lightweight gigachat api wrapper for smolagents

Overview

gigasmol serves two primary purposes:

1. Provides direct, lightweight access to GigaChat models through GigaChat API without unnecessary abstractions
2. Creates a smolagents-compatible wrapper that lets you use GigaChat within agent systems

No complex abstractions — just clean, straightforward access to GigaChat's capabilities through smolagents.

GigaChat API + smolagents = gigasmol 💀

Why gigasmol 💀?

• Tiny Footprint: Less than 1K lines of code total
• Simple Structure: Just 4 core files
• Zero Bloat: Only essential dependencies
• Easy to Understand: Read and comprehend the entire codebase in minutes
• Maintainable: Small, focused codebase means fewer bugs and easier updates

Installation

API-Only Installation (default)

python>=3.8

pip install gigasmol

Full Installation with Agent Support

python>=3.10

pip install "gigasmol[agent]"

Quick Start

Raw GigaChat API

gigasmol

import json
from gigasmol import GigaChat

# Direct access to GigaChat API
gigachat = GigaChat(
    auth_data="YOUR_AUTH_TOKEN",
    model_name="GigaChat-Max",
)

# Generate a response
response = gigachat.chat([
    {"role": "user", "content": "What is the capital of Russia?"}
])
print(response['answer']) # or print(response['response']['choices'][0]['message']['content'])

Usage with smolagents

gigasmol[agent]

from gigasmol import GigaChatSmolModel
from smolagents import CodeAgent, ToolCallingAgent, DuckDuckGoSearchTool

# Initialize the GigaChat model with your credentials
model = GigaChatSmolModel(
    auth_data="YOUR_AUTH_TOKEN",
    model_name="GigaChat-Max"
)

# Create an CodeAgent with the model
code_agent = CodeAgent(
    tools=[DuckDuckGoSearchTool()],
    model=model
)

# Run the code_agent
code_agent.run("What are the main tourist attractions in Moscow?")

# Create an ToolCallingAgent with the model
tool_calling_agent = ToolCallingAgent(
    tools=[DuckDuckGoSearchTool()],
    model=model
)

# Run the tool_calling_agent
tool_calling_agent.run("What are the main tourist attractions in Moscow?")

How It Works

GigaSmol provides two layers of functionality:

┌───────────────────────────────────────────────────┐
│                    gigasmol                       │
├───────────────────────────────────────────────────┤
│ ┌───────────────┐          ┌───────────────────┐  │
│ │    Direct     │          │   smolagents      │  │
│ │ GigaChat API  │          │  compatibility    │  │
│ │    access     │          │      layer        │  │
│ └───────────────┘          └───────────────────┘  │
└───────────────────────────────────────────────────┘
    │                             │
    ▼                             ▼
┌─────────────┐           ┌────────────────┐
│ GigaChat API│           │ Agent systems  │
└─────────────┘           └────────────────┘

1. Direct API Access: Use GigaChat for clean, direct access to the API
2. smolagents Integration: Use GigaChatSmolModel to plug GigaChat into smolagents

Examples

Check the examples directory:

• structured_output.ipynb: Using GigaChat API and function_calling for structured output
• agents.ipynb: Building code and tool agents with GigaChat and smolagents

Acknowledgements

• SberDevices for creating the GigaChat API
• Hugging Face for the smolagents framework