Branches are out of beta and generally available.

If you missed the earlier changelogs, branches give you isolated cloud environments to develop, test, and iterate on your data infrastructure without touching production. And without needing Docker to run Tinybird locally. What's new in GA:

Connector support: Kafka, S3, and GCS connectors now work in branches. Kafka connections start stopped by default with their own consumer group. S3 and GCS connections support sample imports so you can validate schemas before deploying.

Preview deployments from CI: npx tinybird preview creates an ephemeral branch per pull request. It auto-detects tokens in Vercel, GitHub Actions, and GitLab CI, so you can test Endpoints against real data in your CI pipeline.

Agentic workflows: Branches are sandboxes for coding agents. For example, Claude Code can run tb branch create, define Data Sources and Pipes, ingest test data, validate query results with tb dev before deploying to production. All in Tinybird Cloud, without running Tinybird locally and without touching your live environment.

Zero-copy partition sharing: The --last-partition flag brings production data into branches without duplicating storage.

Read the full announcement: Branches are GA: data infrastructure for agents.

Requires upgrading to tinybird 3.5.3. Upgrade with tb update.

Kafka connectors now support the SASL_PLAINTEXT security protocol. Previously, selecting SASL_PLAINTEXT would silently drop SASL credentials and fail with a misleading GSSAPI error. This affected connections to Kafka brokers that use SASL authentication without TLS, like WarpStream.

The TypeScript SDK now supports generic type parameters for all column types, contributed by Wyatt Schulte. When you define a Data Source schema with t.string(), t.int32(), t.dateTime(), etc., TypeScript infers the exact row type at compile time via InferRow<>. You can also pass branded or opaque types like t.string<UserId>() for stronger type safety.

Connections are now included in deploy requests, contributed by Marcos Leal. Previously, tinybird deploy would skip connections, so S3, GCS, or Kafka connectors weren't deployed to the main Workspace. The deploy flow also handles null feedback resources without crashing.

• [All]: DELETE operations with nondeterministic conditions (subqueries, now()) now return a clear error message explaining the issue, instead of a cryptic ClickHouse internal error about allow_nondeterministic_mutations.
• [All]: Fixed a bug where Data Source resource limits could be lost when retry operations were performed on failed ingestion batches.
• [Forward]: Fixed Organization names with special characters being truncated too aggressively in the page title.
• [Forward]: Fixed the cluster size selection table where the vCPU time column wasn't displaying correctly.

We built our own AI coding agent. Here's why we're sunsetting it: Tinybird Code is being deprecated. General-purpose coding agents have matured fast enough that custom agents are no longer the right approach. The domain knowledge lives in portable agent skills now.

How Order Editing replaced a failing DynamoDB pipeline with Tinybird: A 3-person team processing $5B+ in Shopify GMV replaced their DynamoDB analytics pipeline with Tinybird. 3M events/day at ~$300/month.

How we built branches: sharing production data without copying it: A technical deep-dive into zero-copy partition sharing, Kafka consumer group isolation, and schema migrations behind Tinybird branches.