Tinybird Release Notes

Today is the last day of Tinybird Classic on Free and Developer plans. We announced the sunset in July, ran the automatic migration to Forward in September, and gave Workspaces that couldn't be migrated automatically until today to move.

Starting today

• Remaining Classic Workspaces on the Free plan are deleted.
• Classic Workspaces on the Developer plan are no longer supported, and access through the UI is blocked. If you still have one, contact [email protected] and we'll help you migrate it.
• Classic Workspaces on Enterprise plans aren't affected.

If your Workspace was migrated, it's already running on Forward. The guide for after migrating from Classic covers what changed and how to keep working with it. If you're on Enterprise and want to move, the migration guide walks through the process, including the tb migrate-to-forward command.

Forward is where all new Tinybird work happens: local development, branching, CI/CD, SDKs, and agent-friendly workflows. Thanks to everyone who migrated. We also shared the news on X.

Breaking changes

• Tinybird CLI 5.0.0 for Forward requires Python 3.11 or later: Python 3.10 is end of life and is no longer supported. Upgrade to Python 3.11, 3.12, or 3.13 before you update the CLI. The CLI Docker images now use Ubuntu 24.04 with Python 3.12.
• Tinybird Local on arm64 requires a 4 KiB kernel page size: Tinybird Local now ships ClickHouse® 25.8.40.4. The arm64 image no longer starts on hosts with 16 KiB or 64 KiB pages. Use the tinybirdco/tinybird-local:legacy-64kb-pages image on those hosts. It won't receive updates.

Improvements and bug fixes

• [All]: The Tinybird MCP server's execute_query and call_endpoint tools now return up to 1,000 rows by default, or up to 10,000 with an explicit LIMIT. This keeps agent responses fast and within tool output limits.
• [All]: You can open Explorations from a link with a prompt already filled in using the initialPrompt URL parameter. Nothing runs until you submit it.
• [All]: Copy Pipes that run on on-demand compute now get a default max execution time of 30 minutes instead of the shorter shared infrastructure limit.
• [All]: You can now use JSON functions such as JSONExtractString in Data Source TTL expressions.
• [All]: S3 connector appends now have their own rate limit, so they no longer use up the quota for your other append requests.
• [All]: The Events API now reports the correct successful and quarantined row counts for large NDJSON requests.
• [All]: Branch names can now start with a digit, so Git branches like 3-fix-tokens no longer break tb build or CI. Deleting a Branch is also more reliable and no longer touches the main Workspace's resources.
• [All]: Removing a member from a Workspace now also removes their access to its Branches.
• [All]: You can no longer delete the last Workspace of an organization with an active paid subscription through the API or CLI, so you don't lose access to billing.
• [All]: Shared Data Sources are more reliable. Deleting or altering a Data Source shared with a deleted Workspace no longer fails, Forward deployments no longer get stuck after a consumer Workspace is deleted, and unsharing from a Forward Workspace through the API or sharing into a Workspace with a clashing name now returns a clear error instead of a 500.
• [All]: GCS Sinks that fail because of a billing problem on the bucket's Google Cloud project now show a clear error explaining how to fix it.
• [All]: The DynamoDB connector no longer fails when the export bucket includes an s3:// prefix.
• [All]: Queries with debug=analyze that hit the time limit now return a 408 timeout instead of a 500, and requests with a malformed Token to Data Source alter and delete endpoints return a 403 instead of a 500. Branch regression tests now return a 400 for invalid test parameters.
• [Forward]: Deployments that only change resource-scoped Tokens now show the affected Data Sources and Pipes as modified.
• [Forward]: During onboarding, you can now fix an invalid organization name on the first step and go back to edit it if setup fails.
• [Forward]: The Logs table header now stays pinned while you scroll, long values in the Data Source Settings tab wrap instead of overflowing, and the Time Series and Playgrounds sidebars keep your filter when you select a favorite.
• [Classic]: Fixed the processed data total showing 0 in the Metrics and Usage views, and a crash when closing or resetting the Test Parameters panel in the Pipe editor.

Organization admins can now select several members on the Members page and remove them in one step. The confirmation dialog lists who you're about to remove.
The Members page also loads much faster for large organizations, and removing members no longer waits on slow background requests.

See when your dedicated cluster rate limited an Endpoint

Organizations with dedicated clusters can now see cluster protection logs in the Overview and Observability pages. They show when a concurrency limit was applied to an Endpoint, why (for example, the cluster was severely degraded), how it changed while traffic recovered, and when it was cleared.

Failed Sink jobs in sinks_ops_log

Failed and cancelled Sink jobs now appear in the sinks_ops_log Service Data Source and in the Logs view, so you can monitor Sink errors next to successful runs. S3 Sink jobs that fail because of missing bucket permissions now return a clear access denied error that lists the IAM and bucket policy permissions required.

Improvements and bug fixes

• [Forward]: tb pull no longer fails when your Workspace has more Sink or Copy Pipes than your plan allows. The extra ones are pulled as regular Pipes with their SQL intact, and the CLI warns you so you can reconfigure them.
• [Forward]: Deployments that drop indexes or columns on large Data Sources no longer time out during promotion.
• [Forward]: Deployment job details no longer get stuck loading without showing an error.
• [All]: The Tinybird MCP server's call_endpoint tool now passes every value of array parameters to your Endpoints. Previously only the first value was used.
• [All]: Running a Copy Pipe through the API now checks that the Token has PIPES:READ on that Copy Pipe.
• [All]: Retried URL imports no longer create duplicate rows in Materialized Views.
• [All]: Import jobs queued behind large backfills are no longer marked as failed after waiting 4 hours.
• [All]: Dry-run deletes now return HTTP 400 with a clear error instead of HTTP 500 when the delete condition fails at runtime, for example when comparing a Date column with an invalid date string.
• [All]: The resource sidebar on Data Source, Pipe, and Endpoint pages now defaults to a narrower width so the main content gets more room. You can still resize it.
• [All]: Organization members without billing access now see Plan & Billing as an unavailable sidebar item with a tooltip explaining that only organization admins can open it.

Tinybird Local on arm64 requires a 4KiB page size kernel

The arm64 Tinybird Local image requires a Linux kernel with a 4KiB memory page size. Most distributions use 4KiB, so most setups need no change. On a kernel with a larger page size, ClickHouse® does not start.

If your kernel uses a larger page size, switch to a 4KiB kernel flavor if your distribution offers one, use a distribution that defaults to 4KiB, or run Tinybird Local on amd64. See Tinybird Local.

amd64 hosts, Tinybird Cloud, and your Workspaces are not affected.

See secondary indexes in Data Source settings

The Settings tab of a Data Source in Forward now lists its secondary indexes, including the expression, type, and granularity of each one. Settings you haven't configured show an empty state, such as "No indexes", instead of disappearing, and long values like sorting keys wrap without breaking the layout.

GET /v0/datasources now returns secondary indexes in the engine.indexes field, matching the single Data Source endpoint. In Classic, index granularity now shows the correct value instead of repeating the index type.

Disable block deduplication on Data Sources

Data Sources now accept the replicated_deduplication_window engine setting. It sets how many recent inserted blocks ClickHouse® remembers to detect and skip duplicated data. Set it to 0 when you delete and reingest the same data, or when a Materialized View writes identical blocks that you need to keep:

ENGINE "MergeTree"
ENGINE_SORTING_KEY "timestamp"
ENGINE_SETTINGS "replicated_deduplication_window=0"

You can only lower this setting. The default and maximum value is 10000. See engine settings.

Breaking changes

• Tinybird Local on arm64 requires a 4KiB page size kernel: The arm64 image runs only on hosts whose Linux kernel uses a 4KiB memory page size. On a larger page size, switch to a 4KiB kernel flavor, use a distribution that defaults to 4KiB, or run Tinybird Local on amd64. amd64 hosts are not affected.
• SQL API queries require a read scope: The SQL API now returns HTTP 403 when the Token has none of the ADMIN, WORKSPACE:READ_ALL, DATASOURCES:READ, PIPES:READ, PIPES:CREATE, or ORG_DATASOURCES:READ scopes. This includes queries that don't read any table, such as a SELECT 1 health check with a DATASOURCES:APPEND Token. Use a Token with a read scope for queries.
• Pipes with JWT fixed parameters are only queryable through their own Endpoint: Reaching a Pipe protected by JWT fixed_params through the SQL API or from another Pipe now returns HTTP 403. Call the Pipe's Endpoint instead, where the fixed parameters are always applied.
• Tinybird MCP returns tool results as text: Tools no longer include structuredContent in their results. call_endpoint now rejects parameters that the Endpoint doesn't define instead of ignoring them, and Endpoint failures now come back as tool errors. Update clients that read structuredContent to parse the text result.
• Deployments reject raw DateTime partition keys: New or changed Data Sources can no longer use a bare DateTime or DateTime64 column as ENGINE_PARTITION_KEY, because it creates too many parts. Use an expression such as toStartOfHour(timestamp) or toDate(timestamp).
• Stricter rules for display names: Token, Connection, Organization, and variable names are now limited to 100 characters and can't contain <, >, quotes, or backticks. Create and update requests with an invalid name return HTTP 400.

Improvements and bug fixes

• [Forward]: Deployments that alter a Data Source in place now validate ENGINE_SETTINGS against the same allowlist used for new Data Sources.
• [Forward]: Deleting a deployment or discarding a staging deployment no longer leaves behind Materialized Views that could block later deployments.
• [Forward]: Deployments now check bucket access for S3 Connections that use static access keys, so a missing or inaccessible bucket fails the deployment instead of a later import or sink job.
• [Forward]: FORWARD_QUERY errors now name incompatible columns with their expected and actual types, and include the ClickHouse® error for invalid queries.
• [Forward]: Materialized View backfill warnings now name the affected Pipe instead of an internal table identifier.
• [Forward]: The Query tab now shows node errors in a single panel, marks missing required parameters, and scrolls to the parameter bar when you test values.
• [Forward]: Refreshing a Token by name or ID, including with tb token refresh, now also rotates it in active deployments.
• [All]: Renaming a Token to a name that's already in use now fails, matching Token creation.
• [All]: Force-deleting a Data Source now cleans up all its Copy Pipes, even when some references are stale.
• [All]: Pipe parameters named after other Python builtins, such as property or object, work again. A Pipe with an invalid template no longer breaks the OpenAPI document for the whole Workspace.

Paid Workspaces can now run scheduled Copy Pipes on isolated on-demand compute

Add ON_DEMAND_COMPUTE true to the Copy node in your datafile:

NODE copy_sales
SQL >
SELECT *
FROM sales
TYPE COPY
TARGET_DATASOURCE sales_copy
COPY_MODE append
COPY_SCHEDULE 0 * * * *
ON_DEMAND_COMPUTE true

Scheduled executions then use on-demand workers instead of your Workspace compute.

Manual runs still require tb copy run --on-demand-compute in Forward or tb pipe copy run --on-demand-compute in Classic. Update to Forward CLI 4.6.20 or Classic CLI 6.5.7 to configure scheduled runs from a datafile. Read On-demand compute for scheduled Copy Pipes for use cases, tradeoffs, and a complete example. See on-demand compute for Copy Pipes and on-demand CPU pricing for reference.

Tinybird MCP uses one tool for all Pipe Endpoints

Tinybird MCP now exposes a single call_endpoint tool instead of generating one tool for every published Pipe Endpoint. Agents can use list_endpoints to discover an Endpoint, then pass its name and parameters to call_endpoint. This keeps the MCP tool list small even when a Workspace publishes hundreds of Endpoints.

This replaces the generated per-Endpoint tools. Update prompts or clients that call those tools by name to use call_endpoint. The deprecated explore_data tool now returns migration guidance instead of running its previous multi-step agent. Use list_datasources, list_service_datasources, and execute_query for data exploration.

Remote file imports use the v1 jobs API

Appending or replacing Data Source rows from a remote CSV, NDJSON, or Parquet URL now uses the v1 import jobs API by default in the UI and CLI. Existing tb datasource append and tb datasource replace commands keep the same interface. Forward CLI 4.6.19 and Classic CLI 6.5.7 include the change.

More metrics for dedicated clusters

The /v0/metrics API and organization.metrics_logs Service Data Source now expose additional metrics for replication health, queues and lag, MergeTree storage and part counts, background work, connections, caches, and disk space. The Service Data Source description now explains how to filter and aggregate the metrics, and includes organization_id in its schema.

Improvements and bug fixes

• [Forward]: Deployments can skip unnecessary backfills when changes to Null engine landing Data Sources don't require rebuilding downstream data.
• [Forward]: Re-running a failed Pipe query now clears the previous error and shows a loading state while the new query runs.
• [Forward]: Selecting text in a time-series legend no longer toggles the series when you release the mouse.
• [All]: Import jobs now run on a worker pool that scales automatically with demand, so imports queue less and finish faster during load spikes. No configuration changes are required.
• [All]: Missing-resource errors now name the Pipe that references the missing Data Source or table.
• [All]: SQL template parameters named type or dir no longer fail validation.
• [API]: Deployments containing an unsupported file type now return an HTTP 422 validation error instead of HTTP 500.

Tinybird MCP execute_query and generated Pipe Endpoint tools now accept a response_format of CSVWithNames, JSON, or Markdown. CSVWithNames remains the default, so existing clients keep the same output. Agents can also pass explain=true and optional explain_settings to execute_query to inspect a query plan before running the query.

The Classic migration period has ended

The migration period for Tinybird Classic has ended. We attempted to migrate all remaining Free and Developer Workspaces to Forward, but a small number had issues that prevented automatic migration. If yours is one of them, resolve the blocking issues and migrate it by October 16, 2026, the end of the final grace period.

• Free plans: Classic Workspaces that remain after the grace period will be deleted.
• Developer plans: Classic Workspaces will not be deleted, but Tinybird will stop providing Classic support and block access through the UI.

After migration, your data, Tokens, and Endpoint URLs remain unchanged. The new After migrating from Classic guide explains what changed and how to use version-controlled projects, Tinybird Local, Cloud Branches, deployments, SDKs, and agent skills in Forward.

Inspect Endpoint errors in pipe_stats_rt

The tinybird.pipe_stats_rt and organization.pipe_stats_rt Service Data Sources now expose an error_message column with the request error text. You can inspect error details alongside request statistics without querying the separate tinybird.endpoint_errors Service Data Source. The existing error flag remains unchanged, so current queries continue to work.

Breaking changes

• tinybird.query_validator_log has an updated schema: pipe_name is now resource_type and resource_name, and run_validation is now validation_time. fix_suggestion now contains the preferred suggestion, and the fix_suggestion_ai, workspace, database, host, and region columns have been removed. Update queries that reference the previous column names.
• Pipe control-plane requests now have a per-Workspace safety limit: Pipe metadata and Data Sink configuration routes allow 150 requests per second with a burst of 300. Clients that exceed the limit receive HTTP 429 and must retry according to the Retry-After header. Published Endpoint requests, the Query API, and the Events API are unchanged.
• New and changed Data Sources now validate TTL and partition-key compatibility: Tinybird rejects TTLs finer than the partition key and partition keys finer than hourly. Adjust the TTL or partition key before deploying. Existing unchanged Data Sources receive a warning instead of blocking the deployment. Compatible tables use part-level TTL drops by default to reduce retention work.

Improvements and bug fixes

• [API]: Published CSV Pipe Endpoints now accept __tb__csv_header=false to omit the column-name row. This lets clients concatenate compressed, paginated CSV responses without adding a header between pages. The response Content-Type includes header=absent, and the default CSV response remains unchanged.
• [Forward]: Branch creation now reuses the origin Workspace's trusted external ID for S3, DynamoDB, and Kafka Connections that use AWS IAM roles. Migrated Connections no longer require trust-policy changes before creating a Branch.
• [Forward]: Branch creation no longer blocks unchanged legacy TTL and partition configurations. Column reordering also no longer counts as a schema change during TTL validation.
• [All]: Free-plan Organizations now distinguish approaching and reached limits for daily requests, LLM usage, and storage on the Overview and billing pages.
• [All]: Copy Pipe query-analysis timeouts now report the actual 10-second analysis limit and recommend splitting complex SQL instead of suggesting a plan upgrade. Failed launches also record the request parameters, execution type, and mode in jobs_log.
• [API]: Empty Parquet uploads to the Analyze API now return the existing unsupported-file error instead of HTTP 500.

Breaking changes

• tinybird.rate_limits_log is now tinybird.cluster_control_log: The rate limits Service Data Source now uses the new name. Queries that read tinybird.rate_limits_log must switch to tinybird.cluster_control_log, which keeps the same schema and a six month retention.

Improvements and bug fixes

• [API]: A Token with DATASOURCES:APPEND scope for a Data Source can now read the import jobs it started. Append-only clients can monitor job status without a separate Token with DATASOURCES:CREATE scope.
• [All]: Copy Pipes with complex queries now get up to 30 seconds to resolve and validate their schema, instead of using the 10-second default. Table functions keep their existing 60-second timeout.
• [All]: Delete operations now wait correctly for mutations across ClickHouse® command formats. Large delete conditions no longer duplicate the full condition in the mutation status query, avoiding query-size errors while waiting for completion.
• [Forward]: Time-series charts and tables now display groups produced by conditional expressions, including 0 and 1, and selecting a group filters the chart correctly.
• [All]: Organization pages now show the Organization name immediately in the header, explain inaccessible Workspaces on hover, sort empty metrics alongside zero values, and keep status labels and replica rows on one line.
• [SDK]: TypeScript SDK 0.0.84 waits up to five minutes for branch creation, reports cancelled jobs immediately, and returns a clearer timeout error when branch creation doesn't finish.

Forward CLI 4.6.17 adds tb job retry for eligible import and S3 or GCS sync jobs in an error or cancelled state. The command starts a new job and returns its ID, so you can retry without switching to the API.

The JSON data type is no longer in private beta

Every workspace can create JSON columns and query them with no flag or request needed. If you were holding back dynamic or semi-structured event data because of the beta, you can start using JSON columns today.

Organization pages work without workspace access

Organization pages (Overview, Observability, Logs, Cluster) now load for members who don't belong to any workspace. You can read organization-level data with your user token alone instead of an empty screen.

Improvements and bug fixes

• [All]: Resource descriptions in the sidebar render as markdown, and the sidebar scrolls correctly instead of clipping long description lists.
• [All]: Authenticated requests resolve token scopes faster. Scope lookup no longer loads every Data Source in the workspace for each request, which reduces endpoint latency in the regions where it was most costly.
• [API]: Query requests that time out now return HTTP 408 instead of 500, so client-side timeouts no longer look like server failures.
• [Forward]: The Logs page Resources filter now matches job rows by pipe or Data Source name, including deleted resources, and ignores case.
• [Forward]:
  tb deployment promote --wait
  and
  tb deployment discard --wait
  exit with the deployment error instead of polling forever when removing the old deployment fails.
• [SDK]: The TypeScript SDK now serializes JSON parameters correctly.

From the blog

Cluster Management: keep the controls, skip the cluster ops: Observe workload signals, resize Dedicated clusters, and rebalance traffic from the Organization UI or API. Part of the Cluster Management campaign that kicked off this week.

The default execution timeout for Query API and Pipe endpoint requests is now tiered by plan: 10 seconds on the free plan, 20 seconds on paid plans with shared infrastructure, and 30 seconds on dedicated infrastructure. Workspaces with a custom timeout override keep their existing setting.

New replica column in pipe_stats_rt

The pipe_stats_rt service Data Source now includes a replica column showing which ClickHouse® replica executed each request. Use it to spot load imbalance or latency differences across replicas.

Improvements and bug fixes

• [All]: Import jobs now use a routing algorithm that picks the least-loaded ClickHouse® replica instead of always using a single fixed server, on clusters with multiple writer replicas.
• [API]: Fixed browser-based uploads to the v1 Data Source append and replace Endpoints, which previously failed CORS preflight checks.
• [API]: You can now retry failed or cancelled S3 and GCS sync jobs. Retries skip files that earlier attempts already imported instead of re-importing everything, and work on jobs with many files.
• [Classic]: Fixed the dashboard Jobs widget showing no results after filtering while on page 2 or later.
• [Forward]: Deployments no longer fail when a materialized pipe has a leftover alter deployment method directive with nothing left to change; Tinybird shows a warning instead.

Until now, changing the right side of a Materialized View join forced a rebuild of the Materialized View and every resource downstream of it. Deployments now follow only left-table dependencies, so a change to the joined table alone skips those backfills. For a table joined by several Materialized Views, changing its partition key now rebuilds only that table.

Improvements and bug fixes

• [Forward]: Sending events to /v0/events for a Data Source that does not exist yet returns HTTP 404 with an explanation instead of a 500. Data Sources in Forward Workspaces are created through deployments.
• [All]: Ingestion failures caused by duplicate columns are now treated as user errors. They are no longer retried, so the error surfaces immediately instead of after retries that cannot succeed.

From the blog

How Guideless uses Tinybird to build training people actually finish: Play rates, completion rates, and step-level drop-off for every published guide, powered by Tinybird instead of a separate analytics backend.

The v1 Data Sources API can now append or replace rows from a remote CSV, NDJSON, or Parquet file. Pass the file URL and format to the existing /append or /replace endpoint instead of downloading and uploading the file yourself.

Improvements and bug fixes

• [Forward]: Fixed a bug that made deployment backfills fail with certain partition keys.
• [Forward]: Deployment validation no longer repeats Connection and bucket checks when multiple Sink Pipes use the same Connection.
• [Forward]: Promoting a missing or deleted deployment now returns HTTP 404 instead of an internal error.

Forward now retains the existing quarantine table when a Data Source deployment changes its sorting key, partition key, or column types without changing the set of columns.

Quarantined rows remain available to users after the deployment instead of requiring support help to recover them.

Seven-day workspace trends

The Organization Workspaces table now shows seven-day sparklines for vCPU time, memory, and errors. Totals remain available if a trend query times out.

Improvements and bug fixes

• [Forward]: ALTER-based deployments can now drop a column across a landing Data Source and Materialized View flow. The Materialized View query is updated before the referenced column is removed.
• [Forward]: DynamoDB-backed Data Sources now show the correct type and detail link in the UI.
• [All]: Invalid UTF-8 ingestion now returns a clear HTTP 400 error instead of an internal failure.
• [Forward]: Classic to Forward migrations no longer fail when a legacy Data Source lacks cached engine metadata.
• [All]: Copy job timeout and memory errors now point to on-demand compute when it is applicable.
• [Classic]: The region list in the workspace menu no longer reorders while workspace information loads.
• [All]: Delete, Copy, and Sink job concurrency errors now clarify that working and queued jobs both count toward the limit.

From the blog

Better organization-wide visibility, for every Tinybird user: A tour of the new Organization Overview and Logs pages available to organization members.

Slate scopes product ideas around users, not infrastructure: How Slate replaced a Redis bottleneck with Tinybird to deliver real-time prices and profit and loss data.

From OLTP to OLAP: 7 ClickHouse® Optimization Mistakes: Seven common OLAP modeling and query mistakes, with practical fixes.

Tinybird CLI 4.6.14 for Forward and 6.5.6 for Classic now use the v1 import flow by default when appending or replacing a Data Source from a local CSV, NDJSON, or Parquet file. The --experimental=use_v1 option is no longer required.

Forward CLI 4.6.14 returns the queued job ID by default. Pass --wait to wait for completion. Classic CLI 6.5.6 waits for completion by default. Pass --no-wait to return the queued job ID immediately. tb datasource append --events also uses the v1 import flow. Imports from remote URLs keep their existing behavior.

The Tinybird UI and Tinybird Local now use the same v1 flow for local file append and replace operations.

This requires the latest Tinybird Local version because local v1 imports depend on the newer append and replace endpoints.

Conditional TTLs

Data Source deployments now support conditional TTL rules, including multiple rules and DELETE WHERE conditions. TTL validation uses ClickHouse® syntax directly, so ENGINE_TTL accepts the same expressions as ClickHouse®. You can now do things like:

ENGINE_TTL "ts + INTERVAL 6 MONTH WHERE client_id != 'long-ttl-client', ts + INTERVAL 1 YEAR WHERE client_id = 'long-ttl-client'"

Live data-movement progress for deployments

Forward deployment pages now include a Data movements tab when a deployment requires a backfill. It shows each affected Data Source, its current status, elapsed or estimated time, and progress. Deployment and job status update automatically while work is running.

Unified job type for deployment backfills

Deployment backfills now log with job_type = 'deployment_backfill' regardless of how they move the data.

Until now, backfills that ran as a populate were logged as job_type = 'populate', with no metadata linking them to the deployment that triggered them.

Now all backfills log the same way in tinybird.jobs_log and tinybird.datasources_ops_log, including deployment_id and deployment_job_id in job_metadata, so you can tie every backfill back to its deployment.

Rows written before this change keep their original job type. To find old and new deployment backfills in the same query, filter on both:

SELECT * FROM tinybird.jobs_log WHERE job_type IN ('populate', 'deployment_backfill')

Updated Tinybird agent skills

The Tinybird agent skills now guide coding agents to parse JSON once when extracting several fields in a Materialized View, instead of parsing the same payload once per field.

The skills also recommend aligning partition granularity with Data Source TTL windows and using ttl_only_drop_parts=1 when appropriate, allowing ClickHouse® to drop complete expired parts instead of rewriting partially expired ones.

Organization info available to all users

Organization visibility is no longer limited to admins. Every member of an Organization can now see the Overview, Logs, Observability, Workspaces, Members, and Managed regions pages, so the whole team gets the same picture without asking an admin for access.

Admin-only actions stay protected: managing the cluster, viewing Billing, and adding, removing, or changing member roles in Workspaces they don't administer.

Improvements and bug fixes

• [All]: Data Source creation no longer selects Tuple or Map columns as inferred sorting keys. Schemas containing only composite columns now fall back to ORDER BY tuple() instead of failing to create the table.
• [All]: Strict NDJSON ingestion now handles compatible named objects for Map(String, T) columns, including numeric and nullable value types.
• [All]: Organization observability charts now use the same color for each replica in cluster load and memory charts as in the replicas table.
• [All]: You can now use do_not_merge_across_partitions_select_final as a query setting. When querying with FINAL, this setting merges parts only within each partition, which can speed up queries on well-partitioned Data Sources.
• [Forward]: Tinybird CLI 4.6.14 fixes tb migrate-to-forward so it finds and offers to delete all Classic branches before migration, including branches not created by the user running the command.

JSON ingestion preserves integer values

/v0/events ingestion into JSON and Array(JSON) columns now preserves integers when later values exceed the type inferred from the first value. Previously, a sequence such as 50 followed by 1944 could store the second value as -104.

Rows already affected are not repaired automatically and must be re-ingested.

Three months of real-time Pipe statistics

The tinybird.pipe_stats_rt and organization.pipe_stats_rt service Data Sources now retain three months of real-time Pipe request statistics instead of one month. This provides a longer window for investigating historical requests without falling back to daily statistics.

Experimental v1 ingestion in the CLI

Tinybird CLI 6.5.4 for Classic and 4.6.13 for Forward add experimental v1 ingestion for local files:

• tb datasource append --experimental=use_v1
• tb datasource replace --experimental=use_v1

The commands support CSV, NDJSON, and Parquet files, including gzip-compressed files. They return the queued import job immediately. Pass --wait to wait for completion. Remote URLs are not supported with this experimental option.

Breaking changes

• Pipe template parameter validation: Tinybird now rejects Pipe template type functions that declare a parameter without a valid, non-empty string name. Existing Endpoints continue serving traffic, but affected templates must be corrected before their next push, build, or deployment. The validation is included in Tinybird CLI 6.5.4 for Classic and 4.6.13 for Forward.

Classic: Up to 3 mutations in a single alter

The /v0/datasources/(.+)/alter endpoint now allows up to 3 mutating operations (such as DROP COLUMN or MODIFY COLUMN) in a single request, combined in one ALTER statement that runs as a single mutation. Previously, only one mutating operation was allowed per request, even alongside a metadata-only change.

Metadata-only operations, such as ADD COLUMN, don't count towards this limit and can still be combined freely with mutating operations. Requests that generate more than 3 mutating operations are rejected with an error message.

Improvements and bug fixes

• [All]: Added an email notification for users whose Workspaces are scheduled to be upgraded to ClickHouse® 25.3, ahead of the upcoming deprecation of 24.12.
• [Forward]: Failed deployment jobs no longer remain stuck as working when a concurrent Workspace update prevents the deployment status from being saved.
• [Forward]: Tinybird CLI 4.6.13 includes vendored Data Source files when running tb build, preventing false drift errors in branches that reference shared Data Sources.
• [Forward]: Authenticated users without a Workspace now see a "No workspace found" page instead of entering a redirect loop.
• [Forward]: The copyable prompt under Overview > Integrate now includes every Data Source and Endpoint in the Workspace instead of truncating each list after 10 resources.

Tinybird Classic is being sunset for Free and Developer plan Workspaces. If you have a Classic Workspace on one of these plans, migrate it to Tinybird Forward by September 15, 2026. After that date, remaining eligible Classic Workspaces will be scheduled for automatic migration. Enterprise plan Workspaces are not affected.

Forward is the main Tinybird product experience. Schema changes, local development, branching, CI/CD, SDKs, and AI-assisted workflows are all better in Forward. The migration guide walks through the process step by step, including the tb migrate-to-forward command.

Classic branches (part of the Versions functionality) are not migrated to Forward and will be removed during migration. Review how you use Classic branches before proceeding.

As a thank you for migrating early, we're offering a coupon for two months at 50% off your current plan, or one month free if you upgrade from Free to a paid plan. This offer is valid until August 15, 2026. Contact [email protected] after completing your migration to claim it.

Classic Workspace creation in the UI is now restricted to organizations on an Enterprise plan.

DynamoDB sampling for cloud branches and local

DynamoDB Data Source imports now support sampling in cloud branches and Tinybird Local. Main branches always get a full export on creation. Cloud branches and local can sample by row count or data size limits, or request a full export. This avoids slow exports on branches and reduces unnecessary egress costs.

tb_secret in Forward sink export settings

Forward Sinks now support {{ tb_secret(...) }} in EXPORT_* metadata fields like EXPORT_BUCKET_URI. Secrets are resolved at deploy time and baked into the Sink configuration. Empty secrets fail clearly instead of producing an invalid s3:// path. Changing a Workspace secret used in an export does not update existing Sinks: redeploy to pick up the new value.

v1 replace endpoint

New v1 endpoints are available for local file imports:

• POST /v1/datasources/{name}/append?format={csv|ndjson|parquet}
• POST /v1/datasources/{name}/replace?format={csv|ndjson|parquet}

These endpoints replace the equivalent v0 import request:

• POST /v0/datasources?name={name}&mode={append|replace}&format={format}

For replace operations, the v1 endpoint also accepts:

• replace_condition={SQL condition} — equivalent to --sql-condition
• skip_incompatible_partition_key=true — equivalent to --skip-incompatible-partition-key

The v0 import flow returns an import result that the client processes as a synchronous operation. The v1 endpoints stage the upload and enqueue an import job, returning immediately with the job response and import_id. Use GET /v0/jobs/{import_id} to follow the import until completion.

Improvements and bug fixes

• [All]: Log detail rows now resolve to clickable resource links for Pipes, Endpoints, Sinks, and Data Sources. Data Source operation logs show a related Pipe row when metadata is available. The logs table adds a hover state and persistent highlight for the row open in the split panel.
• [All]: ALIAS_REQUIRED and DECIMAL_OVERFLOW errors are now classified as user errors and no longer retried by the platform. Queries hitting these conditions fail faster instead of accumulating retry attempts.
• [Forward]: TypeScript SDK 0.0.82 changes the default branch_data_mode to no data. Cloud branches are now created empty by default. Set branch_data_mode: "last_partition" in your config to keep copying the last partition of production data into new branches.
• [All]: Copy Pipes that reference a deleted or missing target Workspace now return HTTP 404 instead of a generic error.
• [All]: Pipe creation with valid JSON but wrong field shapes now returns HTTP 400 early, instead of failing later when the handler treats values as mappings.
• [Forward]: Changing index_granularity now forces a new table plus backfill instead of attempting an ALTER, since this setting changes how data is stored on disk.

From the blog

The DynamoDB Connector is now in Tinybird Forward: How to mirror DynamoDB tables to Tinybird Forward with a .connection file and a single deploy, including the Classic-to-Forward migration path.

The CLI now picks the target Workspace in Tinybird Local more predictably during branch-based development. By default, commands point to the local Workspace mapped to the current Git branch. If you pass --token or set TB_TOKEN, the Token overrides Git branch detection and the command runs against the Workspace that belongs to that Token.

Sink jobs now pick the replica with the lowest load, using the same placement logic as Copy jobs and Data Source imports. Replica selection also accounts for running Sink queries, so Copy, import, and Sink placement all see the same picture of cluster load. This reduces the chance of a Sink job landing on a busy replica and slowing down.

Improvements and bug fixes

• [All]: The hostname and ProfileEvents columns are now exposed when querying system.query_log, making it easier to see which replica ran a query and more insights of the query execution.
• [All]: Import jobs created with a capitalized format, such as Parquet instead of parquet, no longer get stuck in job tracking. Previously they held a worker slot for up to 5 minutes after finishing, which throttled subsequent imports.
• [All]: Truncating a Data Source with a long history of deduplicated inserts no longer fails. When the regular TRUNCATE cannot complete, Tinybird now swaps in an empty copy of the table instead of returning an error.
• [All]: Fixed Omni integration through the ClickHouse HTTP Interface.
• [Forward]: Tinybird CLI 4.6.7 fixes an intermittent tb login failure where the browser flow returned a JSON decode error, and now raises an actionable message when authentication cannot complete.
• [Forward]: The Endpoint integration snippets in the UI now build the correct URL when working against Tinybird Local.
• [Forward]: TypeScript SDK 0.0.81 uses server-side auto-promote for deployments instead of polling and manually setting the deployment live and deleting the previous one. tb deploy now accepts --wait / --no-wait and --auto / --no-auto flags (both default to true), matching the CLI.