Subscribe
0%
Illustration for: DeepSeek puts its app's search, file reading and voice into the API and open-sources a chat client. What it costs

DeepSeek puts its app's search, file reading and voice into the API and open-sources a chat client. What it costs

DeepSeek's cheapest model can now search the web, read links and parse PDFs on the server side, with standalone OCR, speech-to-text and text-to-speech alongside. An MIT-licensed browser client shows how to rebuild the DeepSeek app on top of it, but tool fees stack on top of token fees and public apps must switch on safety checks.

Listen to this article0:00 / 5:39

DeepSeek has moved the features behind its consumer chat app into its developer API. Its V4.1-Flash model (deepseek-flash) can now search the web, read links and parse uploaded files on DeepSeek's servers, and the company has published standalone search, fetch, OCR, speech-to-text and text-to-speech APIs, according to a November 13 notice on its API docs site. To show how the pieces fit, it also open-sourced chat-demo, an MIT-licensed chat client that runs entirely in the browser.

Why it matters: until now, anyone building a DeepSeek-powered assistant with web search or document upload had to wire up their own search provider, scraper and PDF parser, then feed the results back into the model. DeepSeek now runs that loop itself, at prices that are low per call but billed on top of tokens. The demo repo picked up about 2,500 GitHub stars in its first three days, according to GitHub's search API.

What the API can now do

DeepSeek says the built-in tools work across its Messages, Responses and Chat Completions request formats, with tool calls and context assembly handled server-side:

  • Web search of pages or images, with citations, in non-thinking mode.
  • Deep thinking search, which searches and reads pages during multi-step reasoning.
  • File input for PDFs, Word and PowerPoint files and images, parsed into the conversation.
  • URL reading, which automatically fetches links in a user's message.
  • Safety checks that screen inputs and outputs and flag refusals.

Developers switch on search and link reading by declaring the deepseek_web_search and deepseek_read_url tools, according to DeepSeek's chatbot guide. The guide also shows an option to auto-generate a conversation title. When tools, the high safety level or titles are enabled, several sampling parameters, including temperature, top_p and the thinking token budget, are ignored.

The safety-check requirement

DeepSeek tells developers that any app offered to the public should enable the safety check ("safety_check": {"level": "high"}). It describes this as basic risk control only: developers remain responsible for their own content review, model registration and AI-content labelling. A blocked answer comes back with stop_reason: "refusal", and DeepSeek says apps should replace it with their own refusal message.

What it costs

Tool fees are charged per use on top of normal token charges. These are the rates on DeepSeek's pricing page at the time of writing:

Item Unit Price
Search (text or image) per 1,000 searches $0.75
Web fetch per 1,000 fetches $0.75
File parsing and OCR per 1,000 pages $0.225
Text to speech per 10,000 characters $0.225 (limited-time; list $0.45)
Speech to text per hour of audio $0.058 (limited-time; list $0.116)
deepseek-flash input (cache miss) per 1M tokens $0.30 peak / $0.15 off-peak
deepseek-flash output per 1M tokens $1.20 peak / $0.60 off-peak

DeepSeek gives no end date for the 50% speech discounts. Its peak window is 01:00–04:00 and 06:00–10:00 UTC on weekdays. Search and fetch are billed the same whether called directly or through the model's tools.

For scale, a DEV Community breakdown works out that a chat with five searches and a 10-page PDF adds about $0.006 in tool fees. The same author notes that search is the cost that grows: an agent that searches on every turn could see tool fees overtake token fees over a month. They also report that the standalone Search and Fetch APIs date back to October 17. What's new on November 13 is running them inside a conversation.

What's in chat-demo

The repo's README describes a static TypeScript app with no backend. The browser talks directly to DeepSeek's API using the key you enter in Settings. It supports multiple conversations with message branches, streaming replies, search with citations, link reading, attachments, voice input and read-aloud, Markdown with math and Mermaid diagrams, and Chinese and English interfaces. By default it uses deepseek-flash through the Anthropic-compatible Messages and Files APIs. DeepSeek's notice says it can also import conversation history exported from the DeepSeek app.

Two details are worth knowing before you use it:

  • Your API key sits in the browser. The README says the key and settings are stored in localStorage, and chats and attachments in IndexedDB. That's fine on your own machine. If you host it for others, each user should bring their own key, and don't bake yours into a public build.
  • Deleting chats doesn't delete uploads. The README says files already uploaded to DeepSeek's API are not removed when you clear local history.

Running it locally needs Node.js 22.12 or later and pnpm. DeepSeek also publishes a prebuilt archive that opens from index.html.

Caveats

Several writeups so far, including CellCog's and DEV's, say their authors didn't call the APIs themselves, and TechMeasured hasn't tested them either. DEV reports that the built-in web and file tools work only on deepseek-flash, not on deepseek-v4-pro. DeepSeek's own pages don't state any per-tool rate limits or file-size caps.

Who should care

If you already use DeepSeek's API and run your own search or PDF pipeline, it's worth comparing your current per-query cost with $0.00075 per search and $0.000225 per page. If you send user documents or chats to any third-party API, check where the data goes first: these tools process content on DeepSeek's servers.

Sources

  1. DeepSeek API Docs — The full DeepSeek App experience, now via API api-docs.deepseek.com
  2. DeepSeek API Docs — Build a Chatbot api-docs.deepseek.com
  3. DeepSeek API Docs — Models & Pricing api-docs.deepseek.com
  4. GitHub — deepseek-ai/chat-demo github.com
  5. DEV Community — DeepSeek Put Its App Inside the API: Search, Files, Voice — and a Bill With Three Extra Lines dev.to
  6. CellCog — DeepSeek API Adds Web Search, File Input and Voice Tools cellcog.ai

#DeepSeek#deepseek-flash#V4.1-Flash#API#web search#OCR#speech#open source#chat-demo