🧠 CYNAPSE - Ghost Shell AI Agent

Modular AI Agent System for the Terminal

CYNAPSE is a modular AI agent that runs entirely in your terminal. Built with Go and Bubble Tea, it combines the power of multiple LLM providers with a beautiful TUI interface and extensible synapse (plugin) system.

Philosophy: Like Arch Linux for AI - install only what you need, control everything, no bloat.

---

⚡ Quick Install

curl -fsSL https://raw.githubusercontent.com/Alartist40/cynapse/main/install.sh | bash

That's it! The installer will:

• Auto-detect your OS and architecture
• Install Go and dependencies if needed
• Build CYNAPSE
• Set up ~/.cynapse/ directory
• Add cynapse to your PATH

---

🚀 Usage

Run Interactive Chat

cynapse

Manage Synapses (Extensions)

# List installed synapses
cynapse synapse list

# Install a synapse
cynapse synapse add leafcutter

# Remove a synapse
cynapse synapse remove git-tools

# Search available synapses
cynapse synapse search inference

Configuration

# Show config location
cynapse config

# Create default config
cynapse config init

# Edit config
nano ~/.cynapse/config.yaml

---

📦 Synapses (Optional Extensions)

CYNAPSE uses a modular synapse system - install only what you need:

Synapse	Description	Install
leafcutter	CPU-optimized LLM inference engine	cynapse synapse add leafcutter
git-tools	Repository management and analysis	cynapse synapse add git-tools
web-automation	Browser control and scraping	cynapse synapse add web-automation
speedtest	LLM benchmarking and metrics	cynapse synapse add speedtest

Like biological synapses: Each extension creates new neural connections, expanding capabilities without bloating the core.

---

🎨 Features

• ✅ Beautiful TUI - Purple & orange themed terminal interface
• ✅ Streaming Responses - See text appear word-by-word
• ✅ Multi-LLM Support - Ollama, Anthropic, OpenAI, Gemini
• ✅ Model Switching - Change models on the fly with / menu
• ✅ Memory System - Persona-driven memory with SQLite storage
• ✅ MCP Integration - Model Context Protocol for tool calling
• ✅ Modular Synapses - Extend functionality with plugins
• ✅ Request Cancellation - Cancel long-running requests
• ✅ Session Management - JSONL conversation logs
• ✅ Cross-platform - Linux, macOS, Windows

---

🧠 DENDRITE Memory System

CYNAPSE uses DENDRITE, a neural-inspired graph memory system. Knowledge is stored as interconnected nodes (neurons) rather than flat files.

• Interconnected Knowledge: Uses [[wiki-links]] to create relationships between concepts.
• Bi-directional Discovery: Backlinks are maintained automatically for every link.
• Neural Context: The agent intelligently scores and retrieves relevant nodes based on conversation context.
• Visual Explorer: Interactive D3.js graph visualization (works 100% offline).
• Search: Full-text search powered by SQLite FTS5.

Graph Nodes (Neurons)

When you reference a node that doesn't exist yet (e.g., [[future-node]]), DENDRITE creates a "placeholder" to hold the backlink. This is then "upgraded" to a full node when you create it. This ensures connections are never lost, even if nodes are created out of order.

Best Practice: Create nodes before linking them to ensure all metadata is persisted immediately.

---

🏗️ Architecture

CYNAPSE Core (Your Brain)
├─ TUI Interface
├─ Agent System
├─ Memory & Persona
└─ MCP Manager
    └─ Synapses (Load on demand)
        ├─ LeafcutterLLM
        ├─ GitTools  
        ├─ WebAutomation
        └─ Your custom extensions...

---

💻 Manual Installation (Advanced)

If you prefer to build manually:

# Clone repository
git clone https://github.com/Alartist40/cynapse.git
cd cynapse

# Install dependencies (Linux/Debian)
sudo apt-get install build-essential pkg-config libopenblas-dev libsqlite3-dev

# Build
go build -o cynapse ./cmd/cynapse

# Install
sudo mv cynapse /usr/local/bin/

# Setup home directory
mkdir -p ~/.cynapse/{synapses,data,logs}
cp config.yaml ~/.cynapse/

# Run
cynapse

---

🔧 Configuration

CYNAPSE uses ~/.cynapse/config.yaml:

# LLM Provider
llm:
  provider: "ollama"              # ollama | anthropic | openai | gemini
  model: "qwen3.5:9b"
  ollama_base_url: "http://localhost:11434"
  max_tokens: 4096
  temperature: 0.7

# Memory System
memory:
  persona_path: "~/.cynapse/data/persona"
  sessions_path: "~/.cynapse/data/sessions"
  db_path: "~/.cynapse/data/memory.db"

# MCP Servers (auto-loaded from synapses)
mcp:
  enabled: true
  servers: []

# Tools
tools:
  profile: "standard"             # minimal | standard | full
  work_dir: "./workspace"

---

🎯 Building Your Own Synapse

Want to build custom extensions? Use the synapse template:

# Clone template
git clone https://github.com/Alartist40/cynapse-synapse-template my-synapse
cd my-synapse

# Implement your tools in internal/tools.go
# Update synapse.yaml with metadata

# Build
go build -o my-synapse ./cmd/synapse

# Test
echo '{"jsonrpc":"2.0","id":1,"method":"initialize"}' | ./my-synapse

# Users can install with:
# cynapse synapse add my-synapse

See Synapse Development Guide for details.

---

📊 System Requirements

Minimum:

• Go 1.22+ (auto-installed by installer)
• 2GB RAM
• 500MB disk space

Recommended:

• 4GB+ RAM (for larger models)
• 8GB+ RAM (for LeafcutterLLM with 7B models)

Supported Platforms:

• Linux (x86_64, arm64, armv7)
• macOS (x86_64, arm64)
• Windows (x86_64)

Tested On:

• Raspberry Pi 5 (8GB)
• Raspberry Pi Zero 2W
• Ubuntu 22.04+
• macOS Ventura+
• Windows 10/11

---

🤝 Contributing

Contributions welcome! Please read CONTRIBUTING.md.

Areas to contribute:

• New synapse development
• LLM provider integrations
• Documentation improvements
• Bug fixes and performance optimizations

---

📜 License

MIT License - see LICENSE file.

---

🙏 Acknowledgments

• Built with Bubble Tea TUI framework
• Inspired by biological neural networks
• MCP protocol by Anthropic

---

📬 Contact

Author: Alartist40
Repository: github.com/Alartist40/cynapse
Website: alartist40.github.io

---

Built with 💜 for the terminal