Skip to main content

Convex Backend Bootstrap

Status note​

This doc captures the initial Convex backend slice landed for #54, plus the first profile foundation slices from #9 through #13.

It is intentionally narrow: enough structure to run Convex locally, generate typed backend helpers, prove the backend is wired, and define profile identity/authority contracts without prematurely locking auth providers, account links, public write flows, billing, or rich presentation features.

Locked decision​

  • the first backend surface lives in the repo-root convex/ directory
  • the initial bootstrap should stay schema-light and friendly to anonymous local development
  • generated Convex types under convex/_generated/ should be committed

Current implementation​

  • convex/schema.ts defines the first durable profiles table for people and communities
  • convex/health.ts exposes a minimal public query, health:status, that confirms the backend is reachable without hard-coding early product domain records
  • convex.json pins Convex to Node 22 so local backend runtime expectations stay aligned with the repo's current Node baseline
  • convex/tsconfig.json provides the TypeScript settings Convex uses to typecheck backend source files
  • the web app consumes health:status as the first real Next.js -> Convex runtime path, mounting a client-side provider baseline and surfacing the result in the homepage without inventing early product schema

Local workflow​

  1. Install dependencies with pnpm install.
  2. Bootstrap an anonymous local deployment and verify the health query with pnpm bootstrap:backend:local.
  3. Keep pnpm dev:backend:local running while editing backend files.
  4. Run pnpm run:backend:health:local from a second terminal when you want the same one-shot placeholder health check without using the verify script name directly.
  5. Run pnpm dev:web if you want to confirm the homepage can render the live health:status response through the web app.
  6. Run pnpm typecheck:backend before shipping backend edits.
  7. Run pnpm check:backend:generated before pushing schema or function changes that may affect convex/_generated/.

If you want the full repo gate, use pnpm verify. If you only need the web app checks, keep using pnpm verify:web.

Notes:

  • Convex writes deployment configuration to the repo-root .env.local file.
  • The local Convex wrapper mirrors the repo-root CONVEX_URL into apps/web/.env.local as NEXT_PUBLIC_CONVEX_URL so the web app can follow the normal client-side Convex + Next.js convention without leaking a non-public variable through server props.
  • Anonymous local backend state for this repo is kept under .convex-home/ and .convex-tmp/ so the bootstrap does not collide with other Convex projects on the same machine.
  • Production Convex deploys run from the baseline GitHub Actions workflow when CONVEX_DEPLOY_KEY is configured; otherwise the deploy job records a skip summary and exits cleanly.
  • Deploy-time data backfills are declared with @convex-dev/migrations and run through migrations:runAll after production function deploys.
  • Committed files in convex/_generated/ are treated as checked-in build artifacts and should remain diff-free after pnpm check:backend:generated.

Structure rule​

Keep the initial backend slice simple:

  • add new schema and functions under convex/
  • avoid introducing product tables until the relevant issue defines the domain slice
  • prefer small, typed functions over placeholder complexity

Follow-on issues​

  • #55 wires the web app to the first Convex client/runtime path using health:status
  • profile schema, auth, billing, and production deployment posture should land in their own issues instead of bloating the bootstrap
  • #9 adds the first product table, profiles
  • #10 through #13 add profile slugs, type-aware fields, permission contracts, and claim-state helpers while keeping auth/account links deferred

App Router baseline​

The current rule is intentionally narrow:

  • use useQuery inside client components when the UI should stay reactive after first render
  • use fetchQuery inside server components when the page only needs a server-side read
  • defer preloadQuery until a real feature needs server-rendered first paint plus a hydrated reactive handoff

This keeps the first App Router pattern copyable without introducing multiple competing baselines before auth and domain work arrive.