Object storage gets branch-aware semantics

Neon branches everything — including your objects

The biggest news this week is Neon's branch-aware Object Storage, now in beta for all users. The headline feature is that buckets and objects inherit and fork alongside Postgres branches — copy-on-write isolation means a branch's objects diverge from the parent the moment you write to them, and lineage-based versioning gives you full history per branch. You access everything through standard S3 tools using Neon credentials, and the neon.ts SDK can provision buckets alongside your database. For teams that already use Neon's branching for schema migrations and ephemeral environments, this closes a real gap: test data, file uploads, and assets can now follow the same lifecycle as the database. Standing up a preview branch with its own isolated object store is no longer a multi-provider chore.

Neon also shipped a reworked Console layout that puts the Postgres database, Auth, Object storage, Functions, and AI Gateway at the same level under each branch. The AI Gateway expanded its model lineup with Kimi K3, GLM-5.2, and Inkling alongside new Gemini and GPT options. On permissions, per-project roles are now generally available with org-level Admin, Editor, Viewer, and Collaborator roles, plus project-level grants — the old collaboration model is deprecated. The CLI gains api-keys and profile commands for scoped credentials and multi-account work. A separate blog post argues that Castform's post-trained 4B model matches frontier search accuracy at 100x lower cost, running against Lakebase Search on Neon.

Supabase: pipelines, AI sign-in, and a cleaner JavaScript client

Supabase's monthly roundup bundled several notable items: Pipelines CDC now streams Postgres changes to BigQuery in near real time (public alpha on all paid plans), sign in with ChatGPT is in beta, and Unified Logs reached open beta with Grafana Cloud observability available on every plan including Free. The company also deprecated extension version pinning — explicit versions in CREATE/ALTER EXTENSION are now ignored — and locked the realtime schema against modification. That last bit matters if you've been relying on pinned extensions for reproducibility.

For developers building on realtime, the Postgres Changes update is the more immediately useful piece. You can now compose AND filters across columns, use operators like like, ilike, is, match, imatch, and isdistinct, and opt into trimming payload columns to cut bandwidth — the column selection requires @supabase/supabase-js 2.109.0+. The JavaScript SDK also moved OpenTelemetry tracing to an opt-in /tracing subpath, so the default import stays lean. A few fixes round it out: setAuth no longer disables realtime token refresh, duplicate realtime bindings are prevented, and validateUUID now accepts uppercase UUIDs. Supabase also became a connector on Perplexity Computer, letting the agent read and write Postgres tables and carry state across runs — useful for debugging from support tickets or scheduling recurring queries.

On the CLI front, Supabase continued its native TypeScript migration: v2.112.0 deprecates db diff --use-pg-schema and keeps db pull --experimental delegated to Go rather than retiring it, while v2.113.0 ports shell completion and db reset local recreate to native TS. A steady stream of beta releases carried the same work: bootstrapping db push, db test, and container bootstrap all moved to TypeScript, and there's a new option to skip vault sync. It's churn, but the direction is toward a single maintainable implementation.

PlanetScale doubles down on Vitess operations

PlanetScale's week was about operational control. Parallel deploy requests let two schema changes run concurrently as long as they touch different tables — slow migrations no longer block the queue, and you get --strategy parallel in the CLI. The VTGate management commands for show, resize, and status cover size, count, and autoscaling with a CPU utilization target. And read-only region support is here: list regions, create passwords scoped to a specific read-only region, and dump from a chosen region instead of the primary.

The CLI releases behind these features also landed, along with a few smaller quality-of-life fixes like dedicated replica PgBouncers supporting zone affinity after creation and preserved raw API error codes in JSON output. Taken together, this is a clear statement that PlanetScale wants schema and infrastructure management to live comfortably in your terminal.