#Advanced API Usage

Once the basics work, this page covers what production use actually needs: streaming, common parameters, vision input, SDKs, and retry strategy. Replace MODEL_ID in the examples with a real ID copied from the model plaza; minimal request examples live in API Endpoints.

#Streaming

Conversational apps almost always want streaming (typewriter effect). Add "stream": true to the request body and the response arrives chunk by chunk over SSE (Server-Sent Events):

curl https://tokens.byteseek.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "stream": true,
    "messages": [{ "role": "user", "content": "Write a short poem about the sea" }]
  }'

Notes:

• Each chunk starts with data: ; the stream ends with data: [DONE];
• SDKs handle chunking for you (see below); if parsing yourself, split by line and skip blanks;
• With reasoning models, time-to-first-token stays long even with streaming — the thinking phase produces no visible output, so set generous client timeouts;
• The Anthropic format supports "stream": true too, with a different event structure — use the Anthropic SDK.

#Common parameters

Parameters follow the upstream vendor's docs for each protocol; the most used ones:

#Vision input (images)

Vision-capable models accept images in messages. OpenAI Chat format:

curl https://tokens.byteseek.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "messages": [{
      "role": "user",
      "content": [
        { "type": "text", "text": "Describe this image" },
        { "type": "image_url", "image_url": { "url": "https://example.com/photo.jpg" } }
      ]
    }]
  }'

• Images can be public URLs or inline data:image/jpeg;base64,... data;
• Images count as input tokens — large images cost noticeably more, compress first;
• Whether a model supports vision follows the capability labels in the model plaza.

#SDKs

Official SDKs only need base_url and api_key changed:

from openai import OpenAI

client = OpenAI(
    base_url="https://tokens.byteseek.ai/v1",
    api_key="sk-your-key",
)

stream = client.chat.completions.create(
    model="MODEL_ID",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://tokens.byteseek.ai/v1',
  apiKey: 'sk-your-key',
});

const stream = await client.chat.completions.create({
  model: 'MODEL_ID',
  messages: [{ role: 'user', content: 'Hello' }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '');
}

from anthropic import Anthropic

client = Anthropic(
    base_url="https://tokens.byteseek.ai",
    api_key="sk-your-key",
)

message = client.messages.create(
    model="MODEL_ID",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
print(message.content[0].text)

Keep keys in environment variables (OPENAI_API_KEY / ANTHROPIC_API_KEY) — SDKs read them automatically. Never hard-code keys into committed code; see the security requirements in the Acceptable Use Policy.

#Timeouts and retries

For production robustness:

• Generous timeouts: reasoning models respond on the scale of minutes; set client timeouts ≥ 300 s;
• Exponential backoff: retry 429 and 5xx with 1s → 2s → 4s intervals, 3–5 attempts max; other 4xx errors are request problems where retrying is pointless;
• Don't blast concurrency: dense retries without backoff get automatically deprioritised, see Rate Limits & Concurrency;
• Log request IDs: the request ID in response headers is the one handle for debugging and support tickets.

#Related

• API Endpoints — addresses and minimal examples per protocol
• Choosing a Model — pick by task
• Rate Limits & Concurrency — limit layers and raising them
• Error Codes — handling by status code