← सभी गाइड पर लौटें

OpenAI Chat Completions: Zero to Production

What the chat completions format does

The OpenAI-compatible chat completions format gives your application one predictable way to send conversation context to a language model and read its answer. You submit a JSON request to a chat completions endpoint, identify the model, provide an ordered messages array, and receive the generated content inside a structured response.

This compatibility is useful when you want to switch between GPT models and supported Claude families without rewriting your application. 59API provides a low-cost, pay-as-you-go relay at https://api.59api.com, with access to native official-quality models rather than downgraded substitutes. Exact model IDs and availability can change, so check the 59API dashboard before deploying.

Understand the request schema

A basic request contains four important parts: model, messages, generation settings, and optional streaming or tool parameters. Each message has a role and content. The system role sets behavior, the user role contains the current request, and assistant messages preserve earlier model replies.

For example, a minimal JSON body is: {"model":"YOUR_MODEL_ID","messages":[{"role":"system","content":"Answer clearly and briefly."},{"role":"user","content":"Explain DNS in one paragraph."}],"temperature":0.2,"stream":false}. Use a model ID listed by your provider instead of assuming that a friendly name such as Claude Sonnet or GPT is accepted.

Make your first HTTP request

Create a 59API key and store it in an environment variable named OPENAI_API_KEY. Never place a live key in browser JavaScript, a public repository, or a mobile application. Then call the conventional OpenAI-compatible route:

Method and URL: POST https://api.59api.com/v1/chat/completions

Headers: Authorization: Bearer YOUR_API_KEY; Content-Type: application/json

Body: include the model and messages JSON shown above. With curl, send the body using a file or the command-line JSON option, and inspect the HTTP status before parsing the response.

A successful response normally contains choices[0].message.content, while token accounting appears in usage. Do not parse the entire response as if it were plain text. Check for an error object and handle an empty choices array defensively.

Use an OpenAI SDK

Most OpenAI-compatible SDKs allow you to change only the API key and base URL. In Python, install the OpenAI package, then initialize it with the 59API base URL ending in /v1: client = OpenAI(api_key=os.environ["OPENAI_API_KEY"], base_url="https://api.59api.com/v1"). Call client.chat.completions.create with model and messages, then read response.choices[0].message.content. In JavaScript, the official OpenAI client follows the same pattern with new OpenAI({ apiKey, baseURL }).

Manage conversation state correctly

The API does not automatically remember previous requests. Your application must send relevant history on every call, usually as an array such as system, user, assistant, user. Keep the system instruction stable, append each completed turn, and remove or summarize old messages when the context becomes too large. A practical production strategy is to retain recent turns and create a compact summary for older information.

Model behavior is not identical across providers. Temperature, token limits, tool calling, vision input, and other optional fields may vary by model. Start with the common fields, test each selected model, and consult the provider documentation before relying on advanced parameters.

Prepare the integration for production

• Set connection and read timeouts. Retry temporary 429 and 5xx responses with exponential backoff, but do not blindly retry every error.
• Validate user input, cap message size, and enforce an application-level spending limit. 59API's pay-as-you-go pricing can be economical, but uncontrolled loops can still create unexpected usage.
• Log request IDs, latency, selected model, status, and usage without logging API keys or sensitive message content.
• Use stream: true when your interface should display tokens progressively. Streaming responses arrive as events or chunks, so your parser must assemble content and handle the final usage or termination signal.
• Pin tested model IDs and maintain a fallback policy. A fallback should preserve the same response contract and be tested for quality, latency, and cost.

Why use 59API?

For teams comparing model providers, 59API combines OpenAI-compatible integration with cheap, pay-as-you-go access to GPT models and Claude Opus, Sonnet, Haiku, and Fable offerings. That lets an existing SDK-based application experiment with different model families while keeping one familiar request format. If the pricing and supported model list fit your workload, sign up for 59API, create a key, and begin with a small test budget before moving the integration into production.

मूल्य · दस्तावेज़