OpenAI-Compatible API

You might say

My AI tool only has settings for OpenAI, but I want to use DeepSeek. Can I just swap the Base URL and API Key?

A third-party model endpoint conforming to OpenAI request and response formats, allowing compatible clients to switch providers seamlessly.For example, a tool that only sends OpenAI-shaped requests may connect to another model service after changing the Base URL, API key, and model name. The service may implement that shape directly or through a translation layer. Compatibility usually covers only some endpoints and fields, so tool calls, images, streaming, and other capabilities still need separate checks.

Common OpenAI API request formats

Chatmessages / input

Send the question and context, then receive a text answer.

basic format

These formats come from different OpenAI API endpoints; compatibility only means a similar request shape, not that every service supports every capability.

What problem does an OpenAI-compatible API solve?

The tool sends a familiar request: The tool sends a request with familiar paths and fields, so connecting to different model services looks similar.

The model service generates the result: The model service receives the request, processes it, and generates the answer. Different services may use different models and processing rules.

Check each capability separately: For example, a text-only model may accept an image field without being able to understand the image. Tool calls and streaming also need separate tests.

You can say this to an AI Agent

Check which OpenAI-compatible endpoints the tool and model service both support, then use non-sensitive requests to test ordinary chat and tool calling separately. Record the actual URL, status code, and missing capabilities without printing the real key.

Learn next

1Base URL→2Provider→3Tool Calling→

Recommended tool

Web fine-tuning tool · Selector Pro

Select page elements for precise AI edits

Click a button, text, or card, note the change, then copy its exact location to AI.

Open Selector Pro website

The endpoint is compatible, but the model says it does not support tool calls. Which judgment is correct?

A. A configuration flag can add tool-calling capability that the model does not haveB. Protocol compatibility does not mean the underlying model supports tool callsC. A compatible endpoint must provide every capability of a GPT model