Swamp, an AI agent workflow tool, defaults to local SQLite storage which fragments agent state across machines. The swamp datastore primitive solves this by providing a pluggable backend interface that supports remote shared storage. Two implementations are covered: a GitLab backend that repurposes GitLab's Terraform state HTTP API (requiring no extra infrastructure), and a PostgreSQL backend targeting AWS RDS/Aurora with row-based distributed locking and heartbeat TTL for crash recovery. The adoption path is incremental — start with local SQLite, add a datastore config line when needed, and scale to team-wide shared state without refactoring existing models or workflows.
Table of contents
What a datastore is #Two backends, one interface #The pattern for a single developer #From one machine to a team #Why this matters #The opt-in adoption curve #5 Impressions