// the find
firecrawl/open-scouts
🔥 AI-powered web monitoring platform. Create automated scouts that search the web and send email alerts when they find what you're looking for.
Open Scouts is a self-hostable web monitoring tool that lets you define natural-language queries, runs them on a schedule via Firecrawl + GPT-4, and emails you when it finds something. It's aimed at developers and power users who want to automate the kind of thing they'd otherwise do manually in a browser. The stack is Next.js 15 + Supabase, so it deploys on Vercel in about 20 minutes if you already have accounts everywhere.
The dispatcher architecture is well thought out — pg_cron fires individual HTTP requests per scout rather than running everything in one job, so a slow or failing scout doesn't block the others. Using Supabase Vault to store the service role key and edge function secrets avoids the common mistake of hardcoding credentials in cron SQL. The per-user Firecrawl API key approach in Settings is practical: it sidesteps the shared-quota problem without requiring a payment system. pgvector on execution summaries is a nice touch for finding patterns across scout history, even if it's probably overkill at this scale.
The entire scheduling layer lives inside Supabase (pg_cron + pg_net), which means you can't run this fully locally — the README glosses over this with 'deploy edge functions' but there's no local development story for the cron-driven execution path. OpenAI is hardcoded as the only LLM; there's no abstraction, so swapping models or providers means touching the edge function directly. The README describes a 'partner integration' for Firecrawl key provisioning as 'closed beta enterprise only', but the code for it is still in the repo — that's confusing for anyone trying to understand what's actually available. No deduplication logic is mentioned: if a scout runs hourly and finds the same results, it's unclear whether it suppresses repeat notifications or spams you.