AIApiTracer

A reverse proxy for local development environments that intercepts requests to AI services like OpenAI, Anthropic, Azure OpenAI, and xAI, allowing you to trace request and response content.

Simply put, it's like Cloudflare AI Gateway Logs specialized for local development environments. Note that it's not intended for auditing, monitoring, or team usage.

Features

• Monitor locally without using external services
• Human-readable display of requests and responses
• Support for multiple AI services
  • OpenAI
  • Anthropic
  • Microsoft Azure OpenAI
  • xAI
  • OpenAI Compatible APIs
• In-memory: No data persistence, retains up to 1000 records

Quick Start

1. Run AIApiTracer

Using Docker

Run AIApiTracer using the Docker image (ghcr.io/cysharp/aiapitracer).

docker run -p 8080:8080 ghcr.io/cysharp/aiapitracer:latest

Using pre-build binary

You can download and run the pre-built application from the Releases page.

./AIApiTracer --urls http://localhost:8080/

2. Open in your browser

Once started, you can view request traces by accessing http://localhost:8080 in your web browser.

3. Change Various Endpoints to AIApiTracer

OpenAI

Specify http://localhost:8080/endpoint/openai/v1 as the endpoint instead of https://api.openai.com/v1.

curl http://localhost:8080/endpoint/openai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
  "messages": [
    {
      "role": "user",
      "content": "Hello."
    }
  ],
  "model": "gpt-4.1-nano"
}'

var openAIClient = new OpenAIClient(new ApiKeyCredential(apiKey), new OpenAIClientOptions
{
    Endpoint = new Uri($"http://localhost:8080/endpoint/openai/v1"),
});

Anthropic

Specify http://localhost:8080/endpoint/anthropic as the endpoint instead of https://api.anthropic.com/.

Claude Code

export ANTHROPIC_BASE_URL=http://localhost:8080/endpoint/anthropic

Azure OpenAI

Specify http://localhost:8080/endpoint/aoai/<resource-name> as the endpoint instead of https://<resource-name>.openai.azure.com/.

var azureClient = new AzureOpenAIClient(
    new Uri("http://localhost:8080/endpoint/aoai/my-resource"),
    new AzureKeyCredential(credential)
);

xAI

Specify http://localhost:8080/endpoint/x/v1 as the endpoint instead of https://api.x.ai/v1.

curl http://localhost:8080/endpoint/x/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
  "messages": [
    {
      "role": "user",
      "content": "Hello."
    }
  ],
  "model": "grok-3-latest"
}'

var openAIClient = new OpenAIClient(new ApiKeyCredential(apiKey), new OpenAIClientOptions
{
    Endpoint = new Uri($"http://localhost:8080/endpoint/x/v1"),
});

OpenAI Compatible

Specify http://localhost:8080/endpoint/openai-compat/<openai-compatible-endpoint> as the endpoint.

⚠️ Security Notice: The OpenAI compatible endpoint is disabled by default for security reasons. You must explicitly enable it by setting the environment variable AIApiTracer__EnableOpenAICompatForwarding to true. When enabled, you should absolutely avoid making it accessible from external networks as it would become an open proxy.

Specify the address of the OpenAI compatible API endpoint after openai-compat/, including the scheme. For example, use http://localhost:8080/endpoint/openai-compat/http://localhost:5273/v1.

curl http://localhost:8080/endpoint/openai-compat/http://localhost:5273/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
  "messages": [
    {
      "role": "user",
      "content": "Hello."
    }
  ],
  "model": "Phi-4-mini-instruct-cuda-gpu"
}'

var openAIClient = new OpenAIClient(new ApiKeyCredential(apiKey), new OpenAIClientOptions
{
    Endpoint = new Uri($"http://localhost:8080/endpoint/openai-compat/http://localhost:5273/v1"), // `http://localhost:5273/v1` is Foundry Local
});

TODO

• [ ] Support for more AI services (Google Vertex AI, Amazon Bedrock)

License

MIT License