Synapse Swarm

A three-agent cognitive swarm system for real-time visual analysis using Hermes Agent.

When to Use

• Analyzing camera or image input with multi-perspective AI reasoning
• Running a chained agent pipeline: vision → strategy → arbitration
• Building operator dashboards with real-time agent collaboration
• Voice-controlled agent workflows with text-to-speech feedback
• Tactical or creative analysis requiring multiple AI viewpoints

How It Works

Synapse Swarm chains three specialized agents in sequence:

1. ZERO (Vision Analyst) — Receives the operator's command and a camera snapshot. Describes key visual details with tactical precision.
2. NOVA (Tactical Strategist) — Reads ZERO's visual scan and the original command. Suggests a creative strategic angle based on the observed object or scene.
3. TITAN (Arbiter) — Receives all prior context — the command, ZERO's scan, and NOVA's tactic — then delivers a final one-sentence verdict. TITAN's response is also spoken aloud via text-to-speech.

Each agent calls hermes chat under the hood, forming a chain-of-thought pipeline where each agent builds on the previous one's output.

Architecture

Operator Command (voice or text)
        │
        ▼
   ┌─────────┐
   │  ZERO   │  ← Camera snapshot + command
   │ (Vision)│
   └────┬────┘
        │ visual scan
        ▼
   ┌─────────┐
   │  NOVA   │  ← ZERO's scan + command
   │(Tactical)│
   └────┬────┘
        │ tactical angle
        ▼
   ┌─────────┐
   │  TITAN  │  ← All context → final verdict (+ TTS)
   │(Arbiter)│
   └─────────┘

Features

• Chained multi-agent reasoning — Each agent receives the full context chain from previous agents
• Customizable system prompts — Modify each agent's core directives via the sidebar
• Dual input modes — Voice commands (auto-detected language: TR/EN) and text input
• Text-to-speech output — TITAN's final verdict is spoken aloud automatically
• Export reports — Download the full swarm analysis as a .txt file
• Cyberpunk Streamlit UI — Animated GIF avatars, translucent panels, neon accents

Requirements

• Python 3.8+
• Hermes Agent CLI (hermes chat)
• Streamlit
• Pillow, SpeechRecognition, gTTS, audiorecorder
• Camera access (for live snapshot input)

Quick Start

pip install streamlit pillow SpeechRecognition gTTS streamlit-audiorecorder
streamlit run app.py

Customization

Agent prompts can be edited in the sidebar at runtime. Each prompt supports these variables:

• {command} — The operator's directive
• {zero_scan} — ZERO's visual analysis output
• {nova_tactic} — NOVA's tactical suggestion

Source

• Repository: github.com/devorun/synapse-swarm
• Author: @Devran1An