Skip to content
/ feagi Public

Evolutionary AGI framework using Spiking Neural Networks (SNN) for embodied systems. Deep robotics integration enabling Physical AI deployment.

License

Unknown, Apache-2.0 licenses found

Licenses found

Unknown
LICENSE
Apache-2.0
LICENSE.txt
Notifications You must be signed in to change notification settings

feagi/feagi

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

FEAGI Python SDK

Complete SDK for building FEAGI agents, controlling the neural engine, and creating marketplace packages

Version 2.0.0 - Framework for Evolutionary Artificial General Intelligence

Overview

FEAGI SDK provides everything needed to:

  • Build agents for robots, simulators, and embedded devices
  • Start and stop the FEAGI neural engine from Python
  • Manipulate genomes and connectomes at runtime
  • Create packages for the FEAGI marketplace

Installation

pip install feagi

For video processing support:

pip install feagi[video]

Quick Start

Basic Agent

from feagi.agent import BaseAgent
from feagi.pns import FeagiAgentClient, AgentType

class MyRobotAgent(BaseAgent):
    def initialize_hardware(self):
        # Connect to your robot/simulator
        pass
    
    def map_sensors(self, hw_data):
        # Convert robot sensors → FEAGI format
        return {"camera": image_bytes}
    
    def map_motors(self, feagi_output):
        # Convert FEAGI commands → robot format
        return motor_commands

# Run the agent
agent = MyRobotAgent("my-robot-01")
await agent.connect()
await agent.run()

Direct PNS Communication

from feagi.pns import FeagiAgentClient, AgentType

# Create client
client = FeagiAgentClient("my-agent", AgentType.SENSORY)
client.configure(feagi_host="localhost")
await client.connect()

# Send sensory data
await client.send_sensory_data(sensor_data)

Architecture

feagi/
├── engine/          # Engine control (start/stop FEAGI)
├── agent/           # Agent framework (BaseAgent templates)
├── genome/          # Runtime genome manipulation
├── connectome/      # Runtime connectome operations
├── packaging/       # Build marketplace packages
├── pns/             # Peripheral Nervous System (communication)
└── cli/             # Command-line tools

Modules

feagi.engine

Start and stop the FEAGI neural engine from Python.

  • FeagiEngine - Control local FEAGI instance
  • PyO3 bindings to Rust feagi::FeagiInstance

Status: Planned for Phase 2

feagi.agent

Agent framework with base classes and templates.

  • BaseAgent - Abstract base class
  • SDKRobotAgent - For SDK-based robots (Cozmo, NAO)
  • SimulatorAgent - For physics simulators (Webots, Gazebo)
  • EmbeddedAgent - For embedded devices (ESP32, Arduino)
  • VirtualAgent - For game engines (Unity, Unreal)

Status: ✅ BaseAgent implemented

feagi.genome

Manipulate genomes on a running FEAGI instance via REST API.

  • GenomeAPI - Add/remove cortical areas, modify parameters
  • GenomeLoader - Load genome from local file
  • GenomeValidator - Validate genome structure

Status: Planned for Phase 3

feagi.connectome

Download and upload connectomes via REST API.

  • ConnectomeAPI - Snapshot/restore trained brain state
  • Download from running instance for backup
  • Upload pre-trained connectomes

Status: Planned for Phase 3

feagi.packaging

Build marketplace packages locally.

  • PackageBuilder - Create .feagi-personality, .feagi-firmware, etc.
  • PackageValidator - Validate package structure
  • ManifestGenerator - Generate manifest.json

Status: Planned for Phase 4

feagi.pns

Peripheral Nervous System - Communication layer.

  • FeagiAgentClient - Modern Rust-backed client (RECOMMENDED)
  • CapabilitiesManager - Device capability management
  • MotorProcessor - Motor command processing
  • VisionProcessor - Vision data processing
  • SegmentedVisionProcessor - Advanced segmented vision
  • Supports ZMQ and WebSocket transports

Status: ✅ Fully implemented

feagi.cli

Command-line tools for development.

  • feagi create-agent - Scaffold new agent from template
  • feagi build-package - Build marketplace package locally

Status: Planned for Phase 4

Development Workflow

1. Create Agent

# Scaffold new agent (future)
feagi create-agent cozmo --template sdk_robot

# Edit generated agent code
vim cozmo_agent.py

2. Test Locally

# Run agent with local FEAGI
python cozmo_agent.py --genome ~/.feagi/genomes/vision_nav.json

3. Build Package

# Create marketplace package (future)
feagi build-package \
    --genome vision_nav.json \
    --docs README.md \
    --output vision_nav_v1.feagi-personality

4. Upload via UI

Upload the package through Neurorobotics Studio or FEAGI Desktop.

Examples

See examples/ directory for complete agent implementations:

  • Basic sensory agent
  • SDK robot agent (Cozmo example)
  • Simulator agent (Webots example)
  • Vision processing examples

Documentation

What's New in 3.0.0

Complete rewrite with clean architecture:

  • ✅ Renamed from feagi_connector to feagi
  • ✅ Modular structure: engine, agent, genome, connectome, packaging, pns, cli
  • ✅ No legacy code, no deprecated APIs, no fallbacks
  • ✅ Modern BaseAgent framework for agent development
  • feagi.pns module for Peripheral Nervous System communication
  • ✅ Prepared for engine control (PyO3 bindings in Phase 2)
  • ✅ Prepared for runtime genome/connectome manipulation (Phase 3)
  • ✅ Prepared for marketplace package building (Phase 4)

Breaking Changes:

  • Package renamed: feagi_connectorfeagi
  • All legacy clients removed (FeagiClient, FeagiAgentConnector)
  • Only modern FeagiAgentClient supported
  • Python 3.10+ required
  • Imports changed: from feagi_connector import Xfrom feagi.pns import X

Migration from 2.x

# Old (2.x)
from feagi_connector import FeagiAgentClient

# New (3.0)
from feagi.pns import FeagiAgentClient

Requirements

  • Python 3.10+
  • feagi_rust_py_libs (for Rust-backed performance)
  • NumPy 1.20+
  • PyZMQ 24.0+
  • aiohttp 3.9+

Contributing

See CONTRIBUTING.md for guidelines.

License

Apache-2.0 - See LICENSE.txt for details.

Links

Authors

Neuraville Inc. - [email protected]

Copyright 2016-2025 Neuraville Inc. All Rights Reserved.

About

Evolutionary AGI framework using Spiking Neural Networks (SNN) for embodied systems. Deep robotics integration enabling Physical AI deployment.

Resources

License

Unknown, Apache-2.0 licenses found

Licenses found

Unknown
LICENSE
Apache-2.0
LICENSE.txt

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

No packages published

Contributors 40