Post mvp factory extensions bots agents

Milestone: Factory grows beyond push-button workflows: triggered and recurring automations, inbox management, IoT integrations. External agents interact with your data via the public API. Parrot (voice dictation) also lives here. Learning payoff: Agent interoperability, webhook-driven automation, workflow design patterns.

Recommended Learning Order

If you're new to some of the tech in this stack, here's the minimum viable reading before each phase:

Architectural Invariants to Review After Every Phase

Before moving to the next phase, verify:

• Every new API route uses withApiGuard()
• No packages/* imports from apps/*
• Every new user-owned table has RLS enabled with the canonical combined policy (user isolation + soft-delete in one USING clause)
• Every new syncable table has no_hard_delete_[table] and no_update_deleted_[table] triggers applied
• Every new table with user content uses withRLS() via the guard, not inline
• All mutations go through the repository layer
• Syncable tables include createdAt, updatedAt, deletedAt
• Soft deletes only — no hard deletes on user data
• All queries on syncable tables filter where(isNull(table.deletedAt))
• Content tables in the embedding pipeline have embeddingStatus
• DATABASE_URL connects as app_runtime (not superuser) — verify if changing connection config

Rough Effort Estimates (Solo, AI-assisted, Learning pace)

Sessions are loosely defined as focused 2–3 hour working blocks. Estimates assume you're reviewing every line and asking questions — that's the point.