provider-streaming

What problem does it solve?

Standardizes streaming patterns for LLM providers to ensure reliable, tool-enabled interactions across diverse backends and tool schemas, reducing integration friction.

Core Features & Use Cases

• Provides a canonical streaming workflow that guarantees tools are passed to both complete() and stream() calls.
• Documents provider-specific quirks and rules (e.g., comparator behavior for Anthropic, Gemini, Ollama, OpenAI, LiteLLM) to avoid common streaming pitfalls.
• Use Case: When integrating a new provider or debugging a streaming session, follow the provider-adapter guidance and validated tests to ensure correct event sequencing.

Quick Start

Implement a new provider's streaming by wiring the provider layer, ensure the stream() method accepts tools, and run the streaming tests to verify tool event ordering.