DeepSeek now sells its whole app as an API: web search, document reading, voice, and an open-source demo

Until now, an AI model's API was usually just a chat endpoint: send a question, get an answer back. According to a post by the Hỏi Dân IT Official page about an update dated 13 November, DeepSeek has expanded its API to integrate many capabilities familiar from its own app — from web search, reading link content, and document processing to voice functions — along with a demo interface with open-source code. The post also flags the trade-offs to know: control over generation parameters and tool-call costs.

Quick answer: According to the shared information about the 13 November update, DeepSeek has expanded its API from a text chat endpoint into an integrated toolset: web search, URL reading, processing of PDF/Word/PowerPoint documents and images, and voice functions (listening and speaking). DeepSeek has also open-sourced a demo interface that supports using your own API key and importing conversation history exported from DeepSeek Chat. Trade-offs to know: when some tools are enabled, DeepSeek may decide generation parameters itself, and a single request can incur both token fees and tool-call fees.

A tool-integrated API is a programming interface where the model provider does not only return text but also operates the supporting tools — search, web reading, file processing, voice conversion — so developers can call them together in one request instead of wiring up each external service themselves.

How did DeepSeek change its API in the 13 November update?

Previously, if you wanted AI to search the Internet, read documents, process files, or talk by voice, you had to connect many external services yourself. According to the post, DeepSeek is changing that: capabilities from the DeepSeek app are now integrated on the API service side.

Put simply, as the post puts it: instead of only renting the AI's brain, developers can now also use the tools that help that brain work — and the change can save a great deal of product development effort.

What do built-in web search and URL reading do for developers?

Imagine a customer-service chatbot. A customer asks: "What reviews does this product currently have on the Internet?"

Before, answering that could mean connecting a search service, fetching results, visiting each page, extracting content, and only then handing it to the AI to summarise — not to mention handling broken pages, junk content, or irrelevant results.

With the new approach, search and URL-reading tools are integrated on the service side. Developers can ask the AI to use them and receive results with source information to show users.

What is notable about PDF, Word, PowerPoint, and image processing?

According to the post, this is probably the part most worth attention for anyone building AI applications for work. An assistant built on the new API can:

• Read PDF contracts and summarise the key terms.
• Analyse Word documents dozens of pages long.
• Extract content from images.
• Read PowerPoint slides and help synthesise information.
• Search for answers based on provided documents.

Instead of stitching together many document-processing tools, developers can use the built-in capabilities.

What applications do listening and speaking AI open up?

Another interesting point is the voice functions: users can speak directly to the assistant instead of typing, and the AI's answers can in turn be converted to audio.

The post gives examples: an English-learning app, an assistant for older people, or a tool letting staff look up documents by voice. Products that previously needed several services combined now have another option to build with.

Why is the open-source demo interface rated so highly?

According to the post, this is the part many developers care about most. Instead of only providing API documentation and leaving everyone to figure it out, DeepSeek also offers a frontend demo showing how to build an application.

According to the shared introduction, the demo supports using your own API key and importing conversation history exported from DeepSeek Chat. For people learning AI programming, it is a resource worth studying: not just reading theory, but examining how a real application connects to and uses AI features.

What trade-offs should you know when using a tool-integrated API?

The post flags two technical points for AI builders to note:

• Control over generation parameters: according to the shared information, when some tools such as web search, URL reading, or safety checks are enabled, DeepSeek may decide generation parameters itself rather than letting developers fully customise them. This affects applications that need tight control over how the AI responds, such as chatbots with a fixed communication style or systems requiring consistent results.
• Cost: one request can incur both token fees and tool-call fees, rather than token counts alone — so usage costs need careful calculation.

Should you use the full API package or build your own to avoid dependence?

The post raises the core question: are we paying for things we could previously have built ourselves?

Approach	When it makes sense	Risk
Use the integrated services	Small projects needing a prototype in days; no wish to maintain many systems	At hundreds of thousands of users, every tool call becomes a significant cost; if the provider changes prices, limits, or policies, the product is affected too
Build everything yourself	Large products where operating cost, provider replaceability, and data control matter	Months can be spent building functions others already do very well

According to the author, the problem is not simply using a ready API versus building from scratch. The hard part is knowing which parts to outsource for speed and which parts are worth investing in to keep long-term control. For a small project, deployment speed may be priority number one; for a product with a large user base, operating cost, provider replaceability, and data control matter far more — and that is the point that makes DeepSeek's update worth watching, rather than just its feature list.

Frequently asked questions

1. What is new in DeepSeek's 13 November API update? According to the shared information, the API was expanded from a chat endpoint into an integrated toolset: web search, URL reading, document processing, and voice, plus an open-source demo interface.

2. Which document types can the new API process? According to the post: PDF, Word, PowerPoint, and images — reading, summarising, extracting content, and answering questions based on provided documents.

3. What does the open-source demo interface support? According to the shared introduction, the demo supports using your own API key and importing conversation history exported from DeepSeek Chat, helping developers study how a real application is built.

4. What are the trade-offs of the integrated tools? When some tools are enabled, DeepSeek may decide generation parameters itself; and costs can include both token fees and tool-call fees.

5. When should you build your own instead of using the full API? When the product is already large and operating cost, provider replaceability, and data control matter more than initial deployment speed.

This article was rewritten from a post by the Hỏi Dân IT Official page on 14 November 2026. The information about the 13 November API update and the views expressed follow the original post's source and perspective.