RainDigest

Automates the process of fetching video bookmarks from Raindrop.io, extracting metadata/transcripts, and generating structured AI summaries (using Gemini or OpenAI). It even captures keyframes for a richer reading experience in Readwise.

Features

• Auto-Sync: Scans your Raindrop.io collections for new video bookmarks.
• Auto-Classification (Beta): Automatically moves items from "Unsorted" to the most relevant collection using LLM analysis.
• AI Director Mode (Beta): Automatically identifies high-value moments (charts, lists, key slides) in videos under 10 minutes and captures keyframes.
• Smart Summarization: Generates structured notes including "The Gist", "Key Takeaways", and "Critical Perspectives".
• Multi-Source Support: Handles YouTube, Instagram/Facebook Reels, and more via yt-dlp.
• Hybrid Transcription:
  • Uses existing subtitles when available (fast & cheap).
  • Falls back to AI audio-to-text processing (Gemini Multimodal or OpenAI Whisper).
• Readwise Integration: Automatically syncs summaries, metadata, and visual highlights to Readwise Reader.
• Cloud Storage: Optional Cloudflare R2 integration for hosting captured keyframes with automatic 30-day retention cleanup.
• Auto-Tagging: Marks processed items with a #summarized tag in Raindrop.

Quick Start

# 1. Clone
git clone https://github.com/John-Thunder/RainDigest.git
cd RainDigest

# 2. Configure
cp .env.example .env
# Edit .env and fill in your tokens (see below)

# 3. Run (pulls pre-built image from Docker Hub)
docker compose up -d

Want to build from source instead?

docker build -t jjdock1112/raindigest:latest .
docker compose up -d

Setup

Prerequisites

• Docker & Docker Compose
• Raindrop.io Account
• AI API Key: Gemini API Key (Recommended, free tier available).

Configuration

Required:

Variable	Description	Where to get it
RAINDROP_TOKEN	Raindrop.io Test Token	Raindrop Integrations
GEMINI_API_KEY	Google Gemini API Key	Google AI Studio

Optional:

Variable	Default	Description
GEMINI_MODEL	gemini-2.0-flash	Gemini model to use (e.g. gemini-2.5-pro)
GEMINI_PROCESSING_TIMEOUT	300	Max seconds to wait for Gemini file processing
READWISE_TOKEN	—	Sync summaries to Readwise Reader
R2_ACCOUNT_ID	—	Cloudflare R2 (for hosting keyframe images)
R2_ACCESS_KEY_ID	—	Cloudflare R2
R2_SECRET_ACCESS_KEY	—	Cloudflare R2
R2_BUCKET_NAME	—	Cloudflare R2
R2_PUBLIC_DOMAIN	—	Cloudflare R2
MAX_ITEMS	50	Max bookmarks to process per run
DRY_RUN	false	Test fetching without consuming LLM credits
ENABLE_AUTO_ORGANIZER	true	Auto-sort Unsorted items via LLM

Run

Full Service (Summarizer + Organizer):

docker compose up -d

One-shot (no loop):

docker compose run --rm app python -m src.main

Organizer Only:

docker compose run --rm app python organize.py

How It Works

1. Fetch: Scans Raindrop collections.
2. Organize: (If enabled) Checks "Unsorted" collection, asks AI to classify items, and moves them to the best matching collection.
3. Download: Uses yt-dlp to fetch metadata and subtitles. If no subtitles exist, it downloads the audio.
4. AI Director: For short videos, it asks Gemini to "watch" or "listen" and find the best timestamps for screenshots.
5. Summarize: Sends the transcript/audio to the LLM to generate a structured summary.
6. Sync:
   • Saves a local Markdown file in ./output.
   • Uploads keyframes to R2 (if configured).
   • Posts the final summary to Readwise Reader.
   • Tags the bookmark in Raindrop as #summarized.

Customization

1. Modifying AI Prompts

• Core Summary: Change DEFAULT_SYSTEM_PROMPT in src/config.py. This defines the structure (e.g., Gist, Takeaways).
• Title Generation: Modify generate_concise_title in src/llm.py.
• Visual Analysis: Modify analyze_visual_cues or analyze_visual_cues_from_video in src/llm.py to change how the AI finds "important" frames.

2. Adjusting Logic

• Download Parameters: Check src/media.py for resolution or format settings.
• Metadata/Tags: Check src/readwise.py to customize how items appear in Readwise.
• Filtering: Check src/main.py to modify collection filtering or processing priority.

Privacy & Costs

Data Retention

• Google Gemini: Files uploaded to Gemini (Audio/Video) are stored in your Google AI Studio account and usually expire after 48 hours.
• Cloudflare R2: If enabled, keyframes are deleted automatically after 30 days of retention to minimize storage costs and privacy risks.

Cost Estimation

• Gemini: Using gemini-2.0-flash is currently free within generous rate limits (1500 requests/day).
• OpenAI: Using gpt-4o + whisper-1 is paid and costs roughly $0.05 - $0.50 per hour of video depending on complexity.

Roadmap

• AI Director (Visual Highlights)
• Cloudflare R2 Integration
• Support for long-form video (>30 mins) chunking.
• Support for Podcasting feeds.
• Web Dashboard for status monitoring.

Build & Publish

For maintainers — build a multi-arch image and push to Docker Hub:

docker build --platform linux/amd64 -t jjdock1112/raindigest:latest .
docker push jjdock1112/raindigest:latest