High Level Customization

For power users, FeedCraft offers a dashboard to customize how feeds are processed.

Accessing the Dashboard

1. Deploy FeedCraft using Docker (see Quick Start).
2. Navigate to http://your-server-ip:10088.
3. Log in with the default credentials:
   • Username: admin
   • Password: adminadmin (Please change the password immediately after logging in)

Core Concepts

AtomCraft

An AtomCraft is a single processing unit. While there are built-in AtomCrafts (like translate-title, fulltext), you can create your own based on templates.

Example: Custom Translation Prompt You can create a new AtomCraft named translate-to-french based on the translate-content template, but with a custom prompt instructing the AI to translate to French.

FlowCraft

A FlowCraft is a sequence of AtomCrafts. This allows you to chain multiple operations together.

Example: Fulltext + Summary + Translation You can define a FlowCraft named digest-and-translate that consists of:

1. fulltext (Extract content)
2. summary (Generate summary)
3. translate-content (Translate the result)

Managing FlowCrafts

You can create and manage FlowCrafts in the FlowCraft section of the dashboard. The editor allows you to add AtomCrafts and arrange their execution order. Use the arrow buttons (⬆️/⬇️) to reorder AtomCrafts, or the trash icon to remove them from the FlowCraft. (See 2b2439b)

Recipe

A Recipe binds a specific RSS feed URL to an AtomCraft or FlowCraft. This allows you to create a persistent, customized feed URL.

Example:

• Input URL: https://news.ycombinator.com/rss
• Processor: digest-and-translate (The flow created above)
• Result: You get a new FeedCraft URL that serves Hacker News with full text, summaries, and translation.

Advanced Configuration

Docker Environment Variables

You can configure FeedCraft using environment variables in your docker-compose.yml.

• FC_PUPPETEER_HTTP_ENDPOINT: URL of a browserless/chrome instance. Required for fulltext-plus.
• FC_REDIS_URI: Redis connection string. Used for caching to speed up processing and reduce AI costs.
• FC_LLM_API_KEY: API Key for OpenAI or compatible services (DeepSeek, Gemini, etc.).
• FC_LLM_API_MODEL: The default model to use (e.g., gemini-pro, gpt-3.5-turbo). Supports multiple models: You can provide a comma-separated list of models (e.g., gpt-3.5-turbo,gpt-4). FeedCraft will randomly select one for each request and automatically retry with others if a call fails.
• FC_LLM_API_BASE: The API endpoint. Must end with /v1 if using OpenAI compatible APIs.
• FC_LLM_API_TYPE: (Optional) openai (default) or ollama.

External Services

To fully utilize FeedCraft, you should deploy it alongside Redis and Browserless.

version: "3"
services:
  app.feed-craft:
    # ... (see Quick Start)
    environment:
      FC_PUPPETEER_HTTP_ENDPOINT: http://service.browserless:3000
      FC_REDIS_URI: redis://service.redis:6379/
      # ...

  service.redis:
    image: redis:6-alpine
    container_name: feedcraft_redis
    restart: always

  service.browserless:
    image: browserless/chrome
    container_name: feedcraft_browserless
    environment:
      USE_CHROME_STABLE: true
    restart: unless-stopped