A multi-agent AI system where specialized agents debate, critique, and synthesize decisions — just like a real engineering council.
Tiny Council is an experimental multi-agent AI architecture where multiple autonomous AI agents collaborate, debate, and challenge each other to arrive at high-quality, explainable decisions.
Instead of relying on a single AI response, Tiny Council forms a council of experts, each with a distinct role — mimicking how real-world engineering teams think.
Traditional AI systems:
- Provide single-point answers
- Lack internal critique
- Hide reasoning paths
Tiny Council introduces:
- 🗣️ Inter-agent debate
- ⚖️ Conflict-driven reasoning
- 🔍 Explicit critique and validation
- 🧩 Consensus-based decision making
This makes the system more robust, explainable, and realistic.
User Query
|
v
+-------------------+
| Council Manager |
+-------------------+
| | |
v v v
+---------+ +---------+ +---------+
| Planner | | Critic | | Expert |
+---------+ +---------+ +---------+
\ | /
\ | /
v v v
+-------------------+
| Consensus Engine |
+-------------------+
|
v
Final Answer + Reasoning Trace
| Agent | Role |
|---|---|
| 🧠 Planner | Breaks down the problem and proposes a solution |
| 🔎 Critic | Challenges assumptions, finds flaws and risks |
| 📚 Domain Expert | Ensures factual and technical correctness |
| 🧩 Synthesizer | Merges arguments into a final response |
Each agent operates independently and communicates through a structured debate loop.
- Language: Python
- Backend: FastAPI
- AI Layer: Pluggable LLM interface (OpenAI / Local models)
- Data Models: Pydantic
- Architecture: Modular, agent-based design
tiny-council/
│
├── agents/
│ ├── base_agent.py
│ ├── planner.py
│ ├── critic.py
│ ├── expert.py
│ └── synthesizer.py
│
├── council/
│ ├── manager.py
│ └── consensus.py
│
├── prompts/
│ ├── planner.txt
│ ├── critic.txt
│ └── expert.txt
│
├── api/
│ └── main.py
│
├── tests/
├── requirements.txt
└── README.md
- ✔️ Modular agent system
- ✔️ Role-specific prompting
- ✔️ Turn-based agent debate
- ✔️ Consensus synthesis
- Core agent framework
- Single-round debate
- Multi-round discussions
- Agent scoring & voting
- Persistent memory per agent
- Explainable reasoning logs
- Web UI for live agent debates
- Exportable decision reports
- AI research & experimentation
- Engineering decision support
- Code review & system design critique
- GSoC / research internship portfolios
- Explainable AI demonstrations
This project is licensed under the MIT License — free to use, modify, and distribute.
Lead Dev
Bhavesh
🌐 GitHub
|