How hemanthreddykunduru/multi-agent-orchestration-system works
The implemented product is an interactive command-line assistant that selects one Ollama model per query, returns its response, and records the exchange. The README describes collaboration and aggregation, but the sampled runtime instead routes each query to a single model; those unsupported capabilities are not shown as implemented. The standalone agent scripts provide separate model-specific entry points. The diagram reflects the sampled MAS.py flow and documented model specializations.
Main entry points: MAS.py Interactive CLI · chat.py Mistral · code.py Qwen Coder · img.py LLaMA Vision
Components 12
Interaction
- Interactive CLI
MAS.py - Printed Response
MAS.py
Routing and Inference
- Task Router
MAS.py - Ollama Invocation
MAS.py
Model Agents
- Mistral
individuals/chat.py - Qwen Coder
individuals/code.py - LLaMA Vision
individuals/img.py - Mathstral
individuals/mat.py
Session and History
- Recent Context
MAS.py - Chat History
history.db
Other components
- User
- Ollama Runtime
Connections 13
- User → Interactive CLI: enters query
- Interactive CLI → Task Router: classifies task
- Interactive CLI → Ollama Invocation: invokes model
- Ollama Invocation → Recent Context: reads recent turns
- Ollama Invocation → Ollama Runtime: runs selected model
- Ollama Runtime → Ollama Invocation: returns response
- Interactive CLI → Chat History: saves exchange
- Interactive CLI → Printed Response: prints response
- Printed Response → User: shows result
- Ollama Runtime → Mistral: runs general tasks
- Ollama Runtime → Qwen Coder: runs code tasks
- Ollama Runtime → LLaMA Vision: runs image tasks
- Ollama Runtime → Mathstral: runs math tasks