OpenAI

Access GPT models via the OpenAI Chat Completions API with native function calling, vision, and reasoning model support.

Prerequisites

• An OpenAI API key from platform.openai.com

Quick Setup

1. Get API Key

1. Sign up at platform.openai.com
2. Navigate to API Keys in the left sidebar
3. Click Create new secret key and copy it (starts with sk-)

2. Configure

[default]
provider = "openai"
model = "gpt-4o"

[providers.openai]
api_key = "${OPENAI_API_KEY}"

Or set the environment variable:

export OPENAI_API_KEY="sk-..."

3. Verify

prx doctor models

Available Models

Model	Context	Vision	Tool Use	Notes
gpt-4o	128K	Yes	Yes	Best general-purpose model
gpt-4o-mini	128K	Yes	Yes	Smaller, faster, cheaper
gpt-4-turbo	128K	Yes	Yes	Previous generation flagship
o3	128K	Yes	Yes	Reasoning model
o4-mini	128K	Yes	Yes	Smaller reasoning model
gpt-4	8K	No	Yes	Original GPT-4

Configuration Reference

Field	Type	Default	Description
api_key	string	required	OpenAI API key (sk-...)
api_url	string	https://api.openai.com/v1	Custom API base URL
model	string	gpt-4o	Default model to use

Features

Native Function Calling

PRX sends tools in OpenAI's native function format. Tool definitions include name, description, and parameters (JSON Schema). The provider supports tool_choice: "auto" for automatic tool selection.

Vision

Vision-capable models (GPT-4o, GPT-4o-mini) can analyze images included in the conversation. Images are sent inline through the standard message format.

Reasoning Model Support

For reasoning models (o1, o3, o4-mini), PRX automatically handles the reasoning_content fallback. When the model returns output in reasoning_content instead of content, PRX extracts the reasoning text transparently.

Multi-turn Conversations

Full conversation history is preserved and sent to the API, including system prompts, user messages, assistant responses, and tool call/result pairs in OpenAI's native structured format.

Custom Base URL

To use a proxy, Azure OpenAI, or any OpenAI-compatible endpoint:

[providers.openai]
api_key = "${OPENAI_API_KEY}"
api_url = "https://my-proxy.example.com/v1"

Connection Warmup

On startup, PRX sends a lightweight GET /models request to establish TLS and HTTP/2 connection pooling, reducing latency on the first real request.

Troubleshooting

"OpenAI API key not set"

Set the OPENAI_API_KEY environment variable or add api_key to [providers.openai] in your config.toml.

429 Rate Limit

OpenAI enforces per-minute token and request limits. Solutions:

• Wait and retry (PRX handles this automatically with the reliable provider wrapper)
• Upgrade your OpenAI plan for higher rate limits
• Use fallback_providers to fall back to another provider during rate limiting

Empty response from reasoning models

If using o1/o3/o4-mini and getting empty responses, this is expected behavior when the model's output is entirely in reasoning_content. PRX automatically falls back to reasoning_content when content is empty.