How MySmartCoPilot uses AI

Three ways AI runs

1. On your device. Some browsers have AI built in (for example Chrome’s Gemini Nano, which Chrome downloads and manages itself). Other tools download an open AI model from MySmartCoPilot’s model server into your browser, only after showing you its size and asking. The model stays in your browser for next time; you can see and delete it on Cookies & storage. Nothing you type, paste or open is sent anywhere.
2. On MySmartCoPilot’s cloud AI, only if you choose it. When your device can’t run a model, a tool may offer to send your request to MySmartCoPilot’s server, where openly available AI models that our infrastructure provider runs for us produce the answer. It shows you exactly what will be sent, with personal details such as email addresses and phone numbers replaced by placeholders, and checks that you are not a bot. MySmartCoPilot keeps neither your request nor the answer (see AI features in the privacy policy).
3. Your browser’s speech features. Tools that listen or read aloud may use your browser’s own speech service. Some browsers send the audio or text to their maker for this; the tool says so and asks first.

If none of these can run, the tool offers its basic mode without AI, labelled “Basic mode (no AI)”, or tells you plainly that its AI is not supported on your device.

A few image and audio tools from before the AI platform, such as Background Remover, Object Remover (Magic Eraser), AI Image Upscaler and Audio to Text Transcriber, instead run an AI model of their own in your browser: they download it from mysmartcopilot.com itself the first time you use them, after showing its size, and your browser keeps it with the site’s other files. Your photos and recordings never leave your device, and these tools never use server AI.

Models that can run on your device

Each model is a fixed version, checked against the publisher’s files and copied to MySmartCoPilot’s model server, so your browser never downloads it from anywhere else. Sizes are the full download, before the AI runtime (a few MB, downloaded once).

MySmartCoPilot’s cloud AI

When you choose it, a tool’s request goes to MySmartCoPilot’s cloud AI: openly available AI models that our infrastructure provider runs for us in its data centres. They rewrite, summarise and draft text, answer questions about what you give them, translate, turn speech into text and text into speech, describe and generate images, and check requests and answers for safety. Tools say what the cloud AI will do rather than which model does it. What every tool promises instead:

• you see exactly what will be sent before anything leaves your device, and nothing is sent unless you agree;
• MySmartCoPilot keeps neither your request nor the answer;
• our provider processes your request only to produce the answer, and under its terms does not use it to train AI models or to improve its services;
• every result says that AI made it.

The use rules that come with these models are part of our AI rules, and the notices their licences ask for are on the open-source licences page.

Free daily limits

MySmartCoPilot’s server AI is free and stays free: it runs only on the free daily allocation our infrastructure provider gives us, which we count in “AI units”, and MySmartCoPilot uses at most 9,000 of them a day for everyone together. It is shared, so it can run out: when it does, server AI stops until 05:30 IST (00:00 UTC) and the on-device options keep working. Each AI tool shows an estimate of the uses left for you today.

On top of that, one visitor can use at most 400 AI units a day across all tools, one connection 1,200, and the site owner can lower any of these limits. A short rewrite costs about 5 to 7 AI units; an image description or a story checked by the safety model about 40 to 50.

Safety rules

• MySmartCoPilot’s AI refuses sexual content and anything sexual involving minors, nudify and “undress” edits, deepfakes and face swaps, voice cloning, impersonating real people, brands or officials, fake documents, reviews or engagement, malware, and help with violence, weapons or self-harm. If you mention self-harm, the tool stops and shows where to get help, such as Tele-MANAS (14416).
• Requests and written answers are checked against these rules before you see anything. Open-ended writing, image descriptions and image generation are also checked by an automatic safety model, and open-ended writing and image generation stay switched off until they pass a safety review.
• AI tools never give medical, legal or financial advice: they explain neutrally and point you to a professional. They do not offer AI “detectors”, humanisers or tools to get around safety checks.
• The safety filter can block harmless requests too. If it blocked yours by mistake, report it (below).

Labels and marking

• Every AI result carries a label such as “AI-generated — check before use” and says where it ran: on your device, with the model’s name, or on MySmartCoPilot’s cloud AI.
• Files you download from an AI tool carry a note in their metadata that AI made or edited them; AI images and voices also carry a visible label that tools do not remove. Text you copy is never marked with hidden characters.
• AI makes mistakes. Check results before you use them, and don’t rely on them for medical, legal, financial or safety decisions.

Report a problem

If an AI tool produced something harmful or wrong, or blocked a harmless request, tell us through the contact form (topic “Report a problem with AI output”). Each AI label has a “Report a problem” link that opens the form with this topic chosen and the tool and recipe names filled in (and the model’s name, when the AI ran on your device). It never attaches what you entered or the AI’s answer, so describe what went wrong. Illegal content can also be reported to the authorities, in India at cybercrime.gov.in.