hermes-eeg 0.1.0

EEG/BCI neural interface plugin for Hermes Agent — connects OpenBCI hardware to AI agents via real-time emotion detection

hermes-eeg 🧠

EEG/BCI neural interface plugin for Hermes Agent. Connects OpenBCI hardware to AI agents via real-time emotion detection and AI-readable "felt experience" generation.

The core innovation: AI agents can perceive how humans experience music, content, or any stimulus — enabling a closed feedback loop between human emotion and AI creation.

Features

• OpenBCI Hardware Support — Cyton (8-ch), Ganglion (4-ch), Synthetic (test data)
• Works Without Hardware — Mock mode with realistic simulated EEG for development/testing
• Real-time Emotion Detection — Valence, arousal, attention, engagement at 2Hz
• Musical Chills Detection — Gamma bursts + theta coupling = frisson events
• Felt Experience Format — AI-readable session data with emotional arcs and narratives
• Session Recording — Persistent storage of listening sessions as JSON

Installation

# Core (mock mode, no hardware needed)
pip install hermes-eeg

# With OpenBCI hardware support
pip install "hermes-eeg[hardware]"

# Enable the toolset in Hermes
hermes tools enable eeg

Tools

Tool	Description
eeg_connect	Connect to OpenBCI board (or mock/synthetic)
eeg_disconnect	Disconnect and release resources
eeg_stream_start	Start recording a listening session
eeg_stream_stop	Stop and generate felt experience format
eeg_realtime_emotion	Get live emotional state
eeg_experience_get	Retrieve past session data
eeg_calibrate_baseline	Prepare personal baseline calibration
eeg_list_sessions	Browse recorded sessions

Quick Start

# In Hermes chat:
> Connect to EEG in mock mode and start a test session

# The agent will:
# 1. eeg_connect(serial_port="", board_type="mock")
# 2. eeg_stream_start(session_name="test", track_title="My Song")
# 3. ... record emotional data in background ...
# 4. eeg_stream_stop() → generates narrative + summary

How It Works

Signal Processing Pipeline

1. Raw EEG → Detrend → Notch filter (50/60Hz) → Bandpass (0.5-45Hz)
2. Band Power Extraction (Welch's method):
   • Theta (4-8 Hz) — Emotional processing
   • Alpha (8-13 Hz) — Relaxation
   • Beta (13-30 Hz) — Arousal
   • Gamma (30-45 Hz) — Peak experience / "chills"
3. Emotion Mapping:
   • Valence (-1 to +1): Frontal alpha asymmetry (F4-F3)
   • Arousal (0-1): Beta/alpha ratio
   • Attention (0-1): Theta/beta ratio + gamma
   • Engagement (0-1): Geometric mean of arousal × attention

Felt Experience Format

Sessions are saved as JSON with:

• Per-moment emotional dimensions (valence, arousal, attention, engagement)
• Event flags (attention_shift, emotional_peak, possible_chills)
• Summary statistics
• Natural language narrative for AI consumption

Dependencies

• numpy + scipy — Always required (signal processing)
• brainflow — Optional (real OpenBCI hardware). Falls back to mock/SciPy without it.

Development

git clone https://github.com/buckster123/hermes-eeg-plugin
cd hermes-eeg-plugin
pip install -e ".[dev]"
pytest tests/ -v

Based On

Extracted from ApexAurum — the neural resonance system for AI-human creative feedback loops.

License

MIT