources it read were in another language. Please report it to help@clearafi.ai with the question you asked.

A panel of leading models from several independent providers. The exact line-up is shown in the model picker and changes as models are added, upgraded or retired, so the picker is always the accurate list.

Yes. The model picker lets you select the panel for your question. If you do not choose, Clearafi selects a suitable panel for you.

Between two and five. Two is the minimum, because a "consensus" of one model agreeing with itself is not a consensus. On the free plan a panel can include up to three models.

A model may be unavailable in your region, may not support what your question needs, such as live web search, or may be temporarily unavailable from its provider. The picker shows what you can use right now.

Availability depends on your plan and your region. Some models are hosted in jurisdictions that are not offered to users in certain territories, so the picker adapts.

Clearafi retries once, then continues without it and tells you which model dropped out. As long as at least two models answered, you still get a consensus. Below two, Clearafi tells you rather than presenting a single answer as a consensus.

Yes. For questions that need current information, models that support live web search will search before answering, and the sources they used are listed with your answer.

The web pages the models actually consulted while answering. Each entry links to the page so you can verify a claim at its origin.

No. Web search is a capability of the model, not of Clearafi, and not every model has it. For a question that needs current data, Clearafi favours models that can search, and tells you if a model on your panel could not.

Approximate location is used to make location-dependent answers correct — for local prices, opening hours or "near me" questions. If your question names a different place, the place in your question wins.

When the models disagree substantially, Clearafi automatically runs a second round in which each model sees the others' answers and can revise its own. It often resolves differences that came from a model simply missing something.

Automatically, when the first-round consensus score falls below 60. You do not need to request it.

It can. You are shown both the original score and the post-debate score, so you can see whether the discussion actually moved the models closer together.

Yes. Both rounds are kept and remain visible, so you can compare what each model said before and after seeing the others.

Yes. A conversation keeps its context, so follow-ups build on what has already been said rather than starting over.

On the free plan the daily limit counts new conversations, not turns. Follow-ups within a conversation you have already started are unlimited.

In your history, which lists your conversations with an automatically generated title so you can recognise them at a glance.

They are generated from the actual gaps and disagreements in your answer, so they tend to point at what would most improve it. Tap one to send it.

Yes. Use the microphone in the question box and your speech is transcribed into text you can review and edit before sending.

Up to 30 seconds per recording, which is ample for a spoken question.

Voice needs microphone permission, and support varies between browsers. Check that Clearafi is allowed to use your microphone, and try a current version of Chrome, Edge, Safari or Firefox.

Yes. Upload a file and the models will answer from its contents as well as their own knowledge.

PDF, Word (.docx), Markdown and plain text. Legacy .doc files are not supported — save as .docx or PDF first.

Files are limited to 2 MB. The amount of extracted text that can be used is also capped, and the cap is larger on a paid plan than on the free plan.

The most common cause is a scanned or image-only PDF with no selectable text. If you cannot select text in the file, Clearafi cannot read it either.

Yes. You can supply a URL and Clearafi will fetch the page text for the models to work from.

Yes. Every answer has a share link that opens the full result, including the consensus score, the curated answer and each model's response.

A PDF report, a Word document (.docx), a spreadsheet (.csv) and Markdown. The PDF is the presentation-ready report; the Word file is editable; the spreadsheet contains the claim-by-claim breakdown.

Download the Word file and choose Open with Google Docs, or the spreadsheet and open it with Google Sheets. Copy Answer also pastes into Google Docs with its formatting intact, including tables.

One row per core claim, with its agreed, partial or disagreed status and, for every model on the panel, that model's position and the exact words it used. It is built for sorting and filtering.

Yes. The report renders the languages Clearafi supports, including right-to-left scripts.

Three new conversations a day, with unlimited follow-ups inside each one, and a panel of up to three models per question.

Once a day. If you have used all three conversations, your allowance returns the following day.

Yes. Free answers use web search where the question needs it, and the sources are listed.

No. It is not a trial that expires.

Explorer at $9.99 a month, Insight at $18, Strategic at $30 and Executive at $100.

Monthly capacity, and nothing else. Every paid plan has the same features, the same models and the same quality — a higher plan simply includes more monthly usage credit. Pick the one that matches how much you expect to use.

Your monthly credit instead of a three-a-day limit, panels of up to five models, and a larger allowance for document uploads.

Choose a plan from the plans screen and complete checkout. An upgrade takes effect immediately.

Both are handled from your profile. A cancellation keeps your plan active until the end of the period you have already paid for.

You keep your current plan until the end of the paid period, and the new, lower plan starts at the next renewal.

Cancelling stops future renewals and your plan runs to the end of the period you have paid for. Email help@clearafi.ai about anything unusual on your account.

Payments are processed by Stripe, which accepts major cards. Clearafi never sees or stores your card details.

Emailed to your registered address when each payment is taken.

Paid plans include a monthly amount of usage credit. Each answer uses a small part of it, based on the models on your panel and the length of the work.

Because they are not the same amount of work. A five-model panel costs more than a two-model panel, a long document costs more than a short question, and a question that triggers a debate round runs the panel twice.

No. Plan credit is for the month it was granted and a fresh allowance arrives at renewal.

You can buy a top-up to keep going without changing plan, or move to a larger plan if you are consistently running out.

Top-up credit is consumed within the current billing period and does not carry past your renewal, so buy what you expect to use.

No. Top-ups are for subscribers. On the free plan, your allowance resets daily at no cost.

Yes. Share your personal referral link and earn free months when the people you invite become paying subscribers.

Five referrals earn two free months, ten referrals earn four free months and twenty-five referrals earn six free months. Six months is the maximum.

When the person you referred has held a paid subscription for 30 days. That waiting period is what keeps the programme honest.

In your profile, alongside a count of how many referrals you have earned and how many free months you have left to use.

Open your profile from the account menu. It shows your plan, your remaining monthly credit, your usage, and your account controls.

Yes. Your display name is editable from your profile and is used in the app interface.

Your profile shows your remaining credit for the current billing period and your recent activity.

Yes. Any conversation in your history can be deleted individually.

Yes. Conversations can be pinned so they stay easy to find in your history.

Your question is sent to the AI providers on your panel so they can answer it, and your conversation is stored in your account so you can return to it. Do not paste secrets, credentials or other people's personal data into any AI tool, Clearafi included.

Your history is yours. An answer becomes visible to other people only if you choose to share its link.

Every answer can be downloaded as a PDF, a Word document, a spreadsheet or Markdown, so your results are yours to keep. There is no self-service bulk export of the rest of your account data yet — email help@clearafi.ai if you need a copy and we will arrange it.

From your profile. Deletion is confirmed explicitly and then scheduled, with a 30-day grace period during which you can cancel it and keep everything.

Your account and its data are permanently deleted. That step cannot be undone.

No.

Yes. Create an API key in your profile under API Keys, then call Clearafi from your own software to request a multi-model consensus. Keys are named so you can tell them apart, and any key can be revoked from the same screen.

Yes. Clearafi runs an MCP endpoint at https://app.clearafi.ai/mcp, authenticated with your API key as a Bearer token, so a compatible assistant can request a consensus as a tool. Your profile has ready-made configuration for Cursor, Claude Code and Claude.ai — copy it, paste in your key, and you are connected.

Yes. Each answer can produce an embeddable badge showing its consensus score, linking back to the full result.

A full run asks several models, waits for the slowest, and then judges and writes the combined answer, so a couple of minutes is normal — longer if a debate round is triggered. The progress panel shows the current stage.

The models' individual answers are kept even when the combined step fails, so nothing you waited for is thrown away. You can retry the analysis without re-running the whole panel.

That happens when fewer than two models produced a usable answer. Clearafi shows you what it did get and labels it clearly, rather than presenting one model's answer as a consensus.

Sources are the pages the models actually read, and pages move or disappear. If the link fails, the page may be in a web archive.

Please report it to help@clearafi.ai and include the link to the answer. A wrong quote, a wrong attribution or a wrong number is exactly what we want to hear about — real examples are how this improves.