Eigen RadarAI
Analysis

DeepSeek opens its app’s search, file and voice tools to developers

DeepSeek has made its consumer app capabilities available through its developer interfaces. Search, file input and link reading can be enabled in existing conversations, with the service handling tool calls and context assembly. Voice and document-recognition functions are also offered separately, and a demonstration client can import exported DeepSeek conversation histories.

Artificial Intelligence··Morning
Three developers at a shared desk on a bright open-plan floor of a Chinese AI company; one speaks into a desk microphone, another feeds a folder into a scanner.

Existing conversation interfaces gain app tools

Chinese artificial-intelligence company DeepSeek has opened the tools used by its consumer chat application to outside developers. Its November 13 update covers web search, uploaded files, reading links, voice and safety checks. Developers turn on the relevant parameters in existing requests; the service then executes the tools and puts their results into the conversation context. The supported model is deepseek-flash, through the Messages, Responses and Chat Completions interfaces. This gives applications access to capabilities previously combined inside the DeepSeek app through the same conversation interfaces they already use for model responses.[1], [2]

Search behavior follows the thinking setting

With thinking turned off, the search tool can choose a web or image search and produce an answer with citations. With thinking enabled, the model can search and open links during several reasoning steps. File handling accepts formats including PDFs, Office documents and images. Their contents are extracted as text or parsed with layout information before entering the conversation. A separate link-reading tool detects addresses in the user’s message, retrieves those pages and makes their contents available alongside the rest of the context. The safety parameter checks both sides of the exchange: model inputs and generated outputs. A detected risk returns a refusal indicator.[1]

Voice and document services can run separately

Search and page retrieval can also be requested as standalone services, alongside optical character recognition, speech recognition and speech generation. Developers can therefore call these underlying functions separately from a conversation. DeepSeek has also released code for a demonstration chat client built on the services, including the ability to import an exported DeepSeek conversation history. Its earlier changelog already recorded the opening of the company’s search service on October 17. The November package extends that access to the wider suite of application functions and their conversation integration. For public-facing applications, DeepSeek describes the safety parameter as basic risk control and reminds developers of their responsibilities for content review, model registration and artificial-intelligence labeling.[1]

References

  1. News sourceDeepSeekDeepSeek opens its app search and file tools to developers↩1↩2↩3
  2. News sourceIT HomeDeepSeek releases its full application API suite, opening app features to developers↩