Assess
Reason for selection
Convex promises to collapse our current "Supabase + Drizzle + tRPC + TanStack Query" pyramid into a single code-first backend. Queries are TypeScript, live by default, and run within the datastore; local dev is a one-liner with no Docker orchestration. A self-host option on Fly/Docker eases vendor-lock anxiety. If the claims hold up, Convex could greatly simplify new-project bootstrapping and reduce full-stack cognitive load.
Key questions
- How reliable and scalable is the reactive query engine under real-world load?
- What is the DX of modelling relational data and auth without SQL?
- Total cost of ownership vs Supabase + edge functions once traffic scales?
- Can we layer fine-grained RLS / multi-tenant rules cleanly?
- Migration path back to PostgreSQL if we outgrow Convex?
- How painful is self-hosting and upgrades?
Assessment approach
- Spin up a demo in the mo-stack: Auth (Clerk), CRUD screens, optimistic updates, file uploads.
- Benchmark latency and subscription fan-out with a Locust script.
- Compare authoring effort and lines of code with an equivalent Supabase-Drizzle prototype.
- Test self-hosting on Fly.io and Docker compose; document pitfalls.
- Hold a dev-experience retro and decide whether to promote, iterate or drop.
Potential testers
- Cam (lead), two full-stack volunteers, one mobile engineer for React Native SDK.
Notes
Convex's new native auth is still beta and lacks advanced rules; we will pair with Clerk/Auth0. Long-term viability should be tracked.