DeepSeek Launches Search and Fetch APIs for AI Agents
New web-access tools enable real-time data retrieval with a per-request pricing model.
1Real-time Web Access
DeepSeek has introduced Search and Fetch APIs designed to provide agents and RAG applications with live web content. These tools function as raw JSON APIs, utilizing the same bearer authentication as the standard DeepSeek API. Developers can now integrate real-time external data directly into their existing AI workflows and automated agentic systems.
2Search API Functionality
The Search API allows users to execute queries with specific search syntax, including operators like site filtering and date constraints. A successful request returns an HTTP 200 status with a results array containing metadata such as rerank scores, snippets, and summaries. Empty search results do not trigger billing, ensuring cost efficiency for developers.
3Fetch API Capabilities
The Fetch API retrieves page content and converts it into structured Markdown text. This tool is subject to target site availability and specific access policies. If a URL is blocked or lacks extractable content, the API returns a 404 error. It provides a clean, readable format for AI models to process web-based information.
4Per-Request Billing Model
DeepSeek implements a pricing model of $0.75 per 1,000 successful requests. Billing is strictly tied to successful HTTP 200 responses that return actual data. Failed requests, timeouts, or empty search results are not charged. This structure decouples costs from token usage, focusing instead on the successful retrieval of external web information.
5Implementation and Integration
Developers must send their API key and JSON content type in the request headers to interact with these endpoints. The system supports standard HTTP methods and follows OpenAI-compatible error formats for easier integration. Unknown fields in the request body are ignored, allowing for flexible query construction without breaking the API request structure.
6Rate Limiting Mechanisms
Search and Fetch APIs are subject to independent QPS limits calculated at the account level. Beyond these general limits, the Fetch API enforces a per-domain QPS restriction to prevent site overloading. Requests exceeding these thresholds receive an HTTP 429 error, necessitating a retry strategy for robust application performance in high-traffic environments.
7Transition to Paid Access
While current tool calls via DeepSeek Harness, Claude Code, or Codex remain free, the transition to the paid billing model begins on November 1, 2026, at 00:00 Beijing time. Developers should prepare their infrastructure to handle the new cost structure for these tools before the official implementation date to avoid service disruptions.
8Operational Error Handling
The API provides specific HTTP status codes to diagnose failures, such as 401 for invalid keys or 402 for insufficient balance. For transient errors like 429 or 5xx, developers are advised to implement retry logic. However, 4xx errors—excluding 408—should not be retried without modifying the request parameters to ensure successful future execution.
Original article: Hacker News — DeepSeek Launches Search and Fetch APIs for AI Agents