What Supabase actually is
Supabase wraps PostgreSQL with a REST API, authentication, realtime subscriptions, edge functions, and storage. The architecture: every Supabase project is a real PostgreSQL database (not a NoSQL imitation), exposed through PostgREST for REST, GoTrue for auth, Realtime for WebSocket subscriptions, and Edge Functions (Deno-based) for server-side logic. When you create a table posts(id uuid, title text, body text), Supabase instantly exposes GET /posts, POST /posts, PATCH /posts, DELETE /posts at your-project.supabase.co/rest/v1/posts. The dashboard shows the SQL underneath, so you never get locked in. For saas.pet I have a Supabase project running 5 tables, 3 edge functions, and pgvector for review search. The project has been live for 6 months with zero database issues. For most developers building web apps, Supabase is the fastest way from idea to production.
PostgreSQL is the foundation that matters
Firebase uses Firestore, a proprietary document store. Supabase uses PostgreSQL, the most popular open-source database in the world. The difference is not subtle. With Supabase you get real SQL: SELECT posts.title, COUNT(*) FROM posts JOIN authors ON posts.author_id = authors.id WHERE authors.country = 'US' GROUP BY posts.title ORDER BY 2 DESC LIMIT 10. You get real joins, real transactions, real ACID guarantees. You can install extensions like pgvector for vector search, PostGIS for geospatial, pg_trgm for fuzzy text match, and pg_cron for scheduled jobs. For the saas.pet review database, I use pgvector to find similar AI tools by embedding, pg_trgm to handle fuzzy category searches ('coding-tool' matches 'coding', 'code', 'tool'), and standard SQL for everything else. The 6 month uptime for my project is 99.98%. Backups are automatic (point-in-time recovery on paid tier). For developers who want a real database, not a JSON blob store, Supabase is the right choice. For developers who want simplicity over SQL power, Firebase is still easier on day 1. By month 3, Supabase wins.
Auth that does not make you cry
Supabase Auth handles email/password, magic links, OAuth (Google, GitHub, GitLab, Apple, Discord, Slack, Spotify, etc.), phone/SMS, and anonymous sign-in. JWT tokens are issued automatically and refresh handled in the client SDK. Row Level Security (RLS) is the killer feature: write a SQL policy like CREATE POLICY 'user_read_own' ON reviews FOR SELECT USING (auth.uid() = user_id) and the database enforces it for every API call. For saas.pet, every user can only read their own review drafts. The auth flow takes 20 minutes to set up: copy the Supabase client into your frontend, call supabase.auth.signInWithOAuth({ provider: 'github' }), done. For 6 months across 5 projects, auth has never broken. The Google OAuth provider alone covers 80% of signups. For developers who have ever struggled with NextAuth, Clerk, or Auth0, Supabase Auth is the first auth that does not require a config file the length of a novel.
Realtime without writing WebSocket code
Supabase Realtime listens to PostgreSQL logical replication and pushes row changes to subscribed clients. Subscribe to a channel and you get inserts, updates, deletes as JavaScript events. The latency is under 100ms for same-region. For the saas.pet admin dashboard, I subscribe to INSERT events on a scrape_jobs table and the dashboard updates in real time without polling. For a chat feature, subscribe to messages.channel and write a 30 line frontend hook. Compared to building WebSocket servers yourself (which I have done 3 times in my career, and each time regretted the 2 weeks), Supabase Realtime is a 5 minute setup. The downside: no fine-grained filtering server-side (you get all events, you filter client-side). For most use cases this is fine. For high-throughput chat (millions of messages per second), you still need a custom solution.
Edge functions for the small but important backend logic
Supabase Edge Functions are Deno-based TypeScript functions deployed globally. Each function is a small HTTP endpoint. For the saas.pet Cron Health check, I have an edge function that queries the latest scrape status and returns JSON. The function deploys with supabase functions deploy and runs at every continent. Cold start is around 500ms. For 5 projects I use edge functions for: Stripe webhooks, Resend email sends, custom AI model proxying, image resizing, and rate limiting. Each function is 50-200 lines of TypeScript. The total cost is included in the Supabase plan up to 1 million invocations per month on Pro, more on Enterprise. For developers who would otherwise write an Express server, deploy to Railway or Fly.io, set up a CI/CD pipeline, and configure environment variables, edge functions are a 10x productivity boost. For complex backend logic (heavy computation, long-running jobs), you still need a traditional server.
pgvector: the AI feature that built saas.pet
pgvector is a PostgreSQL extension for vector similarity search. Supabase bundles it on every project. Enable it in the dashboard, create a table reviews(id uuid primary key, embedding vector(1536)), and you can call SELECT * FROM reviews ORDER BY embedding <=> '[0.1, 0.2, ...]' LIMIT 10 to find the 10 most similar reviews. For saas.pet I generate embeddings for each review body using OpenAI text-embedding-3-small, store them in pgvector, and expose a /api/reviews/similar endpoint that returns matches in under 50ms for 50K rows. Before pgvector, building semantic search required Pinecone ($0.096/million vectors) or Weaviate (self-hosted complexity). With pgvector in Supabase, semantic search is one SQL query away. For 6 months I have been running semantic review search and it has never been down. For developers building any AI feature (RAG, recommendations, semantic search, deduplication), Supabase plus pgvector is the fastest setup in 2026.
Storage for files, images, and videos
Supabase Storage is S3-compatible object storage with a JavaScript SDK. Upload a file with supabase.storage.from('avatars').upload('user123.png', file), get a public URL or signed URL. The free tier gives 1GB, the Pro tier gives 100GB, and you pay $0.021/GB beyond. RLS policies work on storage (so you can restrict user uploads to a bucket). For saas.pet I store scraped tool logos in a public bucket and use the signed URL feature for user-uploaded review screenshots. For 6 months, storage has been reliable. The main limitation: no built-in image resizing. You need to pair with an edge function or use an external service like imgproxy. For developers building any app with file uploads, Supabase Storage is the easiest option. For heavy media apps (video processing), you still need S3 or Cloudflare R2.
The free tier gotcha nobody warns you about
The Supabase free tier includes 500MB database, 1GB storage, 2GB bandwidth, 50K auth users, 5GB egress. This is fine for solo projects and small side apps. The gotcha: a single Supabase project pauses after 7 days of inactivity on the free tier (you must hit an API call or open the dashboard). For saas.pet I learned this the hard way when the admin dashboard went dark for 3 days and the user-facing site started returning 500s. The fix: add a small cron job (a Vercel cron hitting a Supabase edge function once a day) to keep the project active. The Pro tier ($25/month) does not pause. For developers running production apps, the Pro tier is mandatory. For learning and prototyping, the free tier is great if you accept the 7 day pause.
Pricing compared to Firebase and Pocketbase
Supabase Pro: $25/month includes 8GB database, 100GB storage, 250GB bandwidth, 100K MAU. Firebase Blaze: pay-as-you-go, roughly $30-100/month for similar scale. Pocketbase: free, self-hosted, you run the server yourself. The decision matrix: if you want managed zero-ops, Supabase Pro or Firebase Blaze. If you want cheapest, Pocketbase but you maintain a server. For saas.pet I pay $25/month for Supabase Pro and have never had a billing surprise. The bandwidth pricing is predictable (no surprise bills like AWS). For developers choosing a backend, the math usually works out to Supabase being 30-50% cheaper than Firebase at the same scale, with the bonus of real PostgreSQL. For hobbyists, the free tier plus manual activity is enough. For solo founders shipping SaaS, Supabase Pro is the right starting point.
When NOT to use Supabase
Supabase is not the right choice for: (1) Extreme write throughput (millions of writes per second) - use Cassandra or DynamoDB instead. (2) Heavy Microsoft Azure integration - use Cosmos DB instead. (3) Real-time analytics on large datasets - use ClickHouse or BigQuery. (4) Self-hosted, air-gapped deployments with no cloud access - Supabase has a self-hosted option but it requires maintenance. (5) If you absolutely need Google Cloud Storage, Firebase Auth deep integration - Firebase still wins. (6) If you want a fully managed data warehouse with no SQL - use Airtable or Notion. For 95% of web apps, mobile apps, indie SaaS tools, internal tools, AI RAG apps, and prototypes, Supabase is the right backend. The developer experience is faster than Firebase. The data model is more powerful than Firebase. The pricing is more predictable than AWS. Supabase is what Firebase would be if Firebase was built on PostgreSQL.
The 6-month honest verdict
After 6 months and 5 projects, here is the honest verdict. Pros: PostgreSQL foundation, pgvector for AI, RLS for security, edge functions for backend logic, realtime without WebSocket code, free tier for prototypes, Pro tier for production. Cons: free tier pauses, edge function cold start is slow, no built-in image processing, dashboard is functional but not beautiful, supabase-js client SDK is good but not perfect. For saas.pet, Supabase handles review data, user auth, semantic search, and admin tasks. Zero data loss. One 10-minute downtime incident (caused by me, not Supabase). For developers choosing a backend in 2026, Supabase is the first choice to evaluate. The combination of PostgreSQL plus pgvector plus edge functions plus auth is unmatched at the price. For 6 months and counting, Supabase has earned a permanent slot in my stack.
