DeepSeek api documentation: Quickstart guide

Base URL and Authentication

Our API works with any OpenAI-compatible client by simply changing the base URL and providing your API key. We do not route multiple models; we host one uncensored large language model tuned for adult and coder use cases. Authentication relies on a single API key per account, which you receive immediately after signing up via Google or email. There is no phone number required, and no subscription fees exist. Your prompts are not used for training data, ensuring privacy while you integrate our endpoint into your existing codebase.

Copycurl https://api.deepseekapidocumentation.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "uncensored",
    "messages": [{"role": "user", "content": "Explain the difference between supervised fine-tuning and RLHF."}]
  }'

POST /v1/chat/completions: The core endpoint

Send your conversations to our single endpoint for text-in, text-out responses. You do not need to manage model routing or worry about gateway latency because we serve one tuned endpoint directly. The model id is uncensored, and it responds to standard chat completion parameters. This approach eliminates the complexity of selecting from a pool of models. You get predictable behavior without the noise of multiple vendor versions.

Copyfrom openai import OpenAI

client = OpenAI(base_url="https://api.deepseekapidocumentation.com/v1", api_key="YOUR_KEY")

resp = client.chat.completions.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Summarise this thread without softening it."}],
)
print(resp.choices[0].message.content)

GET /v1/models: Verify your uncensored model

Confirm your connection and the active model by hitting the models endpoint. This returns the specific configuration of our uncensored LLM, distinct from GPT, Claude, or other vendors. You can verify the context window and token limits directly from this response. It is a quick check to ensure your SDK client is pointing to our servers correctly. Use this to debug connection issues before sending heavy payloads.

Copyimport OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.deepseekapidocumentation.com/v1", apiKey: process.env.API_KEY });

const resp = await client.chat.completions.create({
  model: "uncensored",
  messages: [{ role: "user", content: "Draft a villain monologue for my game." }],
});
console.log(resp.choices[0].message.content);

Parameters: Temperature, Top P, and Streaming

Control creativity with temperature and top_p, or enable streaming via Server-Sent Events (SSE). Streaming provides real-time token updates, which is crucial for building responsive chat interfaces. You can also set a seed for deterministic outputs or define stop sequences. Our API returns token usage in the final chunk of a stream. This transparency helps you track costs accurately as the model generates text.

Copystream = client.chat.completions.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Tell the story in second person."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

JSON Mode and Function Calling

Structure your outputs for programmatic use with JSON mode or integrate external tools via function calling. Set the response format to json_object to force strict JSON compliance. Function calling allows the model to invoke tools with specific arguments. This setup is ideal for building agents or parsing structured data. You can also use presence and frequency penalties to reduce repetition in generated text.

Rate Limits and Token Counts

Each key supports 300 requests per minute and allows 8 concurrent requests. The context window holds 100,000 tokens total, with a maximum output of 16,000 tokens per request. If you do not set max_tokens, the limit defaults to 2,048 tokens. Request bodies cannot exceed 8 MB. These limits are strict but predictable, helping you design robust retry logic without unexpected throttling. Errors do not consume credit, so you can test freely.