QuantaLogic is a ReAct (Reasoning & Action) framework for building advanced AI agents.

It seamlessly integrates large language models (LLMs) with a robust tool system, enabling agents to understand, reason about, and execute complex tasks through natural language interaction.

The cli version include coding capabilities comparable to Aider.

📖 Documentation

HowTo Guide

We created QuantaLogic because we saw a significant gap between the advanced AI models developed by companies like OpenAI, Anthropic, DeepSeek and their practical implementation in everyday business processes.

Our mission is to bridge this gap, making the power of generative AI accessible and actionable for businesses of all sizes.

• ReAct Framework: Advanced implementation combining LLM reasoning with concrete actions
• Universal LLM Support: Integration with OpenAI, Anthropic, LM Studio, Bedrock, Ollama, DeepSeek V3, DeepSeek R1, via LiteLLM. Example usage: quantalogic --model-name deepseek/deepseek-reasoner or quantalogic --model-name openrouter/deepseek/deepseek-r1
• Secure Tool System: Docker-based code execution and file manipulation tools
• Real-time Monitoring: Web interface with SSE-based event visualization
• Memory Management: Intelligent context handling and optimization
• Enterprise Ready: Comprehensive logging, error handling, and validation system

• Usage

• Release Notes

• Installation

• Quick Start

• Key Components

• Agent System

• Tool System

• Web Interface

• Examples

• Development

• Contributing

• License

• Documentation Development

Usage: quantalogic [OPTIONS] COMMAND [ARGS]...
Environment Variables: Set OPENAI_API_KEY, ANTHROPIC_API_KEY, and DEEPSEEK_API_KEY for API integration.

Options:

• --model-name TEXT: Specify the model (litellm format, e.g., "openrouter/deepseek/deepseek-chat")
• --log [info|debug|warning]: Set logging level
• --mode [code|basic|interpreter|full|code-basic|search|search-full]: Agent mode
• --vision-model-name TEXT: Specify vision model (litellm format)
• --max-tokens-working-memory INTEGER: Maximum tokens in working memory (default: 4000)
• --max-iterations INTEGER: Maximum task iterations (default: 30)
• --compact-every-n-iteration INTEGER: Compact memory every N iterations (default: 5)
• --no-stream: Disable streaming output (default: enabled)
• --help: Show help message

Commands:

• task: Execute a task with the QuantaLogic AI Assistant
  • --file PATH: Path to task file
  • --model-name TEXT: Specify model
  • --verbose: Enable verbose output
  • --mode: Select agent capabilities
  • --log: Set logging level
  • --vision-model-name: Specify vision model
  • --max-iterations: Maximum task iterations
  • --max-tokens-working-memory: Memory limit
  • --compact-every-n-iteration: Memory optimization
  • --no-stream: Disable streaming

See our Release Notes for detailed version history and changes.

TODO List

To configure the environment API key for Quantalogic using LiteLLM, set the required environment variable for your chosen provider and any optional variables like OPENAI_API_BASE or OPENROUTER_REFERRER. Use a .env file or a secrets manager to securely store these keys, and load them in your code using python-dotenv. For advanced configurations, refer to the LiteLLM documentation.

• Python 3.12+
• Docker (optional for code execution tools)

# Basic installation
pip install quantalogic

git clone https://github.com/quantalogic/quantalogic.git
cd quantalogic
python -m venv .venv
source ./venv/bin/activate 
poetry install

pipx install quantalogic

• code: Coding-focused agent with basic capabilities
• basic: General-purpose agent without coding tools
• interpreter: Interactive code execution agent
• full: Full-featured agent with all capabilities
• code-basic: Coding agent with basic reasoning
• search: Web search agent with Wikipedia, DuckDuckGo and SERPApi integration

Tasks can be provided:

1. Directly via task parameter
2. Through a file using --file parameter
3. Interactively via standard input

Using a task file:

quantalogic task --file tasks/example.md --verbose

Selecting agent mode:

quantalogic --mode interpreter task "Explain quantum computing"

Interactive mode:

quantalogic

from quantalogic import Agent

# Initialize agent with default configuration
agent = Agent(model_name="deepseek/deepseek-chat")

# Execute a task
result = agent.solve_task(
    "Create a Python function that calculates the Fibonacci sequence"
)
print(result)

import os

from quantalogic import Agent

# Verify that DEEPSEEK_API_KEY is set
if not os.environ.get("DEEPSEEK_API_KEY"):
    raise ValueError("DEEPSEEK_API_KEY environment variable is not set")

# Initialize the AI agent with default configuration
agent = Agent(model_name="deepseek/deepseek-chat")

# Execute a sample task
result = agent.solve_task("Create a Python function that calculates the Fibonacci sequence")
print(result)

Watch how QuantaLogic can generate complete tutorials from simple prompts:

Example prompt: 04-write-a-tutorial.md

Here are some practical examples to help you get started:

The core agent implements the ReActparadigm, combining:

• Language model reasoning
• Tool execution capabilities
• Memory management
• Event handling
• Task validation

from quantalogic import Agent
from quantalogic.tools import PythonTool, ReadFileTool

# Create agent with specific tools
agent = Agent(
    model_name="openrouter/deepseek/deepseek-chat",
    tools=[
        PythonTool(),
        ReadFileTool()
    ]
)

The ReAct (Reasoning & Action) framework represents a significant advancement in the development of intelligent agents capable of autonomously reasoning through tasks and taking appropriate actions.

QuantaLogic implements this framework, allowing integration with large language models (LLMs) to construct sophisticated agents that can tackle complex problems through natural language interaction.

A ReAct agent utilizes the synergy of reasoning and action. It not only processes natural language inputs but also executes actions in response to these inputs, utilizing various available tools. This functionality is particularly beneficial for environments where complex tasks can be decomposed into manageable subtasks.

QuantaLogic provides an effective implementation of the ReAct framework with several core components:

• Generative Model: This serves as the agent's brain, enabling it to interpret tasks and generate human-like text responses.
• Memory Management: This capability allows the agent to maintain context, keeping track of previous inputs and interactions to provide coherent responses.
• Tool Management: The agent has access to a diverse range of tools, enabling it to perform actions such as code execution, file manipulation, and API communication.

The following state diagram shows the core workflow of a QuantaLogic agent:

stateDiagram-v2
    [*] --> InitializeAgent
    InitializeAgent --> Idle: Agent Initialized

    state Idle {
        [*] --> WaitForTask
        WaitForTask --> SolveTask: Task Received
    }

    state SolveTask {
        [*] --> ResetSession
        ResetSession --> AddSystemPrompt
        AddSystemPrompt --> PreparePrompt
        PreparePrompt --> EmitTaskStartEvent
        EmitTaskStartEvent --> UpdateTokens
        UpdateTokens --> CompactMemoryIfNeeded
        CompactMemoryIfNeeded --> GenerateResponse
        GenerateResponse --> ObserveResponse
        ObserveResponse --> CheckToolExecution
        CheckToolExecution --> TaskComplete: Tool Executed (task_complete)
        CheckToolExecution --> UpdatePrompt: Tool Not Executed
        UpdatePrompt --> UpdateTokens
        TaskComplete --> EmitTaskCompleteEvent
        EmitTaskCompleteEvent --> [*]
    }

    state CompactMemoryIfNeeded {
        [*] --> CheckMemoryOccupancy
        CheckMemoryOccupancy --> CompactMemory: Memory Occupancy > MAX_OCCUPANCY
        CheckMemoryOccupancy --> [*]: Memory Occupancy <= MAX_OCCUPANCY
        CompactMemory --> [*]
    }

    state ObserveResponse {
        [*] --> ProcessResponse
        ProcessResponse --> ExecuteTool: Tool Identified
        ProcessResponse --> UpdateAnswer: No Tool Identified
        ExecuteTool --> UpdateAnswer
        UpdateAnswer --> [*]
    }



    Idle --> [*]: Task Completed
    SolveTask --> Idle: Task Completed

The following sequence diagram illustrates the workflow of a ReAct agent as it processes and solves a task:

sequenceDiagram
    participant User
    participant Agent
    participant ToolManager
    participant Memory

    User->>Agent: Submit task
    Agent->>Memory: Store task details
    Agent->>ToolManager: Retrieve tools
    ToolManager-->>Agent: Provide available tools
    Agent->>Agent: Prepare prompt for task
    Agent->>Agent: Analyze input and generate response
    Agent->>ToolManager: Execute required tool
    ToolManager-->>Agent: Return tool execution result
    Agent->>User: Present final result

1. User Input: The agent begins by receiving a task or question from the user, which initiates the interaction.
2. Memory Management: Before tackling the task, the agent logs relevant task details into its memory, ensuring it has the necessary context for processing.
3. Tool Retrieval: The agent communicates with the ToolManager to inquire about available tools that can facilitate the required actions.
4. Prompt Generation: The agent constructs a prompt that outlines the task specifics, available tools, and any other pertinent context information.
5. Analysis and Response Generation: The agent uses its generative model to analyze the task input and formulate a response.
6. Tool Execution: If certain tools are needed for the task, the agent instructs the ToolManager to execute those tools, fetching the results for processing.
7. Output to User: Finally, the agent compiles and presents the results back to the user.

The QuantaLogic framework incorporates a well-defined tool system that enhances the functionality of AI agents by enabling them to perform a variety of tasks efficiently. Each tool is designed to address specific needs that arise in the context of complex problem-solving and task execution:

1. Core Functionality: Tools such as AgentTool and LLMTool are fundamental to the agent's operation, allowing it to manage tasks and interact with large language models. The integration of these tools enables the agent to process natural language inputs and execute corresponding actions effectively. AgentTool enables the agent to delegate tasks to specialized agents, and LLMTool provides the agent to explore a specific area of a latent space using role play.

2. Code Execution: Tools like PythonTool, NodeJsTool, and ElixirTool are vital for executing code in different programming languages. This capability allows the agent to handle programming tasks directly, facilitating real-time coding assistance and code evaluation.

3. File Operations: The framework includes tools for file management, such as ReadFileTool, WriteFileTool, and ReplaceInFileTool. These tools are essential for enabling the agent to read from and write to files, as well as update file content dynamically. This functionality supports scenarios where agents need to manipulate data or configuration files as part of the task execution process.

4. Search Capabilities: Tools like RipgrepTool and SearchDefinitionNames enhance the agent's ability to search through codebases and identify relevant definitions. This is crucial when dealing with large volumes of code, allowing the agent to quickly locate information necessary for problem-solving.

5. Utility Functions: Additional tools such as DownloadHttpFileTool, ListDirectoryTool, and ExecuteBashCommandTool provide broader functionality that supports various tasks, from fetching external resources to executing system commands. These utilities expand the operational scope of agents, allowing them to perform diverse actions beyond simple text processing.

6. Documentation and Representation: Tools like MarkitdownTool facilitate the generation of documentation, ensuring that output from the agent can be formatted and presented clearly. This is particularly beneficial for creating reports or guides based on the agent's findings and actions.

By integrating these tools into its architecture, QuantaLogic allows agents to perform a wide range of tasks autonomously while ensuring that they have the necessary resources and capabilities to do so effectively. This tool system is fundamental to the agent's ability to reason and act in sophisticated ways, thereby enhancing the overall utility of the framework in complex scenarios.

For detailed documentation of all available tools, please see REFERENCE_TOOLS.md.

# Clone repository
git clone https://github.com/quantalogic/quantalogic.git
cd quantalogic

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
poetry install

# Run all tests
pytest

# With coverage
pytest --cov=quantalogic

# Run specific tests
pytest tests/unit

# Format code
ruff format

# Type checking
mypy quantalogic

# Linting
ruff check quantalogic

1. Fork the repository
2. Create a feature branch
3. Write tests
4. Implement changes
5. Submit pull request

See CONTRIBUTING.md for detailed guidelines.

Copyright 2024 QuantaLogic Contributors

Licensed under the Apache License, Version 2.0. See LICENSE for details.

Initiated with ❤️ by Raphaël MANSUY. Founder of Quantalogic.