DeepSeek opens its app features to developers through its API
DeepSeek's API now handles web search, file reading and more on its own servers, so developers can build a close copy of its chat app without wiring those tools up themselves.
The short version
- Web search, file input, link reading and safety checks now run on DeepSeek's servers when a developer switches on the matching setting.
- V4.1-Flash supports these features across three API interfaces: Messages, Responses and Chat Completions.
- Search, web fetch, text recognition (OCR), speech recognition and speech synthesis are also offered as separate pieces.
- DeepSeek released an open-source front-end demo that shows the full chat experience.
- The report does not give price figures in its text, and it is about a hosted service, not downloadable model files.
Why it matters
Building a chat product has meant gluing a model to search, file parsing and safety tools by hand. If this works as described, a small team can prototype a full chat app much faster. The trade-off is that your product runs on DeepSeek's servers, under DeepSeek's rules and prices.
What DeepSeek opened up
As Phoenix Tech (ifeng Tech) reported, DeepSeek has put the features of its chat app and website into its API. An API is the doorway that lets other programs talk to a model. Until now, developers mostly got the model's raw answers and had to add the extras themselves.
Those extras include searching the web with citations, reading uploaded PDFs, Office files and images, opening links a user pastes in, and checking inputs and outputs for safety, including images. DeepSeek reportedly says developers can rebuild its full app from them.
The work moves to DeepSeek's servers
The old route meant hosting a model, then connecting outside tools and stitching their results back into the conversation. The report compares that to patching together a Frankenstein.
Now a developer switches on a setting in the existing API, and DeepSeek's servers run the tool calls and assemble the conversation context. For web search and link reading, there are named tools, deepseek_web_search and deepseek_read_url. In thinking mode, the model can search and read links across several reasoning steps. The report says V4.1-Flash supports all this across three interface types.
Building blocks and a demo
DeepSeek also split the underlying abilities into separate interfaces that can be used alone or combined:
- Search, for web lookups
- Fetch, for grabbing web pages
- OCR, which turns text in images into text
- ASR, which turns speech into text
- TTS, which turns text into spoken audio
The company also released a lightweight front-end demo as open-source code. It shows search, deep thinking, file uploads, read-aloud answers, voice input and chart rendering. Per the report, API keys, chat history and images stay in the user's browser.
What this does not give you
The headline says you can build your own DeepSeek, but an API gives you a hosted model, not the model's files. Your data goes to DeepSeek's servers, and DeepSeek sets the limits and rates. What you build is a product wrapped around its model.
The report points to a surprising price list, but the text we have does not state the figures. Read the pricing page and data terms yourself before committing.
What to watch
- What the pricing for these bundled features actually looks like once developers start using them.
- How DeepSeek's terms treat data sent through search, file and link tools.
Try it
Read DeepSeek's chat-demo code repository (github.com/deepseek-ai/chat-demo) to see how a full chat app is put together. Check the pricing and data terms before sending it anything real.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by Phoenix Tech (ifeng Tech) · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
