Frequently asked questions

It sends one question to several AI models at the same time, lays their answers side by side, and then has a mediator model synthesize where they agree, differ and contradict each other — a tool for thinking and deciding.

Where the models disagree, you can have them debate each other directly to dig into the real crux. You're not getting a single answer; you're mapping the landscape of views your decision needs.

One model's answer is one model's point of view. Models often reach different conclusions on the same question, and coThink surfaces those differences on a single screen so you can cross-check them and decide. No more hopping between browser tabs pasting the same question into each service.

OpenAI · Google · Anthropic · Perplexity · xAI — pick 2 to 5 models per question from 25 models across 5 brands. You choose each brand's default model and variant, and you can also switch the mediator model that handles synthesis.

Newly released models are reviewed and added after a weekly check.

Once every model has answered, you can request a synthesis: the mediator model you picked lays out what they agree on, each model's unique insight, where they contradict each other, and a synthesized answer. You can also steer it with a note such as "focus on the cost side".

When the models disagree, the participants talk to each other — rebutting and revising each other's points — while a moderator model frames the crux and delivers a conclusion you can decide on. After every exchange you choose whether to continue, stop or conclude, and you can step in with a comment at any point.

It's where explorations that people chose to publish are collected. Anyone can browse the Garden without signing in and see the questions where the models' judgments split.

Find one you like and you can take it straight to your own pick of models — only the question comes across, never the original answers.

Start straight away with a Google account, or sign up with an email and password. If you sign up by email we'll send a confirmation message — click the link in it to finish.

You can use coThink as soon as you sign up; there's no approval to wait for.

Using coThink requires agreeing to the terms of service and the collection of personal data. You confirm that once, right after signing up, and the survey that follows is optional.

Answer the survey (5 questions, about 30 seconds) and we'll thank you with 100 seeds. Choosing 'later' still lets you use the app immediately, and you can take the survey any time from the user menu.

Each account has exactly one sign-in method. Whichever you use first becomes that account's method, and you can't add or switch to another one later.

Your explorations and seeds are tied to the account, so we don't offer account merging. Pick the method you'll want long-term when you sign up.

There's no monthly subscription — coThink runs on prepaid seeds, and only your actual AI usage is deducted from them. When you sign up you get 50 welcome seeds (valid for 30 days from the day they're granted), plus 100 more for completing the onboarding survey — those never expire.

It depends on how many models you pick and how long your question and attachments are. A typical three-model question runs roughly 30–40 seeds.

Before an expensive action we show an estimated range up front, and you can see what you actually spent on each question card badge and in your seed history. Per-model rates are listed in the price reference inside the app's model settings.

Use the balance chip in the sidebar, or the user menu → Seeds. Packages start at ₩7,700 including VAT, and payments are handled by our payment provider Paddle, which shows the amount in your local currency at checkout. Your first top-up earns a one-time +300 seed bonus.

Unused seeds are fully refundable within 7 days of the top-up. After 7 days the unused balance is refunded minus a small handling fee. Seeds granted for free (welcome and bonus seeds) are not refundable. See Article 8 of the Terms of Service for the full conditions.

To generate answers, your question and attachments are sent to the AI providers you selected (OpenAI, Google, Anthropic and others). Exactly what goes where is listed in full in the cross-border transfer table in our Privacy Policy.

No. We do not use the contents of your threads to train AI models or for marketing.

No. We don't store the original attachments on our servers — only minimal metadata such as the file name and format. The originals stay in the browser you're using.

Anyone with the link can read a shared thread without signing in. Search engine indexing is blocked, and turning sharing off cuts access to the link immediately.

Sharing and publishing are separate — creating a share link never puts a thread in the Garden. Publishing is a separate switch, and published threads can be indexed by search engines.

The question, every model's answer, the synthesis and the debate — all of it. Anyone can read it without signing in and search engines may index it. Information that identifies you, such as your account or email, is never published with it.

That said, anything you typed into the question or answers yourself is published as-is. It's worth a quick reread before you switch it on. The full terms are in Article 10 of the Terms of Service.

Yes. Switch publishing off and it comes out of the Garden right away — the detail page stops opening too. Switching it back on does not re-publish it automatically.

Content search engines have already crawled can take a while to disappear from their results — that part is up to the search engine, so we can't promise a timeframe.

You can delete your account yourself from the account screen; your account and threads are removed immediately and cannot be recovered. Only payment and refund records are kept in separate storage for 5 years as required by Korean e-commerce law, then destroyed.

Yes. You can attach images (PNG, JPG, WebP) and documents (PDF, DOCX, TXT, MD) to a question, and the microphone button transcribes what you say straight into the input box.

With it on, coThink runs a web search when you ask and distributes the fresh results across the models. That improves accuracy on questions about facts after their training cut-off, and because each model gets different sources, the answers stay more varied too.

Long answers can hit a model's output limit. Truncated answers show a "Continue" button — press it to fetch the rest.

No. Switching publishing on makes it a candidate, and only selected explorations are planted in the Garden. Once it goes up, the button changes to "Published" and a link to its card appears.

There are conditions for what can go in the Garden — no attachments (images or documents), at least one synthesis, answers from 2 or more models, and a question of at least 15 characters.

You can still tap the dimmed button — it tells you which condition is in the way.