DeepSeek opens its in-house search engine to developers through Search and Fetch APIs

Search and fetch go generally available on the DeepSeek API

In its October 17, 2026 DeepSeek-V4.1-Flash-1017 release announcement, DeepSeek said the Search and Fetch APIs are now open to everyone. According to the announcement, the endpoints are powered by the company's in-house search engine, which serves hundreds of millions of requests a day worldwide. Search Engine Journal adds that the company's changelog says the engine indexes hundreds of billions of web pages.

The two endpoints do different jobs. Search finds web pages for a query and returns a structured list of results, while Fetch retrieves one page and returns it as Markdown text. DeepSeek's guide presents both as a live-web access layer for agents and RAG applications, and both use the same bearer-key authentication as the main DeepSeek API.

Every result carries an authority tier and a rerank score

Per the guide, a Search query must be 1 to 128 characters, one request can ask for 1 to 50 results, and the default is 10. Queries support the site:, -site:, lang:, after: and before: operators as well as exact-phrase matching. Each result includes a title, a URL, a short display summary (typically 100 to 200 characters), a longer model-facing summary (usually 600 to 800 characters), the site name, the publication date and, optionally, the page content.

Two fields stand out. The authority field gives a page's authority tier as an integer from 0 to 4, where -1 means unknown and 3 or above marks highly authoritative sources. The rerank_score is a 0 to 3 value that blends relevance, authority, freshness and similar signals, with 1.5 or higher counted as high quality. The service is paid: Search and Fetch are billed per successful request, only an HTTP 200 response that produces a result counts, empty results, errors and timeouts are not billed, and the number of results requested does not change the price. Server-side tool calls in DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until 16:00 UTC on October 31, 2026. Account-level limits are 100 requests per second for Search and 25 for Fetch, and Fetch has a separate per-domain limit. On the Fetch side, the guide says: "Fetching is subject to the target site's availability and access policies." Requests can be rejected by a security policy (403) or end in a 404 when site policy blocks them.

A new index that AI answers may draw on is now in play

As agents and AI tools spread, whose index the web passes through matters more. DeepSeek adds a new link to that chain with a direct search and fetch layer. According to Search Engine Journal, a site: query can return up to 50 pages from your own site, each with an authority tier and a relevance score, so checking how your site appears in this index is cheap to do. The same outlet notes that a missing page could be outside the index, caught by the changelog's filters or excluded for another reason, and that neither the guide nor the API reference offers a way to tell which.

Several points remain open. Search Engine Journal writes that the launch pages do not name the crawler behind the search index or the user agent Fetch uses, and we found neither name on the announcement or guide pages when we opened them. The same outlet reports that DeepSeek's privacy policy, last revised February 10, says third-party APIs are integrated for search and that users' input keywords are shared with those companies. A Chinese search API company, Bocha, claims in a GitHub README to be DeepSeek's official web search supplier, yet DeepSeek's launch pages do not mention it. According to SEJ, neither the changelog nor the privacy policy states whether queries still go to a third party. Finally, SEJ says the changelog credits the engine with screening out SEO spam, GEO manipulation and poor-quality AI content, without defining what GEO manipulation is; that claim is not on the announcement or guide pages and rests on SEJ alone.

The sources say nothing about Türkiye, so this section is commentary

None of the three sources mentions Türkiye, Turkish-language content or any Turkish institution. There is no Türkiye-specific availability or pricing, and no data on how well Turkish sites are represented in the index. This section is therefore not a fact drawn from the sources but explicitly UNALSOFT commentary.

Our reading: search and fetch layers now live in several indexes that do not depend on Google or Bing alone, and for a business in Türkiye, how it appears in them becomes something measurable. The unknowns, one by one: Türkiye-specific availability or pricing, how authority tiers are assigned, the name of the crawler behind the index, the user agent Fetch uses, and whether queries go to third parties. We covered another AI-search development on October 15.

Sources

DeepSeek API Docs (announcement) · DeepSeek API Docs (Search guide) · Search Engine Journal