iRemembo

iRemembo is a lightweight photo-memory project.

It stores:

• a remembered image copy in Dropbox
• searchable metadata in a local SQLite database
• optional embeddings in a local embedding table

The intended primary workflow is interactive:

1. the user sends an image in chat
2. the user explicitly says to remember it
3. chat-side image understanding produces structured metadata
4. iRemembo writes that metadata into the local DB and uploads the retained image copy to Dropbox

What the project does

• remembers only images explicitly marked to keep
• stores image identity by dropbox_path
• keeps summary / OCR text / tags / entities searchable locally
• supports optional embedding generation for later semantic search
• keeps secrets and runtime data outside the repo

Repository layout

• src/photo_memory.py — main CLI
• scripts/dropbox_tool.py — Dropbox helper
• config/example.local.json — local config template
• docs/architecture.md — architecture overview
• .env.example — example environment variables

Local-only data

Keep these outside the repo:

• real config
• SQLite DB
• tokens / secrets
• local thumbnails or caches
• logs / runtime state

Setup

1. Copy config/example.local.json to a local-only location.
2. Fill in your local paths.
3. Either set environment variables, or place local-only config files at the default convention paths.
4. Initialize the database.

Example with environment variables:

export IREMEMBO_CONFIG=~/iRemembo-local/config.json
export DROPBOX_CONFIG=~/secrets/dropbox.json
python3 src/photo_memory.py init

Example with default local convention paths:

mkdir -p ~/.config/iremembo
cp config/example.local.json ~/.config/iremembo/config.json
# put your Dropbox secret JSON at ~/.config/iremembo/dropbox.json
python3 src/photo_memory.py init --config ~/.config/iremembo/config.json

Main commands

python3 src/photo_memory.py init
python3 src/photo_memory.py remember-chat /path/to/image.jpg --analysis-json '{"summary":"示例","tags":["標籤1"],"entities":{"objects":["照片"]},"ocr_text":""}' --auto-embed
python3 src/photo_memory.py remember /path/to/image.jpg --summary "示例" --tags "標籤1,標籤2" --auto-embed
python3 src/photo_memory.py add /path/to/image.jpg --summary "示例" --tags "標籤1,標籤2"
python3 src/photo_memory.py annotate 1 --summary "更新後摘要" --tags "標籤1,標籤2,標籤3"
python3 src/photo_memory.py embed 1
python3 src/photo_memory.py inspect /path/to/image.jpg
python3 src/photo_memory.py find 關鍵字
python3 src/photo_memory.py search 關鍵字
python3 src/photo_memory.py search 關鍵字 --semantic
python3 src/photo_memory.py fetch 1

Data model

The photos table keeps:

• dropbox_path
• sha256
• summary
• ocr_text
• tags_json
• entities_json
• user_note
• timestamps
• status
• embedding reference fields

The actual embedding vector is stored in photo_embeddings.

Notes

• CLI-side auto analysis is pluggable through analysis_command.
• CLI does not use OpenAI vision fallback.
• Embeddings currently use OpenAI /v1/embeddings when enabled.
• Duplicate detection is SHA-256 based.
• Retrieval now has two layers: find for plain keyword matching, and search --semantic for embedding-based ranking.
• scripts/remember_to_iremembo.py prefers explicit env vars, but if they are absent it will also look for local-only files at ~/.config/iremembo/config.json and ~/.config/iremembo/dropbox.json.

Config example

{
  "db_path": "/absolute/path/to/iRemembo-local/photo-memory.db",
  "thumb_dir": "/absolute/path/to/iRemembo-local/thumbs",
  "dropbox_base": "/photo-memory",
  "dropbox_tool": "/absolute/path/to/iRemembo/scripts/dropbox_tool.py",
  "embedding_model": "text-embedding-3-small",
  "ocr_command": [],
  "analysis_command": []
}