Mentals AI enables the creation and operation of AI agents using natural language descriptions in .gen files, abstracting away traditional coding for agent logic. It targets developers and researchers seeking a flexible, code-free framework for building complex, multi-agent systems with memory and tool integration.

How It Works

Mentals AI employs a recursive agent executor driven by an LLM that determines subsequent actions based on previous loop outputs. Agents are defined in Markdown-like .gen files, allowing for natural language instructions, tool usage, and control flow. This approach aims to provide a "blank canvas" for implementing diverse reasoning frameworks like ReAct or Tree of Thoughts directly in natural language, without requiring explicit programming.

Quick Start & Requirements

• Install prerequisites: libcurl, libfmt, pgvector, poppler (Linux: apt-get, macOS: brew).
• Clone the repository: git clone https://github.com/turing-machines/mentals-ai.
• Configure config.toml with an OpenAI API key.
• Build the project: make.
• Run an agent: ./build/mentals agents/loop.gen.
• Official documentation and examples are available in the repository.

Highlighted Details

• Agents are defined in .gen files using natural language, supporting instruction chaining and tool integration.
• Supports custom reasoning frameworks (ReAct, Tree of Thoughts) and multi-agent interactions.
• Includes native tools for file handling, Python/Bash execution, and user input.
• Experimental short-term memory for cross-instruction context sharing.

Maintenance & Community

The project is marked as "work in progress" with ongoing development for a Web UI and local vector database. Links to community channels are not explicitly provided in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility with non-OpenAI models like Llama3 is mentioned for instruction calls.

Limitations & Caveats

The project is explicitly labeled as "work in progress" with experimental features. A local vector database and Web UI are under development. The README does not detail specific limitations or known bugs.