azure-openai

Provides a Node.js wrapper that abstracts Azure OpenAI service authentication and endpoint configuration, handling credential injection through environment variables or explicit parameters. The library manages the underlying HTTP client setup for communicating with Azure's OpenAI endpoints, eliminating boilerplate for developers who would otherwise need to manually construct Azure SDK clients or raw HTTP requests.

Executes chat completion requests against Azure OpenAI deployments, supporting both blocking (await response) and streaming (event-based token delivery) modes. The library marshals message arrays into Azure's expected format, handles response parsing, and optionally streams tokens back to the caller via Node.js readable streams or callback patterns, enabling real-time UI updates or token-by-token processing.

Executes text completion requests (non-chat) against Azure OpenAI deployments, supporting legacy GPT-3 models and fine-tuned completions. The library formats prompt strings into Azure's completion API format, handles response parsing, and returns completion choices with finish reasons, enabling use cases that don't fit the chat paradigm (code generation from raw prompts, text continuation, few-shot learning).

Generates vector embeddings for text inputs using Azure OpenAI's embedding models (text-embedding-ada-002 or similar). The library batches text inputs, calls the Azure embedding endpoint, and returns normalized vectors suitable for vector database storage or similarity computations. Embeddings enable semantic search, clustering, and recommendation workflows without requiring separate embedding infrastructure.

Allows developers to specify which Azure OpenAI deployment and model version to use for requests, abstracting the mapping between deployment names and underlying models. The library routes requests to the correct Azure endpoint based on deployment configuration, enabling multi-model setups (e.g., different deployments for chat vs embeddings) and A/B testing across model versions without code changes.

Catches Azure OpenAI API errors (rate limits, authentication failures, model unavailability) and maps them to meaningful exception types or error objects, preserving Azure error codes and messages. The library distinguishes between transient errors (429, 500) and permanent failures (401, 404), enabling developers to implement appropriate retry logic or user-facing error messages without parsing raw HTTP status codes.

Validates input parameters (temperature, max_tokens, top_p, etc.) against Azure OpenAI API constraints before sending requests, rejecting invalid values early with descriptive error messages. The library enforces parameter bounds (e.g., temperature 0-2, max_tokens within model limits) and type checking, preventing malformed requests from reaching Azure and reducing API call failures.