typesense

name	typesense
description	Instant search for CampaignReceipts. 226 politicians, ~4500 promises, typo-tolerant, faceted, sub-10ms. Self-hosted on Render. Use when adding search features, rebuilding indexes, or tuning search relevance.

Typesense (Instant Search)

Status: Deploy ready

Why Typesense

CampaignReceipts' viral mechanic: "who promised to lower insulin prices?" Needs instant, typo-tolerant search across politicians and promises. Supabase full-text search works but Typesense gives faceted filtering and sub-10ms results.

Collections

politicians

Fields: name, party, state, branch, ideology_label, scorecard_percentage_kept, professional_background, profile_narrative

promises

Fields: promise_text, verdict, category, politician_name, politician_slug, case_study_narrative

Deploy Checklist

1. Create Render Docker service: image typesense/typesense:27.1, starter plan with disk (~$7/mo)
2. Set env vars in root .env:
   • TYPESENSE_API_KEY (generate: openssl rand -hex 32)
   • TYPESENSE_HOST (Render service URL without https://)
   • TYPESENSE_PROTOCOL=https
3. Run indexing: node companies/campaign-receipts/scripts/typesense-index.mjs
4. Set up nightly re-index cron on Render (02:00 UTC)

Files in campaign-receipts

• scripts/typesense-index.mjs — reads from Supabase, creates/updates collections
• app/api/search/route.ts — proxy to Typesense multi-search
• app/components/SearchBar.tsx — instant search UI in header

Search API

# Multi-search across both collections
curl -X POST "https://TYPESENSE_HOST/multi_search" \
  -H "X-TYPESENSE-API-KEY: KEY" \
  -d '{"searches":[
    {"collection":"politicians","q":"healthcare","query_by":"name,profile_narrative"},
    {"collection":"promises","q":"healthcare","query_by":"promise_text,category"}
  ]}'

Cost

~$7/mo on Render starter plan with persistent disk