An LLM-powered agent that intelligently manage network based on your intents in natural language in a software-defined networking (SDN) environment.

• Rigid and manual workflows: Administrators must translate high-level business goals into complex, vendor-specific flow rules. This is slow and prone to human error.
• Context-blindness and hallucination: General LLMs lack real-time topology and network state, so they can produce incorrect or unsafe configurations.
• Messy management and static knowledge: There is no unified interface for monitoring and control, and no feedback-driven knowledge base to improve the model over time.

• Integrated dashboard and knowledge base: Provide a real-time dashboard with a feedback-driven knowledge base (vector DB) that stores and retrieves successful configuration samples.
• Context-aware RAG pipeline: Inject live topology and similar configuration samples into the LLM prompt to improve technical accuracy.
• User intent translation: Build a natural language interface that converts high-level intents into ONOS-compatible JSON configurations.

1. Connectivity：

   • HostToHost Intent

2. QoS control:

   • HostToHost Intent with specific traffic types selector and priority
   • Plan to further extend onos intent framework that allows OpenFLow meter injection

3. Load-balancing

   • HostToHost Intent with specific traffic types selector and path
   • Plan to develop a load balancer to be extended in onos intent framework

4. Access control

   • FlowObjective intent with selector and treatment

5. Topology awareness and status retrieval

   • A respontive UI developed using React

1. Dynamic intent conflict resolution
2. Rely on openflow flow rules instead of onos intent framework to expand scope
3. Fine-tune LLM with more data collection
4. Multi-controller, multi language support

Example intents:

• “Throttle bulk backups after 1 a.m. to keep latency low for production.”
• “Give the ‘VideoConf’ app higher priority on VLAN 20 until 6 p.m.”
• “Block SSH to servers outside the bastion for interns group.”

• refer here

• Significant components in this project

FYP/
├─ backend/ 
  ├─ main.py (FastAPI entrypoint)
  ├─ api/ (auth, chat, conversations, devices, hosts, config samples, tests, ONOS proxy)
  ├─ schemas/ (Pydantic request/response models)
  ├─ services/ (auth, chat, devices, config samples, llm, onos, rag, testbed, users)
├─ database/ 
  ├─ models.py (SQLAlchemy models)
  ├─ database.py (engine/session)
  ├─ seed_data.py (seed scripts)
  ├─ rag/ (embeddings + similarity search)
  ├─ data/ (seed JSON files)
├─ diagram/
├─ evaluation/ 
├─ frontend/
  ├─ src
    ├─ auth/ (Auth context)
    ├─ components/ (dashboard, chat, pages, layout)
    ├─ hooks/ (topology data)
    ├─ utils/ (API clients)
├─ llm-engine/ (legacy local LLM experiments)
├─ onos-testbed/
  ├─ scripts/ (topology + test runner)
  ├─ notes/ (testbed docs)
├─ main.py

• SDN Testbed
  • ONOS Controller + Intent Framework
  • Open vSwitch
  • Linux namespaces (Mininet-based testbed)
• Frontend
  • Vite + React + TypeScript
  • Tailwind CSS
  • D3-force + Chart.js
• Backend
  • FastAPI + Pydantic
  • Groq API client (LLM inference)
  • RAG with SentenceTransformers
• Database
  • PostgreSQL + pgvector
  • SQLAlchemy

• Node.js 18+ (LTS recommended)
• Package manager: npm, pnpm, or yarn

1. 8000: Backend API
2. 5173: React frontend
3. 6653: ONOS Listening OpenFlow
4. 8181: ONOS GUI

• CPU: Intel(R) Xeon(R) CPU D-1528 @ 1.90GHz
• Memory: 32 GB
• OS: Ubuntu 24.04.3 LTS
• No. of threads: 12

1. Set required environment variables

export GROQ_API_KEY=your_key
export ONOS_API_URL=http://127.0.0.1:8181/onos/v1
export ONOS_USER=onos
export ONOS_PASS=rocks

2. Start backend FastAPI server

cd FYP
uv run uvicorn backend.main:app --host 0.0.0.0 --port 8000

3. Start frontend (separate terminal)

cd FYP/frontend
npm install
npm run dev

Note: llm-engine/agent.py and llm-engine/ are legacy local-LLM experiments. The current backend uses the Groq API by default.