Structured Query V2 now supports same-project, single-hop lookups through to-one Link fields, including related stored and system values. Each source’s permissions are respected, and missing linked records return null. Inverse relationships, multi-record relationships, cross-project queries, and related filtering, sorting, and aggregation are not supported; totals remain root-only, and relation queries no longer automatically fall back to SQL.
Internal and shared forms now support URL-based pre-filling for visible text, number, and single-select fields. Pre-filled values remain editable, and hidden fields cannot be pre-filled.
Added a Trash entry to the chat-first sidebar for easier recovery of deleted projects.
New App Builder apps now use structured queries for single-table lists and aggregations, with SQL fallback for more complex queries. Query guidance follows each app’s configuration, while existing apps retain their current query workflow.
Routine creation now opens a new AI conversation with guided prompts and five examples. The redesigned routine page brings together details, run history, and linked conversations, with controls for editors to continue chats and launch runs. Manual configuration becomes available after creation.
Bug Fixes & Improvements
Improved AI response cancellation so immediate stop requests work reliably and delayed requests do not interrupt newer replies.
Improved AI conversation reliability during long-running tools and response cleanup, preventing new messages from being rejected while the previous response is still finishing.
Fixed repeated errors when opening, refocusing, or reconnecting a new AI chat before its first message is sent.
Field creation now shows a loading state and prevents duplicate submissions until saving completes, including when saving through the unsaved-changes dialog.
Improved the reliability of computed-field updates and imports interrupted by database connection issues.
Fixed configured defaults not being applied to blank fields in records created through the grid’s add-record dialog. Previously populated fields can still be explicitly cleared.
Share settings now reflect changes immediately and consistently preserve the latest selection during rapid permission updates.
The chat-first sidebar now refreshes automatically when projects are moved, copied, imported, created from templates, or restored.
Improved load times for synchronization run records.
Automation Script nodes now run on Node.js 24 instead of Node.js 22. Review scripts that depend on Node.js 22-specific behavior or compatibility.
Improved routine ownership, alerts, conversation visibility, unread status, and run history, including reliability when owners or permissions change. Editing controls are now hidden for members without update permission.
New apps with structured queries enabled now use queries.yaml for supported record and aggregate queries, including filtering, sorting, pagination, and parameters. Unsupported query shapes are clearly rejected with alternatives, while existing apps retain their current SQL behavior.
Routines now have a dedicated page for configuration and run history, with direct access to each run’s conversation. Editing includes autosave and next-run visibility, and ownership transfers to an eligible base or space owner when needed.
The command palette, opened with Cmd+K on macOS or Ctrl+K on Windows, provides quick access to project resources and common actions. It supports keyboard navigation, recent items, permission-aware results, and natural-language questions through Cuppy.
Structured sorting now supports explicit NULLS FIRST and NULLS LAST ordering for records and aggregate queries while preserving existing defaults.
AI image generation and regeneration are no longer available for lookup, formula, or other computed attachment fields. Editable source attachment fields continue to support generation, and lookup attachments continue to display, preview, and synchronize normally.
Bug Fixes & Improvements
Fixed an issue preventing Claude.ai and Claude Desktop custom connectors from connecting to Teable MCP. Existing connectors may need to be removed and added again.
Fixed conditional rollups with multi-value filters so totals refresh correctly when source values change, including for newly created equivalent fields.
Improved field deletion dependency checks to use the current version of each authorized App, reducing response time and ensuring results reflect current dependencies.
Field deletion confirmations now identify referencing Apps and named queries, including the App, query, version, field, and usage. Queries with unresolved dependencies are marked as unknown, and checks remain advisory without changing queries or blocking deletion.
Improved routine permissions and mobile behavior so read-only members can view routine details without restricted controls, and opened run chats are titled and marked as read correctly.
Chat-first workspaces now open the first project conversation directly when opening Teable or a space, without showing an intermediate page or repeatedly prompting users to update the layout.
Query dependencies can now be retrieved through Teable CLI, MCP, and generated clients, with filtering by table and field and enforcement of existing access permissions.
Added field-level impact analysis to identify apps and queries affected by schema changes, respecting app visibility permissions.
App Builder chats started from Base conversations now support OneDrive authorization without clearing or restarting the chat. Authorization prompts remain available throughout the conversation.
Bug Fixes & Improvements
Improved the chat-first layout on screens 640px wide or narrower. The mobile navigation drawer now closes after navigation. Conversations fill the chat page, while resource and app pages provide an on-demand bottom chat drawer to keep app previews visible at all times.
Fixed an issue where “Open in project” repeatedly opened the mobile chat drawer and changed the user’s desktop docking preference.
Teable CLI 0.6.47 reduces startup memory usage and improves the reliability of help, validation, and individual commands.
When CLI read results or table output are too large, output is now truncated or the request is rejected with a clear error message suggesting pagination or fewer fields, rather than terminating unexpectedly. Node.js heaps smaller than 64 MB are not supported.
Improved the accuracy of cell values returned by the CLI.
Invalid or expired OAuth refresh tokens now return a clear 401 response instead of a server error.
Conditional lookup and Rollup filter conditions are now validated when fields are created or updated. Unsupported operators and incomplete values are rejected with specific validation errors, rather than causing generic calculation failures later.
Improved the timeliness and reliability of query dependencies when apps are published concurrently.
Model selections are now saved per chat and applied consistently to directly sent, queued, and steering messages, including in App Builder and routine conversations. The selector shows the model actually in use, including model changes during a response. If the selected model is unavailable, the administrator-configured default model is used instead.
Fixed an issue on iPad where reopening a chat through a deep link could incorrectly display “New chat” and the default model.
AI context tags now remain stable when selected records or cells refresh. Detail cards prioritize readable table names, record titles, and selected field values over internal identifiers.
Introduced an experimental chat-first web workspace with space-level navigation, a dedicated new-chat page, tabbed outputs, and a persistent conversation dock on resource pages. Users can switch back to the classic layout from the personal menu.
AI chat references now show meaningful record titles, field names, source tables, and selection counts. References remain attached to the original selection as data changes, with inspectable previews and clear indicators for unavailable data.
AI conversations now summarize completed resource changes and provide navigation links. Table and record links open in the current page while preserving the active conversation and its context.
Stripe integrations now support separate test and live accounts alongside an existing connection. Apps can switch between accounts without code changes or affecting other apps.
Added feature-gated, app-local named queries for reusable, read-only data access, including parameterized filtering, sorting, pagination, aggregation, and single-record retrieval. Existing SQL-based Apps remain unchanged.
Named queries now support optional filters and sorts, parameterized sort fields and directions, ordering aggregate results, optional total record counts, and deterministic deduplication for multi-value lookup and rollup results.
Added a feature-gated query debugger in the public API, CLI, and App Builder to inspect queries without executing them or reading business data. App Builder supports both saved and unsaved queries.
Added query manifest linting to the CLI and App-local query tool, with configurable thresholds, JSON output, and an optional strict mode. Improved usage detection reduces false unused-query warnings.
Added configurable retention controls for audit logs, automation runs, record history, and deleted records. Retention is disabled by default, requires administrator configuration, and excludes archived records.
Added a standalone Export my data option in account settings so users can request an account archive without starting account deletion.
Bug Fixes & Improvements
Improved security across Pins, workspace permissions, cross-base links, authentication, administrator connection tests, third-party credentials, file processing, and platform dependencies.
Improved chat-first navigation and layout reliability, including space switching, conversation dock persistence, history controls, project indicators, preview tabs, and collapsed folder state.
AI-assisted actions now open the conversation dock with the intended request and context preserved. Improved imports, project menus, shared-project navigation, search, output refreshes, inline renaming, automation script editing, and personal project ordering.
Structured Query API filter errors now identify the affected field, its type, the rejected operator, and supported alternatives. Unsupported operators now return data_query.filter_operator_unsupported instead of data_query.filter_incompatible.
Large CSV imports from URLs now complete the download before writing rows, reducing partial imports caused by interrupted transfers. Added bounded retries and clearer errors while preserving fast previews and upstream timeout responses.
Fixed blank galleries when the primary field was hidden. Gallery cards and attachment covers now render correctly, and primary fields remain visible through duplication, restore, and other updates.
Record updates that conflict with field deletion now return a clear conflict response instead of an internal server error. Improved reliability when schema changes coincide with single or bulk updates.
Standardized interface colors and transparency so borders, backgrounds, dividers, and badges render consistently in light and dark modes.
Fixed collaborator fields not displaying users correctly and inconsistent field ordering when creating views.
Restored connector brand icons and prevented broken logo assets from interrupting catalog browsing.
Improved automation variable editors across email, HTTP, AI prompt, and record update actions. Fixed chip highlighting and overflow, disappearing values, lost edits, and incorrect variable replacement.
Active subscriptions now remain available after a space is transferred to a valid owner during account deletion, without incorrectly reporting that the seat limit has been exceeded.
Improved the record comment editor while preserving mentions, links, keyboard shortcuts, and image upload, resizing, and previews. Fixed formatted-text submission failures, whitespace-only comments, and duplicate submissions from rapid Enter presses.
Improved filtered list performance for numeric lookup fields, table search performance on large tables, and index recommendations for filtered, paginated lists.
Improved named-query validation and App tooling, including clearer errors for invalid query names and undeclared parameters, preservation of generated query files, and reliable replacement previews after cancellation.
Fixed date filters using an empty string so equality and inequality comparisons correctly match empty and non-empty dates.
Fixed the blank API documentation page in self-hosted environments.
Fixed summary calculations involving lookup aggregations and nested lookup fields.
Fixed invalid filtering when collapsing groups based on multi-select lookup values.
Fixed automation AI actions repeatedly failing when JSON output was enabled but the prompt did not explicitly mention JSON.
Empty record update requests now return a clear validation error instead of an internal server error.
Chart queries referencing deleted aggregation fields now return a clear field-not-found error instead of an internal server error.
Fixed records briefly appearing in the wrong groups or order when opening or switching views. View settings, hidden-field counts, and collapsed groups now remain consistent during loading, without showing stale data.
Invalid display settings left after field type conversions no longer prevent bases, fields, and views from loading. Affected text and number fields fall back to their standard display.
Last modified time fields that track specific fields now remain empty when a new record has no values in those fields. They continue to update when tracked values are added, changed, or cleared.
Improved computed-field update performance for deep lookup chains and workflows with many dependent records.
Fixed lookup and conditional lookup fields referencing AI fields so they can be renamed or switched to another AI field.
Added an official MCP server for connecting Claude Code, Codex, Cursor, and other compatible AI clients to Teable through guided OAuth authorization, without manually creating a personal access token. Self-hosted instances support connections at /mcp without additional configuration.
Connected AI clients can list projects, inspect tables and fields, query and manage records, and discover APIs for advanced operations, all within the user’s existing permissions.
Added Settings → MCP to review and revoke AI client connections. Personal access tokens remain available for CI and other browserless environments.
Bug Fixes & Improvements
Fixed an issue where tables could remain unavailable after a network timeout. Once connectivity is restored, retrying reloads the same table without requiring users to reopen it from the project directory.
Single-line text fields now trim leading and trailing whitespace and replace line breaks and tabs with spaces when records are created or updated. Integrations that rely on preserving this whitespace should account for the change. Long-text fields continue to preserve line breaks.