rusted-chain 0.1.0

LangChain like python library written in rust

Rusted Chain

Rusted Chain is a Rust-powered LLM agent framework for Python. It combines the performance of Rust with the usability of Python, providing a LangChain-compatible API for building AI agents.

Features

• Rust Core: Backend implemented in Rust using pyo3.
• Multi-Model Support: Supports Gemini, OpenAI, and Claude.
• LangChain Compatible: Accepts standard LangChain tools and @tool decorators.
• Auto-Execution Loop: Handles tool calling and execution automatically.
• Simple API: Straightforward interface for creating agents and tools.

Installation

pip install rusted-chain

Quick Start

Basic Usage

from rusted_chain import GeminiModel, OpenAIModel, ClaudeModel

# Initialize an agent
agent = GeminiModel(api_key="your-api-key")

# Simple text generation
response = agent.invoke("Tell me a joke about Rust.")
print(response.text)

Using Tools

You can pass Python functions directly. rusted-chain handles schema generation and execution.

from rusted_chain import GeminiModel

def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"The weather in {city} is sunny and 25°C."

# Create agent with tools
agent = GeminiModel(tools=[get_weather])

# Run the agent
result = agent.run("What's the weather like in Tokyo?")
print(result)
# Output: "The weather in Tokyo is sunny and 25°C."

Using LangChain Tools

You can also use existing LangChain tools or the @tool decorator.

from langchain_core.tools import tool
from rusted_chain import OpenAIModel

@tool
def multiply(a: int, b: int) -> int:
    """Multiplies two numbers."""
    return a * b

agent = OpenAIModel(tools=[multiply])

result = agent.run("What is 123 * 456?")
print(result)

Supported Models

Class	Provider	Env Variable
GeminiModel	Google Gemini	GOOGLE_API_KEY
OpenAIModel	OpenAI (GPT-4, etc.)	OPENAI_API_KEY
ClaudeModel	Anthropic Claude	ANTHROPIC_API_KEY

Advanced Usage

Manual Tool Control

Use invoke() to get the raw response (text or tool call) without auto-execution.

response = agent.invoke("Call my_tool")

if response.is_tool_call:
    print(f"Tool: {response.tool_call.name}")
    print(f"Args: {response.tool_call.args}")
else:
    print(response.text)

Performance benchmark (test_perf.py)

A small benchmarking script is included at test_perf.py to compare the request/response latency of rusted_chain vs a LangChain-based client when calling the Google Gemini model (the repository author used gemini-2.5-flash for tests).

pip install langchain-google-genai python-dotenv

How to run

• Optional environment variables used by the script:
  • RC_BENCH_WARMUP — set to true/false (default: true) to perform warm-up calls before measurements.
  • RC_BENCH_REPEATS — number of timed repetitions (default: 4).

Example usage (from repo root)

# set API key and run the benchmark
export GOOGLE_API_KEY="your-google-api-key"
python test_perf.py

License

MIT