Agentic search

Kimi's Agentic Search is built on an end-to-end agentic reinforcement learning architecture. Unlike traditional tool-orchestration approaches, it lets the AI autonomously decide when to search, which tools to invoke, and how to adjust strategy — enabling complex information gathering and processing tasks.

Compared to traditional keyword-matching search, Kimi's search capabilities have gone through two major evolutions:

Explorer Search (October 2024) combined large language model semantic understanding with real-time information retrieval, automatically decomposing complex questions and tracing sources.
Agentic Search goes further with autonomous planning and tool invocation, supporting financial data queries, academic literature retrieval, image search, and other multimodal tasks.

Web search settings

Web search (Auto/Off) breaks through knowledge freshness boundaries to deliver real-time, trustworthy, and traceable information.

Click the "Web search" toggle below the chat input box. When enabled, Kimi autonomously determines whether the current question requires real-time data and intelligently invokes search engines and vertical databases to retrieve and synthesize information.

Core capabilities

1. Freshness meets authority

Real-time web retrieval covering 100+ verified, trustworthy sources (major news outlets, government announcements, financial data platforms, academic journal databases) — ensuring information is both current and authoritative.

2. Source attribution

All search-based answers include reference source links. Click through to view the original page and fully verify accuracy and context.

3. Global information access

Supports natural-language conversations in multiple languages. Kimi can proactively search non-English sources (e.g., Japanese technical documentation, French news) and automatically synthesize multilingual information.

Recommended use cases

Real-time news and updates (breaking news, stock market data, policy changes)
Fact verification (online rumors, historical data, statistical methodologies)
Professional research (competitor intelligence, academic frontiers, industry standards)

Authoritative source badges

Kimi's web search filters sources across three dimensions — relevance, authority, and freshness — automatically filtering out risky sites, duplicate syndication, low-quality web content, and hallucinated information to ensure reliable results.

High-quality sources that pass Kimi's strict screening are marked with authority badges in responses (e.g., official website certification, academic certification, media certification icons), helping you quickly identify highly trustworthy information.

Source screening mechanism:

Filtered out: Low-quality aggregator sites, unverified self-media, expired cached pages
Prioritized: Government agency websites, authoritative media, academic journals, official financial reporting channels

Financial & academic data sources

In professional scenarios, Kimi supports querying vertical-domain databases for structured, traceable professional data.

Simply state your request in the conversation, and Kimi will automatically select the appropriate data source.

Example queries

"Look up Apple's revenue data for the past five years"
"Search for the latest papers on large language model inference optimization"
"Compare GDP, GDP per capita, and unemployment rates of the world's top three economies"

Supported data sources

Source	Description
Global Finance Data	Global stock, futures, index and other financial data
Academic Data	Journals, papers, preprints, dissertations, patents, etc.
Stock Finance Data	Chinese and global stock, futures, index and other financial data
World Bank Data	GDP, population, employment, trade of countries, etc.

Smart search

Image search

Kimi supports image-based search and understanding. After uploading an image, Kimi can automatically invoke image search tools to identify the content and retrieve related information.

Example queries:

"Where is this location?"
"What's happening in this meme / what is this person saying?"
"Find the real McDonald's"

How to use:

Upload an image directly (supports JPG, PNG, and other formats)
Describe the image content or ask image-related questions in your prompt
Kimi will combine image recognition with web search to provide image origins, similar images, and related information

Typical scenarios:

Identifying unknown objects, landmarks, or product models
Finding the source or origin of an image
Analyzing data in charts or screenshots and retrieving relevant background information

Direct URL access

Kimi Search supports direct URL access to fetch internet resources.

Paste a URL directly into the conversation, and Kimi will automatically fetch and analyze the page content.

Example use cases:

Summarize the key points of this article
Extract the key data from this earnings report PDF
Search for image information from the URL

Tips

Prefer publicly accessible links (e.g., news pages, public PDF documents)
Some password-protected or crawl-restricted pages may not be accessible
For dynamically loaded pages, Kimi may only retrieve the initial HTML content
Kimi cannot re-read shared conversation links (i.e., links to other Kimi chats)