europanite/ai-picture-diary：本地 LLM 与 VLM 驱动的自动图片日记

An automated picture diary driven by the local LLM and VLM.

The project combines:

• a FastAPI backend for authentication, RAG, and text generation APIs
• PostgreSQL for application data
• ChromaDB-backed retrieval for local JSON documents
• Ollama or OpenAI-compatible models for generation and embedding
• an Expo / React Native frontend that can be exported for the web
• an illustration worker that generates images for feed entries
• GitHub Actions workflows for feed generation and GitHub Pages deployment

Main services

Requirements

• Docker
• Docker Compose v2
• Git
• Ollama models
• ChromaDB
• Optional: an OpenAI API key when using the OpenAI provider
• Optional: Hugging Face access/cache settings when using model or LoRA assets that require them

Local development

Start the main stack:

docker compose --env-file .env up --build db ollama backend frontend

Check backend health:

curl http://localhost:8000/health

Open the FastAPI docs:

http://localhost:8000/docs

The frontend container runs Expo. Use the URL printed in the frontend logs. Metro and Expo development ports are exposed by Compose.

docker compose logs -f frontend

Stop the stack:

docker compose down

Remove local database and model volumes only when you intentionally want a clean reset:

docker compose down -v

Backend API

The FastAPI app mounts three router groups.

Health

curl http://localhost:8000/health

Authentication

POST /auth/signup
POST /auth/signin
GET  /auth/me

RAG

GET  /rag/status
POST /rag/reindex
POST /rag/ingest
POST /rag/query

Example:

curl -X POST http://localhost:8000/rag/query \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What should be introduced today?",
    "top_k": 5,
    "include_debug": true
  }'

Sentence generation

GET  /diary/health
POST /diary/generate

Example:

curl -X POST http://localhost:8000/diary/generate \
  -H "Content-Type: application/json" \
  -d '{
    "topic": "diary",
    "max_retries": 3,
    "temperature": 0.9
  }'

RAG data

The backend reads JSON documents from DOCS_DIR, which is mounted from ./data/json in Docker Compose.

To rebuild the ChromaDB index from local JSON files:

curl -X POST http://localhost:8000/rag/reindex

To check the current index status:

curl http://localhost:8000/rag/status

The persisted ChromaDB directory is mounted through ./chroma_db and CHROMA_DB_DIR.

Feed generation

The feed assets are stored under frontend/app/public.

Important paths:

frontend/app/public/latest.json
frontend/app/public/feed/
frontend/app/public/snapshot/
frontend/app/public/posts/
frontend/app/public/image/

Generate a sentence with the local script through the backend environment:

docker compose run --rm backend python /scripts/generate_sentence.py

Illustration generation

The illustrate service reads the latest feed entry and writes/patches image references.

Run it after frontend/app/public/latest.json exists:

docker compose run --rm illustrate

Rebuild the feed index and post pages after modifying public feed assets:

docker compose run --rm illustrate \
  python scripts/build_feed_pages.py \
  --public-dir frontend/app/public \
  --base-url .

GitHub Actions

The repository includes workflows for automated feed generation and deployment.