LumenClip DocsRailway runtime
LumenClip runs on Railway PostgreSQL and a private S3-compatible bucket. Clerk owns browser identity; the application owns domain records under stable owner IDs.
Runtime topology
| Resource | Responsibility |
|---|---|
| PostgreSQL | Domain records, output media, jobs, users, and operational metadata |
| Private bucket | Uploaded media, generated slides, videos, thumbnails, and demos |
| Web | Next.js application and MCP HTTP surface |
| Worker | Durable job execution for generation, analytics, and publishing work |
| Scheduler | Computes due automation slots and enqueues idempotent jobs |
Storage model
The domain_records table is the compatibility boundary for logical JSON
stores. Its table_name, source_key, owner, RID, status, ordering, and JSONB
payload preserve the original record identity while allowing indexed reads.
Output media is normalized into output_media.
Private objects are addressed by deterministic bucket and file IDs derived from the data-relative path. Browser clients never receive bucket credentials; application routes stream authorized media and signed downloads remain short-lived.
Operations
pnpm railway:db:migrateApply this command on deploy and before starting web, worker, or scheduler.
Database migrations are checked in under infra/railway/migrations.
Never print database URLs, bucket credentials, Clerk secrets, or provider keys. Production changes must be verified across web, worker, scheduler, PostgreSQL, and bucket-dependent paths.
Acceptance checks
- A generation creates its run/output records and normalized media rows.
- Generated media streams through application routes and public share URLs.
- Publishing creates canonical post records and publication projections.
- Analytics snapshots attach to the correct post and integration.
- Scheduled jobs are deduped, leased once, retried safely, and dead-lettered.
- MCP can read the same persisted artifacts with the caller's owner scope.