PocketBase, the Empire Backend I Run Across Every Project

PocketBase is the empire's chosen backend, and I run it across taxwallaai, papersadda, autokaam, kaam-tracker, and PramaanAI. Five live deployments. Zero regrets.

The reason is structural. PocketBase is one Go binary that bundles SQLite, auth, file storage, realtime subscriptions, an admin UI, and a JS-extensible API into a single artifact. It runs on a Rs 1,200 a month Oracle ARM box without breaking a sweat. Backups are a single SQLite file. Migrations are JSON-driven through the admin UI or programmatic. The whole stack is one process.

What I deliberately do not run: Supabase, Firebase, Vercel KV, Railway databases, or any other vendor-managed BaaS. Empire ops bans them, and that ban is a deliberate operating choice, not a budget call. Vendor BaaS pricing scales sharply past hobby usage, the data is in someone else's hands, and the migration path off is brutal. PocketBase keeps the data, the schema, and the runtime under my control.

The trade-off is real. PocketBase does not give me horizontal scaling. SQLite-on-one-box has a real upper limit; for the empire's scale (under 100 concurrent users on the largest deployment) it is many multiples below the ceiling, but a SaaS that hits 10K concurrent users will outgrow it. My answer when that day comes: shard by tenant onto separate PocketBase instances. The tools to do this gracefully exist; I have not had to.

The sharpest gotcha I hit was the realtime subscription performance under concurrent admin writes. Early kaam-tracker had three admin users writing the same collection at once and the realtime fan-out lagged. The fix was to disable realtime on that collection for admin users (they got websocket-pinged manually) and keep realtime on for read-only client views. Fine.

The Indian-operator angle is the per-month cost. Supabase Pro is roughly 2,000 rupees a month for the entry tier, plus database egress, plus storage. PocketBase on Coolify on Oracle ARM is my hosting cost (1,200 a month) and supports five empire apps simultaneously. The math is not subtle.

For 2026, the alternative I am watching is Convex, which has a cleaner reactive model but is fully cloud-managed. I will not switch unless the empire develops a workload that needs server-side TypeScript reactivity, and so far it has not.

Pick PocketBase, run it on Coolify, sleep well. This is the empire pattern.

Vendor BaaS pricing has grown sharper through 2025-26 (Supabase added per-MAU pricing, Firebase priced more aggressively). The self-host option used to require a stack of services; PocketBase reduces it to one binary. For solo and small-team builders running up to a few thousand concurrent users, this is the path with the cleanest cost-to-control ratio.