Mod AI Tool

AI-powered moderation assistant that helps Reddit moderators triage content faster.

Comment

Inspiration

Moderating large Reddit communities is exhausting. When a spam wave hits or a thread goes toxic, mods have to manually remove comments one by one while bad content keeps spreading. We wanted to give moderators superpowers — instant bulk cleanup and an AI second opinion, right inside Reddit's native UI.

What it does

Mod AI Tool is a Devvit app that gives moderators two powerful tools accessible directly from any post or comment:

  • Mop: Bulk-remove or lock an entire comment thread (from a selected comment down) or every comment on a post — with options to skip distinguished moderator/admin posts.
  • Analyze with AI: Sends the post or comment to Claude and returns an instant verdict — REMOVE, APPROVE, or REVIEW — along with a confidence score and a one-line reason. Helps mods make faster, more consistent decisions.

How we built it

  • Devvit — Reddit's native app platform, used for context menu items and forms that live inside Reddit's UI
  • Hono — Lightweight TypeScript web framework for routing internal menu, form, and API handlers
  • Claude Haiku (Anthropic API) — Fast, low-latency AI model for content analysis, returning structured VERDICT / CONFIDENCE / REASON output
  • Vite + TypeScript — Build tooling and type safety throughout

Challenges we ran into

  • Structured AI output in a sandboxed environment: Getting Claude to return machine-parseable responses reliably required careful prompt engineering with a strict format and regex parsing.
  • Async comment tree traversal: Reddit's nested comment structure required recursive async generators to traverse arbitrarily deep threads without loading everything into memory at once.
  • Devvit's constraints: The platform's sandboxed runtime and form system have specific limitations around how data flows from menu items to form submissions to backend handlers.

Accomplishments that we're proud of

  • A genuinely useful AI moderation assistant that runs entirely inside Reddit with no external dashboard needed.
  • Bulk comment operations using Promise.all that handle entire post comment sections in parallel — fast enough to be practical on active threads.
  • A clean, permission-checked flow that ensures only moderators with the right permissions can trigger actions.

What we learned

  • How to build on Devvit's web model end-to-end — from devvit.json configuration to Hono routes to Reddit API calls.
  • Prompt design for structured AI output: giving Claude a strict format example in the prompt is far more reliable than asking it to reason freely.

What's next for Mod AI Tool

  • Subreddit-specific rules: Fetch the actual rules of the subreddit and include them in the AI prompt for more accurate, context-aware verdicts.
  • Auto-mod triggers: Run AI analysis automatically on new posts/comments and flag borderline content before a human even sees it.
  • Mod action log: Track what was mopped or AI-flagged so mods can review the history and undo mistakes.
  • Batch analysis: Let mods select multiple comments and get AI verdicts on all of them at once.

Built With

Share this project:

Updates