We have shipped a substantial update focused on data-driven automation, workflow reliability, and platform performance.
The biggest addition is Data Tables, which gives teams a built-in way to store and work with structured information inside Tiledesk. We also introduced JSON Condition V2, safer publishing, faster loading for AI actions, new Knowledge Base chunking controls, and several improvements across Analytics, voice, and the widget.
New releases
Data Tables are now available
Tiledesk now includes Data Tables, a built-in way to organize structured records and use them inside AI Agent workflows.
Instead of relying on an external spreadsheet or database for every simple use case, teams can keep operational information closer to the flow. Data Tables can support scenarios such as:
- Contact and lead records
- Product and warranty information
- Repair or service requests
- Approval records
- Internal workflow data
The first release also includes several controls that make Data Tables easier to manage:
- View the current table size in MB against the maximum allowed size
- Filter records using conditions
- See full table names through sidebar tooltips when names are truncated
- Access a documentation image and guide link directly from the Data Table action tooltip
- See a cleaner action interface, with dividers displayed only when filter conditions are active
This creates a stronger foundation for AI Agents that need to work with structured business data, not only generate or retrieve text.
Read more about the Data Tables here
New JSON Condition V2
A new version of the JSON Condition action is now available.
JSON Condition V2 supports a broader range of operators and more advanced checks, making it easier to build flows that react to structured values, dates, arrays, lengths, and other data conditions.
The previous JSON Condition V1 is now hidden from the action list, but it continues to work in existing flows. This protects older projects while guiding new workflows toward the updated version.
Regular expression chunking for TXT and Markdown
When uploading TXT or Markdown files to the Knowledge Base, you can now define how content should be divided into chunks using regular expressions.
This gives teams more control when working with structured text, such as:
- Logs
- Product catalogs
- Repeated document sections
- Technical records
- Text files with custom separators
The Knowledge Base content filter has also been updated so files using this custom chunking method are correctly included in TXT results.
Expanded ElevenLabs support
Voice settings now retrieve models and voices from ElevenLabs through a more streamlined process. ElevenLabs support has also been added to the relevant project configuration areas.
Changes
Publishing now waits until everything is saved
The Publish button is now disabled while the current project changes are still being saved.
During this process, the button displays a spinner and a Saving… message. The behavior applies both to the main header button and the Publish panel.
This prevents a release from being created before recent flow changes or canvas notes have been fully stored.
AI actions now avoid repeated requests
Pages containing several Ask Knowledge Base, AI Prompt, or AI Condition actions now load more efficiently.
Previously, each action could request the same integration, model, or Knowledge Base information separately. These shared requests are now cached and reused for a short period.
A canvas containing multiple AI actions can now request each shared resource once instead of repeating the same request for every action. This reduces unnecessary network activity and improves the experience when opening larger workflows.
Dashboard core upgraded to Angular 15
The Dashboard core has been migrated to Angular 15, including updated Material and CDK components.
Important interface areas have also been adapted to the newer Material MDC system, including:
- History date selection
- Knowledge Base actions
- Data Tables
- Other frequently used dialogs and controls
Unused legacy components have been removed as part of the upgrade. This provides a cleaner technical foundation for future interface improvements and platform stability.
vLLM integration updated
The vLLM integration has been refactored to manage OpenAI-compatible endpoints and models more consistently.
This makes it easier to connect compatible model infrastructure without requiring a completely separate configuration approach.
Widget installation configuration updated
The widget installation script available from the Knowledge Base page now includes the default department ID and chatbot ID.
This gives users a more complete configuration when copying and installing the widget.
Improvements
More accurate Knowledge Base Preview timing
The response time displayed in the Knowledge Base Preview now represents the time experienced by the client while waiting for the result.
This provides a more realistic indication of how long the preview response actually takes from the user’s perspective.
Better Analytics navigation and loading feedback
Analytics pages now provide clearer loading feedback while data is being prepared.
Navigation between Analytics paths has also been improved, and a dedicated button allows users to return to the legacy Analytics area when needed.
More context in the Activities page
The Activities page now includes the Leave Conversation event.
This gives administrators and team managers a clearer record of when a teammate leaves a conversation.
Better visibility across Data Tables and Knowledge Bases
Tooltips have been added to the Knowledge Bases and Data Tables sidebars so users can read the full name of an item when the label is too long for the available space.
More consistent widget controls
Switch buttons in Widget Setup > Advanced now behave consistently with the other switches on the page.
Teammate and interface refinements
Several frequently used areas have received interface improvements, including teammate displays, Analytics loading states, and Data Table action configuration.
Bug fixes
- Fixed an issue where double-clicking Publish could trigger two publishing requests.
- Fixed a shared-data issue that allowed MCP dialogs to modify integration information unexpectedly.
- Fixed reply filters offering operators that were not supported by the server. The editor now shows only valid reply filter operators.
- Existing filters that use unsupported operators are preserved and displayed with a warning instead of being silently changed.
- Fixed an issue where editing an Ask Knowledge Base name or inserting a parameter with Use Knowledge Base name enabled could save
[object FocusEvent]. - Fixed message retention settings not displaying the user-selected value after changing paths.
- Fixed interface inconsistencies in the Data Table action.
- Improved filtering and empty-state handling across affected views.
Together, these updates make Tiledesk more practical for building data-driven AI Agent workflows. Structured data can now live closer to the automation, complex conditions are easier to configure, publishing is safer, and larger projects load with fewer repeated requests.
Explore how Data Tables, AI actions, and conditional logic can work together in a real operational process
