Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableDeepSeek Blog→ original

DeepSeek Opens Its App Features Through the API: Search, File Reading, Voice, and Safety Checks

DeepSeek has opened up the features of its app through the API: search, reading files and links, safety checks, OCR, speech recognition, and speech synthesis. They are enabled with parameters in a request to V4.1-Flash, and the server handles tool calls and context assembly. As a result, an app-level chat can be built without custom plumbing, but the tools are billed on top of token costs.

AI-processed from DeepSeek Blog; edited by Hamidun News
DeepSeek Opens Its App Features Through the API: Search, File Reading, Voice, and Safety Checks
Source: DeepSeek Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceDeepSeek Blog ↗
Editor-in-Chief
Key metric #1Search is enabled with the deepseek_web_search flag, and link reading with deepseek_read_url
Key metric #2Public applications must enable safety checks
Key metric #3There is an open demo on GitHub that imports DeepSeek chat history

If you have ever built your own chatbot, you know that the model itself is only part of the work. The rest is consumed by plumbing: web search, PDF parsing, reading links from messages, moderation, and voice. DeepSeek has decided to take this routine off developers' hands and has opened up, through its API, the same capabilities that power its own app. The same set can now be connected to your own product.

What Is Available Now

The built-in tools work with the V4.1-Flash model (identifier deepseek-flash). They are supported in all three API formats: Messages, Responses, and Chat Completions. DeepSeek's server runs the tool calls and assembles the context, so you do not need to write the loop yourself in which the model asks for a search, you execute it, and you return the result.

The set includes the following:

  • Web Search. Web or image search with source citations. It works in the non-"thinking" mode. Enabled with the deepseek_web_search parameter.
  • Deep Thinking Search. A search in which the model itself searches and reads pages during multi-step reasoning. It suits questions that cannot be answered from a single query.
  • File Input. Parsing of PDFs, office documents, images, and other files directly into the conversation context. The user uploads a contract or a spreadsheet, and the model answers based on its contents.
  • URL Reading. If a message contains a link, the service loads the page itself and uses it to answer. Enabled with the deepseek_read_url parameter.
  • Safety Checks. Screening of incoming and outgoing messages. If the system detects a risk, it returns a refusal indicator, and the application decides what to show the user.

Standalone APIs for Your Own Scenarios

Not everyone needs a ready-made chat. That is why some capabilities are offered as standalone APIs that can be plugged into any pipeline:

  • Search: search without tying it to a dialogue.
  • Fetch: loading the content of a page.
  • OCR: text recognition in images and scans.
  • ASR: speech-to-text conversion.
  • TTS: text-to-speech synthesis.

For example, a call center can use ASR to transcribe calls, a general model for summaries, and TTS for voice replies. An accounting service can use OCR for scans and leave the rest untouched.

For those who want the "like in the app" experience, DeepSeek has published an open frontend demo in the chat-demo repository on GitHub. It can import conversation history from the DeepSeek app. A step-by-step guide to building a chatbot is available in the API documentation.

Caveats That Are Easy to Overlook

The main caveat concerns money. Extended capabilities are billed separately: the tools are paid on top of the regular token cost. The original announcement includes a price table, but this summary does not list specific rates, so before launch it is worth checking the table and estimating what a search will cost per request.

The second caveat is responsibility. Public applications are required to enable safety checks. DeepSeek also states explicitly that the developer remains responsible for content moderation, for registering the model where local law requires it, and for labeling AI-generated material. The built-in checks do not remove these obligations; they only help you meet them.

What This Means in Practice

For small teams, this saves weeks. Previously, a "chat with search and file upload" meant a separate search service, a document parser, a queue, moderation, and context stitching. Now much of that reduces to two or three parameters in a request. The biggest winners are prototypes and internal tools, where speed matters more than fine-grained control.

There are also obvious trade-offs. Server-side context assembly is convenient, but in exchange you have less control over which pages and fragments end up in the prompt. If you need your own ranker, strict rules for selecting sources, or data storage strictly within your own infrastructure, the standalone Search and Fetch APIs combined with your own logic remain the more flexible option. In addition, support is stated for V4.1-Flash specifically, and the announcement says nothing about other models in the lineup.

Still, the direction is clear: model providers increasingly sell not just tokens but the ready-made infrastructure around them. DeepSeek is taking that step and adding an open demo to start from.

Story

Key Questions Answered

What Is Available Now?

The built-in tools work with the V4.1-Flash model (identifier deepseek-flash). They are supported in all three API formats: Messages, Responses, and Chat Completions. DeepSeek's server runs the tool calls and assembles the context, so you do not need to write the loop yourself in which the model asks for a search, you execute it, and you return the result. The set includes the following: Web Search. Web or image search with source citations. It works in the non-"thinking" mode.

What are the key details on Standalone APIs for Your Own Scenarios?

Not everyone needs a ready-made chat. That is why some capabilities are offered as standalone APIs that can be plugged into any pipeline: Search: search without tying it to a dialogue. Fetch: loading the content of a page. OCR: text recognition in images and scans. ASR: speech-to-text conversion. TTS: text-to-speech synthesis. For example, a call center can use ASR to transcribe calls, a general model for summaries, and TTS for voice replies.

What are the key details on Caveats That Are Easy to Overlook?

The main caveat concerns money. Extended capabilities are billed separately: the tools are paid on top of the regular token cost. The original announcement includes a price table, but this summary does not list specific rates, so before launch it is worth checking the table and estimating what a search will cost per request. The second caveat is responsibility. Public applications are required to enable safety checks.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…