DeepSeek Launches Search and Fetch APIs for AI Agents
New web-access tools enable real-time data retrieval for RAG applications at scale.
1Real-time Web Access
DeepSeek has introduced Search and Fetch APIs designed to provide agents and RAG applications with live web content. Both tools function as raw JSON APIs, utilizing the same bearer authentication as the standard DeepSeek API. These endpoints allow developers to integrate dynamic external information directly into their AI-driven workflows.
2Search API Functionality
The Search API allows users to query the web with specific syntax, including site filters and date operators. Developers can define the result count and request content extraction within the JSON payload. A successful request returns an HTTP 200 status code along with an array of results containing snippets and metadata.
3Fetch API Capabilities
The Fetch API is dedicated to retrieving page content from specific URLs, returning data in Markdown format. This tool is subject to target site availability and access policies. If a page lacks extractable content or is blocked by security policies, the API returns a 404 error, indicating the retrieval failure.
4Billing and Pricing Structure
Starting November 1, 2026, at 00:00 Beijing time, these services will be billed at $0.75 per 1,000 successful requests. Billing is strictly based on successful HTTP 200 responses that return data. Failed requests, empty search results, and timeouts are not billable, ensuring developers only pay for actionable information retrieved.
5Error Handling Protocols
The APIs follow an OpenAI-compatible error format, providing clear status codes for various failure scenarios. For instance, HTTP 429 indicates a QPS limit breach, while 402 signifies insufficient account balance. Developers are advised to retry 408, 429, or 5xx errors, but avoid retrying other 4xx errors without modifying the request parameters.
6Rate Limiting Mechanisms
Search and Fetch APIs operate under independent account-level QPS limits, meaning traffic to one does not consume the quota of the other. Additionally, the Fetch API enforces a per-domain QPS limit to prevent site overloading. Requests exceeding these thresholds will trigger HTTP 429 errors, requiring developers to implement appropriate backoff strategies.
7Integration and Tooling
When using the DeepSeek API through platforms like DeepSeek Harness, Claude Code, or Codex, server-side tool calls automatically utilize these Search and Fetch APIs. While these calls are currently free, they will transition to the standard billing model on November 1, 2026. Developers should monitor their usage patterns accordingly.
8Strategic Implementation Impact
By providing raw JSON access to web data, these APIs empower developers to build more context-aware AI agents. The ability to combine search operators with structured content extraction simplifies the development of complex RAG pipelines. This standardized approach ensures consistent performance and predictable costs for production-grade AI applications moving forward.