ModMail AutoPilot

Inspiration

Reddit moderators deal with a constant stream of ModMail: ban appeals, user reports, spam, and routine questions. Reading, sorting, and replying to each message takes time away from the moderation work that actually needs human judgment. ModMail AutoPilot helps moderators triage messages faster while keeping humans in control of every reply.

What it does

ModMail AutoPilot analyzes incoming ModMail with AI, then presents the results in a clean moderator dashboard where mods can:

• View ModMail with AI-generated summaries, priority levels, and categories
• Review suggested replies before sending
• Edit and send a suggested reply with one click
• Dismiss messages that do not need a response
• Inject sample conversations for demo and testing

How we built it

Built on Reddit's Devvit platform using the v0.13 web app architecture:

• Frontend: React + TypeScript + Tailwind CSS dashboard rendered inside an inline custom post
• Backend: Hono server routes running on Devvit's serverless runtime
• Storage: Redis for conversation data persistence
• AI: Claude Haiku through Lemon API for ModMail categorization, prioritization, summarization, and reply drafting
• Integration: Reddit ModMail API for fetching conversations and sending moderator-approved replies
• Configuration: The LLM API key is stored as a Devvit global secret setting named llm-api-key; it is not hard-coded in source code

Challenges we ran into

Migrating from the deprecated Blocks model to the newer Devvit web app architecture required reworking the app around React, Vite, Hono, Redis, server routes, menu endpoints, and trigger endpoints.

We also had to separate demo behavior from real moderation behavior. Test conversations use synthetic test_... IDs and do not call the live Reddit ModMail reply API, while real ModMail conversations are sent through Reddit's ModMail API after moderator approval.

Accomplishments that we're proud of

• Built a working Devvit web app with a React dashboard and Hono backend
• Added real onModMail trigger processing for incoming ModMail
• Integrated AI analysis that returns structured category, priority, summary, and suggested reply data
• Implemented one-click send and dismiss actions
• Stored the LLM credential securely through Devvit settings instead of source code

What we learned

• How Devvit v0.13 web apps structure client code, server routes, menu endpoints, and triggers
• How to render a React dashboard inside a Reddit custom post
• How to persist app data with Redis
• How to work with Reddit ModMail APIs from a Devvit app
• Why production apps need explicit error handling for LLM failures, missing settings, Redis data issues, and test-vs-real action paths

What's next for ModMail AutoPilot

• Expand real-world testing across additional moderated subreddits
• Add multi-subreddit dashboard support
• Add customizable AI reply templates
• Add bulk actions for similar ModMail conversations
• Add richer audit trails for sent and dismissed conversations

Built With

• devvit
• hono
• react
• redis
• tailwind-css
• typescript