🚄 Railway Central Station AI Helper

A Streamlit app that pulls recent threads from Railway's community forum and uses Claude to draft helpful, engineer-style replies — so you can triage and respond to community questions faster.

What it does

• Fetches the 20 most recent threads from Railway's community GraphQL endpoint
• Lets you filter to just bounty/paid threads with one click
• Shows a preview of each thread's content in an expandable card
• Generates an AI-drafted reply for any thread on demand, written in the voice of a senior Railway engineer
• Caches generated reviews per-session so you don't burn API calls re-generating the same reply
• Lets you pull up a review in a copyable code block

Requirements

• Python 3.9+
• An Anthropic API key with access to Claude

Setup

1. Clone the repo and install dependencies:

   pip install streamlit requests

2. Set your Anthropic API key as an environment variable:

   export ANTHROPIC_KEY="your-api-key-here"

3. Run the app:

   streamlit run railway_helper.py

Usage

1. Click 🔄 Refresh Recent Threads to load the latest threads.
2. Optionally check Show only Bounty threads 💰 to filter to threads mentioning bounties or "$".
3. Expand any thread to read its content.
4. Click 🤖 Get AI Review to generate a suggested reply with Claude.
5. Click 📋 Show for Copy to display the generated reply in a plain-text code block for easy copying.

Notes

• Reviews are cached only for the current session (st.session_state) — refreshing the browser tab clears them.
• Thread data is only refetched when you explicitly click Refresh Recent Threads; otherwise the app reuses what's already loaded, so switching the bounty filter or generating reviews won't reset your view.
• Network calls (to the Railway GraphQL endpoint and the Anthropic API) use explicit timeouts and surface errors in the UI rather than failing silently.

Known limitations

• The "Copy" button doesn't use the system clipboard (Streamlit has no native clipboard API) — it displays the text in a code block for manual copy.
• No pagination beyond the first 20 threads.
• No persistence across sessions or users; this is a single-session helper tool, not a multi-user dashboard.