On November 13, 2026, DeepSeek announced all-in-one app APIs in its API changelog: the web search, file input, link reading, voice interaction and safety capabilities familiar from the DeepSeek app, including the web version, are now open to developers. Developers no longer have to wire up tool calls themselves; switching on the corresponding parameters in the existing API is enough, and the server handles tool invocation and context assembly. The conversational capabilities run on the deepseek-flash model, while voice capabilities are offered as standalone interfaces.
One switch turns on five capabilities
Outside thinking mode, turning on the deepseek_web_search tool lets the model run web or image searches on its own in response to a question and answer with citations; in thinking mode, the same tool lets it search the web and read links by itself across multiple reasoning steps. File input accepts PDFs, Office documents, images and other formats, extracting plain text or parsing mixed text and images before merging the result into the conversation context. Link reading relies on the deepseek_read_url tool: the model spots links in user messages, fetches the pages and answers using the content together with the conversation. Model safety is a separate switch: with the safety check parameter on, DeepSeek applies basic risk controls, reviews model inputs and outputs, and returns a refusal flag when a risk is hit. Services offered to the public must switch it on, and providers remain responsible for content safety review, model registration, AI labelling and similar obligations.
Not just a bundle: the underlying capabilities can be called on their own
Beyond the packaged conversational capabilities, search, fetch, OCR, speech recognition and speech synthesis can each be called as a standalone interface. DeepSeek has also built an open-source demo chat client on top of these APIs that can import DeepSeek conversation history, so developers can compare it against their own implementations. Our earlier guide to using DeepSeek's search and fetch APIs covers usage and billing for the search and fetch pair. This update follows the October 17 opening of DeepSeek's self-developed search API: search on both the DeepSeek website and the API runs on the same in-house engine, which the company says handles hundreds of millions of search requests a day worldwide across an index of hundreds of billions of pages.
What this API saves first is the most tedious layer of chatbot development: deciding what to search, what to read and how to assemble context, without developers writing the orchestration logic themselves. The trade-off is narrower choice: the capabilities currently sit only on the deepseek-flash model, and source selection and citation formatting follow DeepSeek's defaults. Pricing and usage details are on the official 'Models & Pricing' page and the 'Build a Chatbot' guide; the changelog itself does not expand on them. What to watch next is whether other models will get the same capabilities, and how the standalone interfaces are billed.
