Agents get their own databases as sharded Postgres enters preview

Databases that agents can provision themselves

The week's most consequential shift is about who creates the database. Claimable Neon lets an agent provision a temporary project with scoped credentials — no account, no payment details, no email verification loop — and a human claims ownership later with neon claim accept. Lakebase Postgres, the Data API, and Managed Better Auth are supported from day one. Alongside it, Neon introduced @neon/tools, a package that exposes typed function-calling tools from the @neon/sdk with adapters for MCP, Mastra, and Eve, adding more than 70 tools to the hosted Neon MCP server. The same announcement raised shared buffers on large fixed-size computes of 18 CU and above, letting them use up to 75% of available memory — a quiet but meaningful bump for memory-hungry workloads.

Sharded Postgres arrives in preview

PlanetScale opened Neki, its horizontally sharded Postgres, to platform preview. The pitch is deliberately conservative: every shard remains real Postgres, while Neki handles routing, resharding, online schema changes, and version upgrades behind a single connection string. Organization admins opt in from the dashboard or Settings → Platform Preview, after which Neki appears as a database engine at creation time. The PlanetScale CLI picked up Neki support in the same week, including fixes to restore sizing and cluster size. Less glamorous but relevant to regulated teams: PlanetScale Managed was re-certified for PCI DSS v4.0.1 as a Level 1 Service Provider, with updated attestation documents in the trust center. The CLI also added disk autoscaling settings to keyspace commands and fixed D1 CHECK and GENERATED expression rewrites for SQLite-only functions.

Supabase's CLI week: pull, diff, and stack lifecycle

Supabase's CLI shipped a dense batch of work best read as one story about making local and remote state easier to reconcile. The headline additions are config diff and config pull, plus a full workers management suite covering new, push, list, status, and delete, with a SUPABASE_USE_SLIM_IMAGES flag and shadow baseline caching now on by default. A supabase pull orchestrator command followed in beta, and the config package resolved stack environment overrides and encrypted secrets that were previously misread. Local stack management got the most churn: an experimental stack start and stack stop, a feature flag for selecting which stack to start and stop, and a rewritten managed local stack runtime that now consumes --log-level correctly.

Security and correctness fixes ran in parallel: content_path is now confined to the project root for every consumer, closing a path-traversal class of bug, and an explicit --workdir no longer climbs to a parent project. Declarative sync now honors deletion of extension-managed objects rather than silently ignoring them. Elsewhere in the stack, Supabase Auth added MFA recovery codes with full generate/verify/regenerate/delete endpoints, and the JS client SDK surfaced that recovery-code API alongside bucket lifecycle configuration, topk up to 10k, and versionId on create-URL methods.

Prisma 8 rounds into shape

Prisma continued iterating on its next major, finishing the rename from Prisma Next to Prisma 8 across every visible identifier — the old schema header still works, but the old NEXT_-infixed environment variables do not — and adding named model and result types to the emitted contract. The latest RC moved the toolchain onto a new CLI engine that gives every CLI command a Markdown output format.