OpenAI Cookbook

Manages all published content through a centralized registry.yaml manifest file that declares content metadata including title, path, tags, authors, and publication dates. The system uses JSON Schema validation (.github/registry_schema.json) to enforce consistent metadata structure across all entries, enabling automated content discovery, filtering, and publication workflows without manual curation overhead.

Provides ~200 runnable Jupyter notebooks demonstrating concrete patterns for OpenAI API usage including chat completions, embeddings, function calling, fine-tuning, and multimodal inputs. Each notebook is self-contained with imports, API calls, and expected outputs, allowing developers to execute examples locally or in cloud notebooks (Colab, etc.) to understand API behavior through hands-on experimentation rather than documentation alone.

Demonstrates how to build voice-enabled applications using OpenAI's speech and audio capabilities, including text-to-speech synthesis, speech-to-text transcription, and real-time voice interaction patterns. Examples show how to integrate voice I/O with chat completions for conversational AI and handle audio streaming for low-latency interactions.

Provides documented techniques and patterns for improving LLM reliability including chain-of-thought prompting, self-verification, structured outputs, and error handling strategies. Content covers both prompting-level improvements (better prompt design) and system-level improvements (validation, retry logic, fallback mechanisms) with concrete examples and empirical guidance.

Demonstrates how to use embeddings and models for text classification, document clustering, and semantic search tasks. Examples show how to structure classification problems (zero-shot, few-shot, fine-tuned approaches), use embeddings for unsupervised clustering, and implement semantic search with ranking and reranking. Includes patterns for transaction classification, document organization, and search result ranking.

Demonstrates how to build autonomous agents that use models to plan, reason, and execute multi-step tasks with tool use. Examples show agent architectures (ReAct, chain-of-thought with tools), how to structure agent loops (think-act-observe), and patterns for handling tool failures and complex reasoning. Includes examples of coding agents using GPT-5 reasoning modes for complex problem-solving.

Provides guidelines and infrastructure for community contributions to the cookbook, including pull request templates, contribution guidelines, and author profile management. The system enables external developers to submit examples and articles that are reviewed, registered in the manifest, and published to the website. Authors are tracked in authors.yaml with customizable profiles.

Provides documented patterns and techniques for effective prompting with chat completions models, including basic request/response patterns, system message design, few-shot examples, and advanced techniques for reliability. Content covers both GPT-4 and GPT-5 models with specific guidance on reasoning modes, prompt personalities, and structured output formatting through examples and articles.

Demonstrates how to generate embeddings using OpenAI's embedding models and integrate them with vector databases (Qdrant, Pinecone, etc.) for semantic search, classification, and clustering tasks. Examples show the full pipeline: text-to-embedding conversion, vector storage, similarity search, and retrieval-augmented generation (RAG) workflows with concrete code for multiple database backends.

Demonstrates how to use OpenAI's function calling API to enable models to invoke external tools and APIs. Examples show basic function calling patterns (defining schemas, parsing responses), advanced patterns (multi-step tool use, knowledge retrieval via tools), and integration with external services like web APIs and custom knowledge bases. Covers both synchronous tool calling and agentic workflows where the model decides which tools to use.

Demonstrates the complete fine-tuning pipeline: preparing training data in JSONL format, submitting fine-tuning jobs via the API, monitoring training progress, and evaluating fine-tuned model performance. Examples include fine-tuning for specific tasks (classification, QA, retrieval augmentation) and show how to structure training data, handle validation splits, and compare fine-tuned vs base model performance.

Demonstrates how to use GPT-4 Vision to analyze images, extract text via OCR, answer questions about image content, and combine vision with embeddings for image-based RAG. Examples show how to pass images to the API (base64 encoding, URLs), structure vision prompts, and integrate vision capabilities with other OpenAI features like embeddings and function calling for complex multimodal workflows.

Demonstrates how to generate images using DALL-E 3, including prompt engineering for image generation, handling variations and edits, and integrating generated images into applications. Examples show how to structure effective image generation prompts, manage API responses, and combine image generation with other capabilities like embeddings for image-based workflows.

Demonstrates how to use Sora for video generation, including prompt engineering for video content, managing video generation workflows, and integrating generated videos into applications. Examples show how to structure effective video generation prompts, handle asynchronous video generation jobs, and work with video outputs.

Demonstrates how to use OpenAI models through Azure OpenAI Service, including authentication via Azure credentials, endpoint configuration, and API usage patterns that differ from standard OpenAI API. Examples show how to configure the OpenAI Python client for Azure, handle Azure-specific authentication, and deploy models through Azure's managed service.