Feature Updates
- Added admin-configured AI model tiers—Ultra, Smart, Standard, and Lite—with centralized model mapping updates. Tier-based chats and apps follow the current mapping, while explicitly fixed models remain unchanged. When no model is specified, Smart is used by default.
- Image uploads now adapt based on the capabilities of the model assigned to the selected tier.
- Added AI-powered App Builder suggestions with preset prompts, direct prompt submission, and preserved table context.
- After email is configured, registered users can log in with an email verification code.
- After configuration, the web login page supports Sign in with Apple while continuing to support Google, GitHub, and OIDC.
- Personal access tokens and OAuth apps can now use Routine as an authorization scope.
- Routine runs can now be updated, activated, deactivated, or started via the CLI within the existing permission scope.
- Routine run details now display the assistant’s final response for the run when available.
- Public settings now include the server build version, making it easier to identify the version of the connected self-hosted server.
Bug Fixes & Improvements
- Fixed App Builder file links and generated file downloads to ensure they open in the appropriate viewer.
- Improved automation script editing: changes are preserved when closing the dialog, and the latest saved configuration is used during workflow tests.
- Improved stacked notifications to keep concurrent messages visible and easy to read.
- Fixed an issue where the AI chat send button shrank in constrained layouts.
- Improved Routine terminology in Simplified Chinese and refined Routine editor styling and form consistency.
- Updated user-facing and API documentation terminology from “Base” to “Project,” clarified API route titles, and corrected authentication headers in OpenAPI examples.
- Strengthened permission isolation for app and automation credentials to prevent unintended access across projects, spaces, integrations, or tenants. Workflows that rely on broader inherited access may need to be reviewed.
- Improved the reliability of V2 record queries by allowing slow record, count, and aggregation queries to be canceled through configured time limits, with clearer timeout errors. Deployments without configured limits retain the existing behavior.
- Record queries now skip total count calculations unless explicitly requested, reducing database workload while preserving totals in workflows that require them.
- Improved cursor-based record pagination so deleting the record corresponding to a cursor no longer causes missing results or premature termination. Expired or mismatched cursors now return a validation error; clients should continue to treat cursors as opaque tokens.
- Fixed “contains” and “does not contain” filters for JSON and multi-value fields when values contain apostrophes.
- Improved the reliability and performance of computed fields, and ensured that resolved computation warnings are cleared only after all affected fields have been fully recomputed.
- Views restored through undo, redo, or the recycle bin now appear immediately, while unnecessary table and view refreshes are reduced.
- Improved the reliability and compatibility of Composio tools while preserving existing parallel execution behavior.