Full web interface for the chatbot — real-time responses via WebSocket.
FastAPI handles AI inference, Django serves the UI, Redis bridges them with Channels.
| Capability | Detail |
|---|---|
| LLM backend | Mistral-7B-Instruct-v0.2 via LlamaCpp (GGUF, CPU) |
| Quantization | Q4_K_M — ~4.5GB RAM, full Mistral-7B weights |
| Memory strategies | Buffer · Window · Summary · SummaryBuffer · VectorStore |
| Vector store | ChromaDB with sentence-transformers/all-MiniLM-L6-v2 embeddings |
| REST API | FastAPI — /chat, /chat/{id}/history, /chat/{id} DELETE |
| Web UI | Django + Channels — real-time chat via WebSocket |
| Session management | Per-user ChatbotChain instances, auto-generated session IDs |
| Real-time comms | WebSocket (Daphne + channels-redis) |
| Session persistence | Redis (channels layer + session cache) |
C:\Projects\chatbot\
├── api\
│ ├── __init__.py
│ ├── main.py ← FastAPI app entry point
│ ├── routes.py ← /chat, /history, /health endpoints
│ └── session_store.py ← per-user ChatbotChain instances
├── core\
│ ├── __init__.py
│ ├── llm_engine.py ← LlamaCpp (GGUF) or Ollama
│ ├── memory.py ← MemoryFactory (5 memory types)
│ └── chain.py ← ChatbotChain (LLM + memory + Mistral prompt)
├── web_app\ ← auto-created by django-admin
│ ├── __init__.py
│ ├── settings.py
│ ├── urls.py
│ ├── asgi.py
│ └── wsgi.py
├── chatui\
│ ├── __init__.py
│ ├── consumers.py ← WebSocket consumer, proxies to FastAPI
│ ├── routing.py ← WebSocket URL routing
│ ├── urls.py ← HTTP URL routing
│ ├── views.py ← chat_view, clear_chat
│ └── templates\
│ └── chatui\
│ └── chat.html ← full chat UI
├── models\
│ └── mistral-7b-instruct-v0.2.Q4_K_M.gguf
├── chatbot-env\
├── requirements.txt
└── .env
| Tool | Version | Download |
|---|---|---|
| Python | 3.11.x | https://python.org/downloads — tick Add to PATH |
| Redis | 5.0+ | https://github.com/tporadowski/redis/releases |
| Git | Any | https://git-scm.com/download/win |
No GPU required. Minimum RAM: 14GB total (close other apps before running).
Redis must be 5.0+ — older versions don't supportchannels-redis.
cd C:\Projects\chatbot
chatbot-env\Scripts\activate
pip install django channels channels-redis daphne httpxDownload (~4.4GB) and save to C:\Projects\chatbot\models\:
https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf
Window 1 — Redis:
& "C:\Program Files\Redis\redis-server.exe"Verify: & "C:\Program Files\Redis\redis-cli.exe" ping → should return PONG
Window 2 — FastAPI:
cd C:\Projects\chatbot
chatbot-env\Scripts\activate
uvicorn api.main:app --reload --port 8000Window 3 — Django:
cd C:\Projects\chatbot
chatbot-env\Scripts\activate
python manage.py migrate
daphne -b 0.0.0.0 -p 8081 web_app.asgi:applicationOpen http://localhost:8081
Browser (WebSocket)
↓
Django + Daphne (port 8081)
↓ HTTP POST
FastAPI (port 8000)
↓
LangChain Chain + Memory
↓
Mistral-7B via LlamaCpp
↓
Redis (channels layer, port 6379)
Every message goes: browser → Django WebSocket consumer → FastAPI /chat → LLM → back through WebSocket to browser.
| Error | Cause | Fix |
|---|---|---|
unknown command 'BZPOPMIN' |
Redis version too old | Install Redis 5.0+ from tporadowski/redis |
WinError 10048 on port 8080 |
Port already in use | Use --port 8081 instead |
| Duplicate welcome messages | WebSocket reconnecting | Remove greeting from consumers.py connect() |
| Typing indicator stuck, no reply | FastAPI not running | Check uvicorn window for errors |
(chatbot-env) not in prompt |
venv not activated | Run chatbot-env\Scripts\activate |
| Version | Focus | Status |
|---|---|---|
| v0.1 | Core LLM + Memory (local console) | ✅ Done |
| v0.2 | FastAPI REST layer + session management | ✅ Done |
| v0.3 | Django web UI + WebSocket streaming | ✅ Done |
| v0.4 | Image upload + OpenCV + BLIP captioning | 🔜 Next |
| v0.5 | Docker + Docker Compose | — |
| v0.6 | AWS / GCP cloud deployment | — |