DB9 unifies Postgres, file storage for AI agents

The "one workspace per agent run" model is genuinely compelling — eliminating the glue code between Postgres, object storage, and embedding pipelines is exactly the friction that slows agent development.

• –The `embedding()` in-SQL function is the standout feature: call it in a query and get vectors back without managing an external embedding service or extra API keys in app code
• –Built-in `fs9` file system extension means transcripts, session snapshots, and binary artifacts live alongside relational state — a natural fit for long-running agents that accumulate context
• –Environment branching that includes files and cron jobs (not just schema) is unusual and useful for agent testing
• –Caveat worth noting: DB9 is a PostgreSQL-compatible layer over TiKV, not actual Postgres — the homepage obscures this, which drew HN criticism; workloads with complex Postgres internals may hit compatibility limits
• –Zero-config CLI (`curl … | sh`, then `db9 create`) and anonymous account creation lower the barrier to trial significantly