Replacing an official web UI with a compatible provider
Enter the Base URL and key of an OpenAI-compatible provider such as DeepSeek or SiliconFlow in the settings, and you get a chat page that leaves no conversation records behind.
Loading…
Guide
AI Chat talks to any OpenAI-compatible model straight from your browser using your own API key: requests go to the endpoint address you configured under Settings ({Base URL}/chat/completions) with an Authorization: Bearer header, and the answer arrives as an SSE stream rendered token by token. The tool provides no model itself and never relays traffic through this site's servers.
Updated 2026-09-093 min read
AI Chat talks to any OpenAI-compatible model straight from your browser using your own API key: requests go to the endpoint address you configured under Settings ({Base URL}/chat/completions) with an Authorization: Bearer <your key> header, and the answer arrives as an SSE stream rendered token by token. The tool provides no model itself and never relays traffic through this site's servers.
Unlike the AI Copywriter, this is open-ended multi-turn conversation: the system prompt and the entire message history are sent together, whereas the copywriter runs a single task template.
| Input | Output | Notes |
|---|---|---|
Preset “Chinese–English translator”, send: 这个方案下周才能定稿 |
The proposal won't be finalized until next week. |
The preset's system prompt asks for the translation only |
Send any message after entering a wrong key |
Red bubble: Endpoint returned 401: Incorrect API key provided: sk-*** |
Error text redacts the secret to sk-*** automatically |
/chat/completions that supports stream: true, and must allow cross-origin requests from the browser (CORS); otherwise the tool reports “network request failed”.Enter the Base URL and key of an OpenAI-compatible provider such as DeepSeek or SiliconFlow in the settings, and you get a chat page that leaves no conversation records behind.
Pick the Senior programmer preset for technical questions and get a conclusion first, then runnable code; switch to Academic editor and paste a paragraph from a paper.
Change the model name in the settings (deepseek-chat, qwen-plus and so on) and compare answer styles with the same prompt.
No. Requests go from the browser directly to the endpoint address you entered, and a line under the input box always shows “Will be sent to <endpoint domain>”.
In this session's memory only by default, and cleared on refresh; only turning on Remember key in the settings writes it to this machine's localStorage. The endpoint address and model name are always saved.
Providers place their Base URL at different levels, and some need the /v1 segment. Follow the documentation and enter the level directly above chat/completions.
The system prompt, message history and your API key are sent by the browser directly to the endpoint address you configured (api.openai.com by default) and do not pass through this site's servers. The key is kept in memory only until you turn on Remember key, which writes it to this machine's localStorage. Do not submit sensitive information in the conversation.
Updated 2026-09-09
Talk to any OpenAI-compatible model with streaming, role presets and Markdown rendering
AI chat needs an OpenAI-compatible API configured
Fill in the endpoint (Base URL), the API Key and the model name on the Settings page.
Works with any OpenAI-compatible service such as OpenAI / DeepSeek / Qwen / Kimi / SiliconFlow,
DeepSeek, SiliconFlow and other Chinese providers give free credits when you sign up.
The key is kept only for this session (cleared on refresh) and is never uploaded to any server; you can enable "Remember the key on this device" in Settings to keep it (stored in plain text — do not enable it on a shared device)