███████╗██╗ ██╗███╗ ██╗ █████╗ ██████╗ ███████╗███████╗██╗ █████╗
██╔════╝╚██╗ ██╔╝████╗ ██║██╔══██╗██╔══██╗██╔════╝██╔════╝██║██╔══██╗
███████╗ ╚████╔╝ ██╔██╗ ██║███████║██████╔╝███████╗█████╗ ██║███████║
╚════██║ ╚██╔╝ ██║╚██╗██║██╔══██║██╔═══╝ ╚════██║██╔══╝ ██║██╔══██║
███████║ ██║ ██║ ╚████║██║ ██║██║ ███████║███████╗██║██║ ██║
╚══════╝ ╚═╝ ╚═╝ ╚═══╝╚═╝ ╚═╝╚═╝ ╚══════╝╚══════╝╚═╝╚═╝ ╚═╝
A local-first multi-agent AI system. Powered by Ollama. Running entirely on your machine.
Synapseia explores what local, modular, and collaborative AI can become when intelligence is distributed across specialized agents — not centralized in a single model. No API keys. No subscriptions. No cloud. Just intelligence, on your terms.
The demo showcases:
- Multi-agent orchestration routing a complex query through specialized agents
- Financial analysis with automated chart generation
- Web research with source-aware, noise-filtered responses
- GitHub repository intelligence and code tracing
- Long-term memory and user personalization — all stored locally
The orchestration layer manages:
- Agent routing — intelligently selects and chains the right agents per query
- Context handling — injects relevant memory and conversation history
- Validation loops — verifies output correctness before delivery
- Structured outputs — formats responses consistently across agents
- Human-in-the-loop — surfaces ambiguity for user clarification when needed
Synapseia is built around a modular set of specialized agents. Each one is independently capable — and exponentially more powerful in combination.
| Agent | Role |
|---|---|
| 🌐 Web Agent | Contextual web research, noise filtering, source-aware responses |
| 📈 Stock Agent | Financial data retrieval, trend analysis, automated visual plots |
| 🧠 Memory Agent | Local memory storage for personalization and long-term continuity |
| 💻 Code Agent | GitHub repo understanding, structure mapping, function tracing |
| ⚡ Raw Agent | Lightweight direct responses with zero orchestration overhead |
Synapseia is built with a zero-trust approach to external services.
✅ Fully offline execution
✅ No API keys required
✅ No subscriptions
✅ No cloud dependency
✅ No telemetry
✅ Complete data ownership
Your data never leaves your machine. Every LLM call, every embedding, every memory write — local.
LLMs & AI
- Ollama runtime — Llama 3.2
- Embedding Models — Nomic-Embed-Text, all-MiniLM-L6-v2, BGE-Small-En
Backend & Databases
- Python, FastAPI
- ChromaDB (vector store)
- Ollama installed and running
- Python 3.10+
- Docker (optional but recommended)
# 1. Clone the repo
git clone https://github.com/Adiittya/Synapseia
cd synapseia
# 2. Install dependencies
pip install -r requirements.txt
# 3. Start Ollama (if not already running)
ollama serve
# 4. Pull a model (e.g., Llama 3.2)
ollama pull llama3.2
# 5. Launch Synapseia
python app.pySynapseia is a proof-of-concept and ongoing exploration of what happens when AI is:
- Local — running on your hardware, not someone else's
- Modular — specialized agents that do one thing exceptionally well
- Collaborative — agents that coordinate, not just coexist
- Private — no external calls, no data leakage, no corporate dependency
The goal isn't to replicate a cloud AI product. It's to demonstrate that a personal AI system — one that knows you, remembers you, and works for you — can exist entirely on commodity hardware.
MIT License — see LICENSE for details.
Built with curiosity. Runs without permission.
A personal AI system. Running entirely on your machine.