Give AIrealtime webaccess
Append :online to any model name and the AI will search the web before answering, fetching the latest info
Get started in 3 seconds
Pick a model and append :online
glm-4.7-flash:onlineAsk as usual — the AI decides whether to search
"What's the tech news today?"
AI searches the web → injects results → answers from the latest info
Examples
All models support the :online suffix, including the free route
Append the suffix to any model to enable web search
model: glm-4.7-flash:onlineGood for: news, realtime data, latest updates
The free entry also supports web search
model: dxnt.com/free:onlineGood for: trying web search for free
Use in Cursor / Claude Code and similar tools
model: glm-4.7-flash:onlineGood for: looking up latest docs, API changes, technical solutions
Specify the model name in the API request body
{"model": "glm-4.7-flash:online",
"messages": [...]}Good for: custom apps, bots, automation
How it works
Four steps to web search, fully transparent and seamless
Extract keywords
Intelligently extracts search keywords from your last message, automatically stripping redundant prefixes like "look up" or "search for"
Search engine query
Calls a search engine (Bing / DuckDuckGo / custom API) to fetch the 5 most relevant results
Inject context
Appends the search results (title + summary + source link) to your message as reference for the AI
Generate answer
The AI answers based on the search results, citing sources for accuracy and traceability
Supported search engines
Three engines to choose from, configured by admins in the backend
Bing China
Direct connection in China, no proxy needed. Fetches results via HTML scraping and auto-filters Microsoft's own links. Great for everyday use in China.
DuckDuckGo
A privacy-focused search engine. Requires a SOCKS5 proxy in China. Fetches structured results via the official API.
Custom API
Connect any search API (Google, SerpAPI, Tavily, etc.) via a {query} placeholder for flexible configuration.
Auto-trigger
No extra command needed — adding the suffix triggers a search
All models
Works with any model, no type restriction
Smart cleaning
Auto-strips redundant prefixes for precise keywords
Traceable sources
Results include title, summary and source links
FAQ
How is search billed?
Each successful search costs ¥0.10 (deducted from your balance). The search results are injected into your message as context and counted as normal Token usage.
Do search results consume Tokens?
Yes. The search results (about 5 entries, each summary capped at 300 characters) are appended to your message as context and consume input Tokens. Use it when you need realtime info; no need for everyday chat.
Are the results accurate?
The system fetches real results from the search engine and auto-filters noisy links. The AI answers based on the results and cites sources. Quality depends on the search engine and keyword precision.
Is streaming supported?
Fully supported. The search runs at the start of the request (about 1-3 seconds), then the AI streams the answer based on the results — same experience as a regular chat.
Are all protocols supported?
Yes. OpenAI Chat Completions, OpenAI Responses API and Anthropic Messages all support the :online suffix. It works in Cursor, Claude Code, Cline and other tools.
How are search keywords extracted?
The system automatically takes your last message as the search term and smartly strips redundant prefixes ("look up", "search for", "please", etc.) and trailing punctuation for more precise searches.