DeepSeek has handed developers the building blocks of its chat app. Searching the web, opening a pasted link, reading an uploaded file and running safety checks can now be switched on through the API calls apps already make, with DeepSeek doing the tool juggling behind the scenes.
The change log carries the date 13 November. DeepSeek says the app experience is just these parts working beside the model, so an outside developer can now put together something close to it without writing the plumbing.
Search comes in two flavours. With thinking off, the model decides whether a question needs a web or image lookup and cites what it finds. With thinking on, it keeps searching and following links over several steps. Pasted links get a third switch, deepseek_read_url.
The pieces are also offered alone: Search and Fetch, OCR, speech recognition (ASR) and text-to-speech (TTS). DeepSeek’s Search and Fetch guide describes Search as returning structured results and Fetch as returning a page in Markdown. Both are charged for each request that succeeds, with no token billing, and empty searches are not charged. Image search is not in the standalone Search API yet and works only through the web tools.
DeepSeek puts the engine behind all of it at hundreds of millions of daily requests, and says it also tuned its Code Agent and Search Agent and improved Chinese search. This follows the 17 October release that first opened the search API to developers, so a team that wants a DeepSeek-style assistant no longer needs a separate search vendor to get live web results. The release note itself lists no prices for the new standalone endpoints.
