A single HTML file to chat with 16 AI providers, run sourced deep research, create and edit images, make videos, voice-overs and podcasts, and read your PDF, Word and Excel files. No installation, no middle server, no account to create: your keys stay encrypted in your browser.
🔔 Get ChatJPT updates by email
Use your own API keys. Providers are grouped by country, so you can also choose based on where your data is processed.
Everything runs right in your browser: ChatJPT talks to the AI providers without going through a third-party server.
API keys are encrypted (AES-GCM) and stored in your browser. They are only sent to the matching provider, never to the author. One button clears every key at once, handy on a shared computer.
Ask two models the same question, from different providers if you like, and see their answers side by side. Keep the best one in a click; it carries on the conversation.
For models that reason, their thinking appears in a collapsible block before the answer. Adjust the depth or effort per model, or turn it off for faster answers.
Turn on web search for up-to-date answers with clickable sources. Paste a link: the model actually reads the page instead of guessing what it says. With Grok, search X posts too.
Hand a complex question to Gemini Deep Research, Qwen Deep Research or Grok's agents: they search the web for several minutes and write a structured report, with its sources. You can edit the proposed plan before the research starts.
When speed matters, turn on Claude's fast mode, OpenAI Ultrafast or Grok priority processing. The extra cost is shown next to each switch.
Create images with OpenAI GPT Image (animated GIF included), Google Nano Banana (panoramic formats), Qwen Image, Wan (up to 4K), Grok Imagine, MiniMax, Recraft (vector SVG included) or Bria: several variants at once, styles, formats and sizes of your choice. Images and audio are kept in your browser.
Make videos with Google Gemini Omni, Grok Imagine, Alibaba Wan and HappyHorse or MiniMax Hailuo, from a text or your photos: first frame, first and last frame, or reference characters and objects, with sound. Extend a video (up to 40 s with Omni) or change one element with a sentence.
Under each image: edit it with a sentence, erase an object with the brush, mark up or comment on a precise spot, remove the background, change the format or animate it into a short video. The Recraft and Bria menus add vectorizing, upscaling, old photo restoration, colorizing and product photos.
Turn text into speech with MiniMax, Google Gemini, Alibaba Qwen, Grok or OpenAI: hundreds of voices in dozens of languages, easy to find by language, gender or use, with emotions, tone and effects (laughs, whispers…). Design, remix or clone your own voice (Gemini, MiniMax, Alibaba).
Attach a document or start from the conversation: a model writes the script of a two-host podcast (or a monologue), you review it, then it is read with two voices. You get an audio file ready to play.
Attach text, code, images, PDFs, Word or Excel files. Claude and Qwen can read PDFs page by page, charts included, and Claude gives the page of each quoted passage. Scanned PDFs are transcribed once by the AI, and the estimated cost is shown before it starts.
Vibe Coder mode guides beginners step by step to build a page or a tool without knowing how to code. Code mode offers syntax highlighting, file downloads and a live preview of HTML pages, in a secure sandbox.
Ask for a chart: bars, lines, pie, histograms, scatter plots, treemaps, heatmaps… It appears right in the conversation.
Multiple conversations in a side panel, search within the text, edit a message you already sent, regenerate with another model, export to Markdown, JSON or PDF.
Each answer shows the tokens it used, the part read from cache (much cheaper) and, with Grok, its exact cost, with a total per conversation. When something fails, the provider's exact message is shown so you understand the problem right away.
Interface in English, French and Spanish, with a light or dark theme. The AI answers in the language of your message, and your preferences are remembered.
Nothing to install: ChatJPT fits in a single file.
Get ChatJPT_1_35_2.html on GitHub, then open it in Chrome, Edge, Firefox or Safari.
Paste the key of at least one provider (OpenAI, Gemini, Claude…). The included guide explains how to get one, often free to start.
Pick a model and a mode, then start typing. You only pay for what you actually use, directly to the provider.
Don’t miss a thing: subscribe to be the very first to know about upcoming ChatJPT releases and features.
Answers to the questions we hear most often about ChatJPT.
Yes. ChatJPT is freeware: the file is free, with no subscription and no account. You only pay the AI providers you use, directly and per use, with your own API keys.
No. ChatJPT fits in a single HTML file. Download it on GitHub, then open it in Chrome, Edge, Firefox or Safari. No installation, no ChatJPT account.
Your keys are encrypted (AES-GCM) and stored in your browser. They are only sent to the matching provider, never to the author. This encryption prevents the keys from being read directly; it does not replace the security of your computer. On a shared computer, one button clears every key at once.
No. There is no middle server: your browser sends each message straight to the provider you chose. Your conversations stay in your browser, and you can export them to Markdown, JSON or PDF.
The price depends on the provider and the model. A typical conversation usually costs a fraction of a cent; images cost more, and video even more. ChatJPT shows the tokens of each answer and a total per conversation. Some providers offer a free trial, such as Cohere (1,000 calls per month) or Upstage ($10 of credit).
Create an account on a provider's developer console (OpenAI, Anthropic, Google…), open the "API Keys" section and create a key. The key guide published with ChatJPT on GitHub lists each provider's address, in English, French and Spanish.
16 providers, grouped by country: OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Perplexity and Meta (United States); Augure and Cohere (Canada); Mistral AI (France); Recraft (United Kingdom); Bria (Israel); DeepSeek, Kimi, MiniMax and Alibaba (China); Upstage (South Korea).
Yes. Providers are grouped by country to make that choice easier. For example, Augure and Cohere are Canadian, and Mistral AI is French. Alibaba even offers keys per region, including the United States and Hong Kong.
Yes. Image mode works with OpenAI, Google, Grok, Alibaba, MiniMax, Recraft and Bria; Video mode with Google, Grok, Alibaba and MiniMax; Voice mode with MiniMax, Google, Alibaba, Grok and OpenAI, up to two-host podcasts. Each image can then be edited, cut out or animated.
Text, code, images, PDFs, Word and Excel files. Scanned PDFs can be transcribed by the AI, with the estimated cost shown before it starts. Claude and Qwen can also read a PDF page by page, charts included.
The interface is available in English, French and Spanish, with a light or dark theme. The AI answers in the language of your message.
The version number is shown at the bottom of the app. New versions are published on GitHub with the list of changes: download the new file, then open it like the previous one.
You can use ChatJPT and share it freely. However, its license (CC BY-ND 4.0) does not allow distributing a modified version. This rule protects users: a modified copy could, for example, send your keys somewhere else.
Developed by Jean-Philippe Thériault | License: Free / Freeware (free to use)