Feature Updates
- Agents can now find and update existing artifacts instead of creating duplicates, preserving version history across sessions. Artifact pages also show the artifact name in the browser tab. Legacy chat-related artifact parameters are no longer accepted by artifact APIs or CLI commands.
- HEIC attachments now receive PNG thumbnails when supported, with a fallback placeholder when previews cannot be generated.
- Direct API-created tables now start empty unless records are provided, while tables created through the UI continue to include three empty rows.
- Large table lists and grids now load fewer fields, skip total counts by default, and support cursor-based pagination for faster sequential browsing. API clients requiring totals must explicitly request them or use the count endpoint.
- Substring search now focuses on text fields and text-based formula or lookup results, improving relevance and reducing indexing overhead. All-field substring search may change for wide tables and unsupported field types.
- App Builder now provides a clearer Preview and Live switch, remembers the selected mode per app, and offers a guided Publish action when no live version exists.
- Space administrators can configure AI field concurrency limits in space settings. Values above the platform maximum are preserved but clearly shown as capped at the effective limit.
- Skill installation instructions and the Teable CLI now consistently use the current instance endpoint across commands, configuration, authentication, and error messages.
Bug Fixes & Improvements
- Improved record query performance and stability, reducing resource usage on high-traffic read operations without changing query behavior.
- Fixed recent-base ordering so newly copied bases appear first, and consolidated duplicate collaborator entries using the highest available role.
- Fixed conditional rollups so nested OR groups and dynamic match conditions are saved and calculated correctly across the UI and API. Nested controls also remain accessible in narrower settings panels.
- Rollup and conditional rollup APIs now reject unsupported source-field and aggregation combinations with clear validation errors, matching UI rules. Clients submitting previously accepted invalid configurations must update their requests.
- Improved computed-field recalculation, reducing stale or blank downstream formulas, lookups, rollups, and summary values after linked or dependent data changes.
- Improved filtered linked-record rollup recalculation to reduce timeouts when multiple rollups use the same linked records.
- Added reliable deep links for Computed Outbox task lineage and anomaly details, including shared links and direct URLs in Copy for Agent output.
- Fixed the linked-record picker so selected records and checkbox states remain consistent when switching tabs, reopening the picker, or loading records asynchronously.
- Browser-based AI sandbox tasks now use the bundled browser tools without repeated downloads or manually configured executable paths, improving startup speed and runtime consistency.
- Fixed overlapping controls in the AI chat external-link confirmation dialog.
- Short links now redirect directly to their canonical destination, reducing intermediate page loads.
- Fixed Auto Number greater-than filters that could cause page and row-count requests to fail, and ensured saved numeric values display correctly when filters are reopened.
- Improved editable shared-view access controls so record and field changes remain within the permitted view scope, including selection-based paste, clear, and delete actions.
- Fixed cross-base conditional lookup and rollup settings that could incorrectly show linked tables as unavailable after reopening. Empty foreign-base references are now handled consistently as same-base fields.
- Improved selected-row summaries so aggregations more reliably respect filters, sorting, search, row ranges, and collapsed groups, including timestamp fields.
- Fixed single-select rollups so unique values preserve first-appearance order across updates, refreshes, and API responses. Count rollups now count unique non-empty values, which may reduce results where duplicate values were previously counted.
- Improved record creation and schema update reliability when imported or deleted linked-field references are inconsistent, preventing repeated failures and inaccessible updates.
- Reduced repeated node-tree refreshes during AI-assisted script editing, publishing, chat, and automation interactions, improving responsiveness and lowering unnecessary network traffic.
- Fixed record comments under the authority matrix so authorized collaborators can comment on accessible records, read-only users can view threads without write controls, and shared links do not expose comment controls.
- Fixed deleted comments reappearing or blocking later deletions, and improved comment image uploads with clearer progress, failure handling, and send prevention while uploads are incomplete.
- Improved App Builder recovery so temporary connection issues do not unnecessarily reset active workspaces or discard uncommitted changes.
- Fixed App Builder publishing failures caused by version history becoming misaligned during generation cleanup.
- Fixed navigation for non-admin users so inaccessible resources do not appear as empty folders or invalid references across base lists, tree views, cached navigation, and AI mentions.
- Improved the permission matrix with clearer status icons, hover details, and more responsive sidebar behavior in narrow layouts.
- Fixed archive permissions so restricted collaborators can archive records within their authorized scope without delete permission. Archive exports now also respect export permissions, and grouped, sorted, and counted views update correctly.