A unified protocol layer for interacting with AI models across providers and languages.

Early development. Core protocol (v0.1) is being defined.

Working with AI models today is fragmented:

• Every provider has different APIs
• Tool/function calling formats are inconsistent
• Streaming responses behave differently
• State, history, and observability are left to developers

AI Protocols aims to fix this by defining a standard interface for AI interactions.

• Model-independent
  Use any provider without changing application logic

• Language-agnostic
  Same protocol across Python, JavaScript, PHP, and more

• Streaming-first
  Unified event-based streaming across all models

• Tool calling standardization
  One format for defining and executing tools/functions

• Built-in context & observability
  Conversation tracking, metadata, and extensibility from day one

AI Protocols is structured into three layers:

Defines the standard:

• Request / Response schema
• Streaming event format
• Tool calling interface
• Context & metadata

→ spec

Reference implementations of the protocol:

• Python (first implementation)
• JavaScript
• PHP

Thin integrations for frameworks will be introduced after the core language SDKs are stable.

The initial focus is on building a solid, language-agnostic protocol and reference implementations.

Contributions, ideas, and discussions are welcome.

This project is in its early stages—feedback now will shape the direction significantly.