DeepSeek says the capabilities behind its consumer app, from web search to file reading to voice, are now available through its API. A developer can switch them on with a few parameters, or call separate services for OCR, speech-to-text and text-to-speech, and build a chatbot that behaves like the DeepSeek app.
- DeepSeek announced on 13 November that its app features are now open through the API: search, file input, link reading, safety checks and voice
- Built-in features are enabled with parameters on the Messages, Responses and Chat Completions endpoints, using the model deepseek-flash
- It also published an open-source chat demo that can import conversation history exported from the DeepSeek app
- The documentation does not state prices, and DeepSeek says public-facing services must handle their own compliance duties
DeepSeek's account on X announced on Friday afternoon that "the capabilities behind the DeepSeek App are now available through the DeepSeek API". The company's documentation describes how to build a full chatbot on top of them, with the same feature set the app offers to consumers.
What developers can switch on #
According to DeepSeek's guide, the model can search the web or images and cite its sources when a developer enables a web search tool. With thinking mode turned on, the same tool lets the model search and read links across several reasoning steps. A separate tool reads links that users paste into a message, fetching the pages and answering from them. Users can upload PDF, Word, PowerPoint and image files, which are parsed into text or a layout-aware form and added to the conversation.
Other options include a safety check that reviews inputs and outputs and returns a refusal marker when it finds a risk, an automatic conversation title, and a marker that tells the model the current date and the user's location and language. Voice is handled by standalone services: speech-to-text for input and text-to-speech to read answers aloud. DeepSeek also lists standalone OCR.
The features work through three endpoints: a Messages-style endpoint compatible with Anthropic's format, a Responses endpoint and the Chat Completions endpoint. The documentation examples use the model name deepseek-flash, and responses return structured blocks for thinking, search steps, page reads, cited text and refusals, so an app can show each stage to the user.
Breaking News as Researcher Says Six Chinese AI Models Carry the Fingerprint of Banned Nvidia Chips
Researcher Will DePue says forensic analysis shows six AI models from DeepSeek, Moonshot, MiniMax and Ant Group were initialised on Nvidia's export-controlled Blackwell chips. What he claims, what is unverified, and our questions on what it means for Nvidia and Singapore. - Chinese AI models banned Nvidia chips - Nvidia Blackwell fingerprint in Chinese AI models
A demo you can copy #
DeepSeek has published an open-source chat demo. It supports the features above, plus image and document uploads, editing and regenerating replies, Markdown and diagram rendering, Chinese and English interfaces and light and dark themes. Users can import chat history exported from the DeepSeek app, though the page warns that an import replaces the existing history. It says the demo stores the API key, conversations and uploaded images only in the browser, and describes it as for demonstration only.
What the documentation does not say #
The guide does not state prices, rate limits or regional availability. One report says tool use carries fees on top of the usual token charges, but we could not confirm a rate on DeepSeek's own page. Some usual settings, such as temperature and top-p, are ignored when search, link reading, high safety or titles are enabled, and the output token limit applies to each inference step in a round, not to the round's total.
DeepSeek also puts compliance on the developer. With the safety check enabled, the service provides basic risk control, but anyone serving the public must handle their own obligations, including content safety review, large model registration and AI labelling, and follow DeepSeek's platform terms.
Why it matters #
Until now, building a search-enabled, file-reading, voice-capable assistant meant stitching together several vendors. DeepSeek is offering the whole stack from one provider, with the same product it runs for its own users. For small teams that want an assistant without assembling the parts, the appeal is speed. The costs and the compliance duties are the things to read closely before building on it.
This article is posted in News News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.