Sense | Act | Brain | Reflect
SABR is a modular Python framework designed for building autonomous AI agents through a strictly decoupled architecture. It separates environment perception, decision-making, and tool execution into distinct, protocol-driven components.
The framework orchestrates a control loop that abstracts the interaction between a Large Language Model (LLM) and complex external systems through four primary pillars:
- Perceptor: Observes the environment and normalizes data into a structured
Observation. - Brain: A reasoning layer (e.g., Gemini, GPT) that processes observations to decide on an
Action. - Executor: Routes actions to specific tools or executes dynamically generated code.
- Reporter: Provides observability by logging the internal state and decision chain to multiple sinks (Console, Web UI, etc.).
The repository is organized into two main layers to ensure high extensibility:
sab_core/: The hardware-agnostic core engine and base protocols.engines/: Domain-specific implementations (e.g.,engines/aiidafor scientific data management).
SABR is currently in the MVP (Minimum Viable Product) phase.
- Core Engine: Fully asynchronous, supporting non-blocking execution loops.
- Brain Integration: Implementation for Google Gemini 2.0 with native function calling support.
- AiiDA Engine: A reference implementation for the AiiDA provenance database, featuring automated resource scanning and a Python code interpreter.
- Web Interface: A NiceGUI-based dashboard for real-time monitoring and human-in-the-loop interaction.
Clone the repository and install in editable mode:
git clone [https://github.com/your-username/sabr.git](https://github.com/your-username/sabr.git)
cd sabr
pip install -e .