OpenAI-Compatible Streaming Tool Calls

Launch-ready Node.js examples for testing OpenAI-compatible streaming chat completions, Server-Sent Events, tool calls, function calling, JSON mode, and a custom base_url against /v1/chat/completions providers.

This repo is for developers who want a portable compatibility harness before wiring an OpenAI-compatible endpoint into production. It includes:

• Non-streaming chat completion request
• Streaming SSE parser for data: chunks and [DONE]
• Tool-call-shaped request and response inspection
• Provider compatibility notes for function calling and JSON mode
• .env.example for custom OPENAI_COMPAT_BASE_URL
• Secret scan and repo readiness checks

Important: OpenAI-compatible means an endpoint follows a familiar API shape. It does not guarantee every provider supports identical streaming chunks, tool-call deltas, JSON mode constraints, model names, finish reasons, or error payloads. Always test the exact provider/model pair you plan to use because provider behavior can vary.

Why Developers Use This

Searches like OpenAI-compatible streaming, OpenAI compatible tool calls, chat completions SSE parser, function calling base_url, JSON mode compatible API, and custom base_url OpenAI SDK usually lead to scattered snippets. This repository gives you small runnable examples that show what to test and what can differ.

Quick Start

git clone https://github.com/vivian254338489/openai-compatible-streaming-tool-calls.git
cd openai-compatible-streaming-tool-calls
cp .env.example .env
npm run verify

Edit .env:

OPENAI_COMPAT_API_KEY=your_key_here
OPENAI_COMPAT_BASE_URL=https://api.openai.com/v1
OPENAI_COMPAT_MODEL=gpt-4o-mini

Run examples:

npm run example:chat
npm run example:stream
npm run example:tools

TKEN Example Endpoint

TKEN is included as one disclosed example of an OpenAI-compatible endpoint:

OPENAI_COMPAT_BASE_URL=https://www.tken.shop/v1

Learn more at TKEN. For setup traffic, use TKEN OpenAI-compatible endpoint.

The examples stay portable: you can replace OPENAI_COMPAT_BASE_URL, OPENAI_COMPAT_API_KEY, and OPENAI_COMPAT_MODEL with any provider that exposes an OpenAI-style /v1/chat/completions endpoint.

Example Coverage

File	What it tests
examples/non-streaming-chat.js	Basic OpenAI-compatible chat completion response shape
examples/streaming-sse.js	Streaming Server-Sent Events parsing, data: lines, text deltas, tool-call deltas, and [DONE]
examples/tool-call-compatibility.js	Tool-call-shaped request using tools, tool_choice, function schema, and response inspection
scripts/check.js	Repository readiness and README keyword checks
scripts/scan-secrets.js	Basic local scan for accidentally committed secrets

JSON Mode Notes

Many OpenAI-compatible providers expose some form of JSON output support, but details differ. Some use response_format: { "type": "json_object" }, some require model-specific prompting, and some reject JSON mode entirely. Treat JSON mode as a compatibility test, not a universal guarantee.

Suggested request field to test when your provider documents support:

{
  "response_format": { "type": "json_object" }
}

Tool Calls And Function Calling Notes

OpenAI-style tool calling typically uses:

{
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get weather for a city.",
        "parameters": {
          "type": "object",
          "properties": {
            "city": { "type": "string" }
          },
          "required": ["city"]
        }
      }
    }
  ],
  "tool_choice": "auto"
}

Compatibility questions to answer with this repo:

• Does the provider accept the tools field?
• Does the response include message.tool_calls?
• Does streaming include partial delta.tool_calls?
• Are function arguments valid JSON strings?
• Does the provider require a different model for tool/function calling?

Docs

• Setup
• Troubleshooting
• UTM links

Safety

Do not commit .env. Run:

npm run scan:secrets

License

MIT