Give AI Real-time Web Access
Add :online suffix after any model name. AI auto-searches the internet before responding, delivering the latest information.
Three-Second Setup
Pick any model, add :online after its name
glm-5.2-flash:onlineAsk as usual, AI auto-decides whether to search
"What tech news today?"
AI searches the web → injects results → answers with the latest info
Use Cases
All models support :online suffix, including Free Route
Add :online suffix to any model to enable web search
model: glm-5.2-flash:onlineBest for: News, real-time data, latest updates
Free route also supports web search
model: aiapitoken.com/free:onlineBest for: Free users to try web search
Use in Cursor / Claude Code / Cline and other tools
model: glm-5.2-flash:onlineBest for: Latest docs, API changes, technical solutions
Specify model name in API request body
{"model": "glm-5.2-flash:online",
"messages": [...]}Best for: Custom apps, bots, automation flows
How It Works
Four steps for web search, completely transparent and seamless
Extract Keywords
Smartly extract search keywords from your last message, auto-remove redundant prefixes like 'search for' or 'look up'
Search Engine Query
Call search engine (Bing / DuckDuckGo / Custom API), get the 5 most relevant search results
Inject Context
Append search results (title + summary + source links) to your message as reference info for the AI
Generate Response
AI generates response based on search results, cites sources, ensuring accurate and traceable info
Supported Search Engines
Three engines available, configured by admin
Bing China
Direct connection in China, no proxy needed. Gets results via HTML scraping, auto-filters Microsoft links. suitable for daily use in China.
DuckDuckGo
Privacy-focused search engine. May need SOCKS5 proxy in some regions. Gets structured results via official API.
Custom API
Supports any search API (e.g. Google, SerpAPI, Tavily), with {query} as a flexible placeholder.
Auto Trigger
No extra commands needed, just add :online suffix
All Models Support
Every model can use it, no restrictions
Smart Cleaning
Auto-removes redundant prefixes, extracts keywords precisely
Traceable Sources
Results include title, summary, and source links
FAQ
How much does search cost?
$0.10 per successful search (deducted from balance). Search results are injected as context into your message, and token usage is calculated normally.
Do search results consume tokens?
Yes. Search results (~5 items, ~300 chars each) are appended to your message as context, consuming input tokens. Use :online when real-time info is needed; no need for everyday chat.
How accurate are search results?
The system fetches real results from search engines and auto-filters noise links. AI generates responses based on search results with citations. Quality depends on search engine and keyword accuracy.
Does it support streaming output?
Fully supported. Search executes at request start (~1-3 seconds), then the AI streams its response based on results - same experience as normal chat.
Which protocols are supported?
OpenAI Chat Completions, OpenAI Responses API, and Anthropic Messages Protocol all support the :online suffix. Works in Cursor, Claude Code, Cline, and other tools.
How are search keywords extracted?
The system auto-extracts your last message as the search query, and intelligently cleans redundant prefixes (like 'search for', 'look up') and trailing punctuation for more accurate results.