An unofficial Go client that tracks every OpenAI model drop

What it does Provides unofficial Go clients for the OpenAI API, wrapping endpoints for chat completions (GPT-3, GPT-4, 4o, o1), image generation (DALL·E 2, DALL·E 3, GPT Image 1), speech-to-text (Whisper), and embeddings into typed structs. It handles the HTTP boilerplate so you don’t have to hand-roll JSON.

The interesting bit It also speaks Azure OpenAI, letting you swap endpoints and deployment names via a configurable model mapper. That dual-vendor support is unusual for an unofficial client, suggesting the maintainer cares about production portability rather than just weekend demos.

Key highlights

• Covers chat, images, audio, and embeddings in one dependency
• Supports streaming completions for real-time token delivery
• Azure OpenAI compatible, with custom model-to-deployment mapping
• Allows custom HTTP clients and proxy configuration
• Requires Go 1.18 or newer

Caveats

• Explicitly unofficial: breaking API changes depend on community turnaround
• Documentation is example-heavy and omits guidance on retries, rate limits, or production error handling

Verdict A practical pick for Go services that need typed access to OpenAI’s generative APIs without maintaining their own HTTP client. Skip it if you require an officially supported SDK or built-in resilience patterns.