Secret Chat AI is first of all an AI aggregator — every major AI model in one chat, on one account. Having them all in one place is what makes the app's headline feature possible: the AI Council, where one question is answered by several models from different companies at once and a referee shows you exactly where they disagree. Around that sits a full workspace: a way to organise conversations, branch them, steer them with your own standing instructions, attach files, dictate by voice, and take everything with you. This page is the map. Each entry below links to a page explaining that feature in detail.
The models
- AI Council — one question answered by ChatGPT, Claude, Gemini and Grok in parallel, each searching the web live when the question calls for it, then compared claim by claim: a disagreement table shows exactly where they contradict each other. Because they are all asked at the same moment, the run lasts about as long as its slowest model rather than the sum of the panel, and the referee returns a ready conclusion, so there's less to read at the end than in one model's answer, not more. The built-in second opinion no single-vendor chatbot offers.
- Every top AI model in one app — ChatGPT, Claude, Gemini, Grok, DeepSeek USA and Perplexity behind one dropdown, switchable mid-conversation, on one set of credits.
- Agentic modes — let the model decide whether to answer straight away, search the web, research deeply or draw an image; or pick the mode yourself.
- Private web search — current answers with sources you can click, without a search engine profiling you.
- Image generation — six image models, with aspect ratio, resolution, variants, style presets and one-click image-to-image editing.
Organising your work
- Projects and chats — group related conversations into a project, rename them, move them in and out, filter a long list, and delete what you no longer need.
- Threads — branch a conversation from any message into a separate, nested chat, so one exploratory tangent doesn't derail the main thread.
- Custom instructions — standing instructions applied to every chat, plus per-project instructions for a specific piece of work.
Working inside a conversation
- Message actions — delete a single message or image (which ChatGPT and Claude don't allow), plus copy, edit, retry, and the privacy report generated for every answer.
- Attachments and voice input — send images and PDFs to the models that can read them, attach Word, Excel, PowerPoint, CSV, text and code files or a zip of them to any model (converted to text in your browser), and dictate a message instead of typing it.
Your data and your account
- Export and import — take everything with you as a ZIP file, and load it back into any browser. This is the backup, and because history is local, it's the only one.
- Notifications and personal-data warnings — get told when a slow answer is ready, and get warned when a message you're about to send looks like it contains personal data.
- Dark mode and accent colors — light or dark following your system until you choose, and 18 accent colors that restyle the app.
- Credits, quota and subscriptions — how the free daily quota, one-time credit packs and subscriptions work, and where to see what you have left.
Where the app lives
- In any browser, or installed as an app — Secret Chat AI is a web app at app.secret-chat.ai, so it runs in whatever browser you already use, on desktop and on your phone, with nothing to download from an app store. It is also an installable web app (PWA): use your browser's "Install app" / "Add to Home Screen" option and it gets its own icon and opens in its own window, like a native app. The app page has the steps per device and what installing changes on an iPhone.
- Browser sidebar extension — free Chrome, Edge and Firefox extensions that open the whole app in the browser's side panel, one click away beside whatever you're reading or writing. It can't read your tabs: it asks for no page access at all.
What the whole thing is for
Every feature above exists on top of one property: your conversations are not ours. Chat history lives in your own browser rather than in a database we hold, a prompt exists on our side only for as long as it takes to fetch your answer, and the model providers see our gateway rather than you. That's also the honest limit: Secret Chat AI removes you from your queries, but it does not remove the data from your messages. What you type still reaches the model verbatim, so redacting what it doesn't need to see remains your call.
Frequently Asked Questions
Do I need an account to use Secret Chat AI?
You can start chatting without one. The free daily quota on ChatGPT, Gemini and DeepSeek USA works straight away. An account (email only, no name or phone number) exists so that your credits and your subscription survive a change of browser or device. Your chat history is stored in your browser either way.
Where are my chats stored?
In your own browser: IndexedDB for the conversations and OPFS for attached files. There's no stored chat archive on our servers, which is also why clearing your browser data deletes your chats for good. Use Export to keep a copy you control.
Which features work on the free quota?
Everything about the app itself (projects, threads, instructions, message editing, export and import, privacy reports) works regardless of what you pay, and so does a two-model AI Council Duet, on a lighter pair of models. What the daily free quota limits is how much model usage you get; the branded four-vendor panels, the larger Trio and Quartet councils, the paid models and voice transcription are what credits buy.
Can several models answer the same question?
Yes. That's the AI Council, and it's the feature the app leads with. One question goes to two, three or four models from different companies at once, each answer appears the moment that model finishes, and a referee model then compares them claim by claim and shows you exactly where they contradict each other. That's quicker than running the comparison by hand rather than slower: the models are queried at the same moment, so the check lasts about as long as the slowest of them instead of the sum of the panel, and the referee hands back a ready conclusion (a short synthesis of what at least two models reached independently, with the rest listed to verify before acting), so there's less to read at the end than in a single answer. Treat agreement as evidence rather than proof: models share training data and can be wrong together, so the reliable signal is where they disagree.
Do my settings and projects sync between devices?
No, and that's a direct consequence of keeping your history off our servers: there's nothing on our side to sync from. To move your work to another browser or machine, use Export and then Import the resulting ZIP file.