Forge (🔥)

A minimal-core, zero-magic, native async/streaming framework for LLM tools & agents.

Forge is built as a lightweight, transparent alternative to LangChain. Instead of deep nested wrappers, magic execution chains, and implicit retries, Forge provides a flat execution model: agent -> tool call -> result.

⚡ Key Features

• Flat Execution Model: Explicit, readable loops. Debugging is simple because there are no hidden layers.
• Typed Tool Schemas: Use Python function signatures + Pydantic for automated, clean JSON schema generation.
• Native Async & Streaming: Built from the ground up on async/await for fast, non-blocking execution.
• Built-in Local Tracing: Full visibility into exact request/response payloads without relying on external paid services.
• Zero Bloat: Lightweight dependencies. Optional provider packages (e.g. anthropic).

🚀 Quickstart

import asyncio
from forge import Agent, tool, AnthropicProvider

@tool(description="Calculate the result of a mathematical expression")
def calculate(expression: str) -> str:
    return str(eval(expression))

async def main():
    provider = AnthropicProvider(model="claude-3-5-sonnet-20241022")
    agent = Agent(
        provider=provider,
        tools=[calculate],
        system_prompt="You are a helpful assistant.",
        verbose=True,
    )

    history = await agent.run("What is 128 * 42?")
    print("Answer:", history[-1].content)

if __name__ == "__main__":
    asyncio.run(main())

🛠️ Testing & Development

Run unit tests via pytest:

pytest