Before the end of the year, we are shipping one of our most complete updates so far. This release brings new AI model options, better WhatsApp controls, major performance improvements, and several enhancements that make day‑to‑day operations smoother and more reliable.
Ask Knowledge Base Now Supports More LLMs
We’ve expanded support for the Ask Knowledge Base action with a wide range of the most advanced language models available today. You now have full control over which LLM powers your AI Assistant, allowing you to optimize for speed, reasoning depth, or cost depending on your use case.
Available models include:
-
OpenAI: GPT‑5.2, GPT‑5.1
-
Anthropic: Claude 4.5 (Opus, Sonnet, Haiku), Claude Opus 4.1
-
Google: Gemini 2.5 Flash Light, Gemini 3 Pro
-
DeepSeek
-
Cohere: Command RFB, Command R+
-
Grok: Qwen 3
This flexibility lets you fine‑tune how answers are retrieved and prioritized, from fast and lightweight responses to deep and highly contextual ones. Model selection is available directly from the AI Settings panel.
WhatsApp Gets More Practical for Daily Operations
WhatsApp workflows continue to improve with features focused on real operational needs:
-
You can now send a WhatsApp message to a single contact directly from the Contact List page, without creating a broadcast or campaign
-
The Sent Messages list now shows a preview of the message template, making it easier to review what was actually sent
-
Fixed an issue affecting PDF template selection in WhatsApp send actions
These improvements make WhatsApp interactions faster, clearer, and easier to manage.
Faster Platform with Smart Caching
We introduced a new caching strategy to improve performance across the platform:
-
Project data, project users, and departments are now cached to reduce redundant HTTP requests
-
Cache invalidation is automatically triggered when data changes, ensuring consistency while keeping performance high
In addition, we fixed a REST call that could generate a 504 error when loading conversation details, improving stability in high‑traffic environments.
Monitor and History: Better Handling of Unassigned Conversations
Managing abandoned or unassigned chats is now easier:
-
The Monitor page includes a dedicated area for unassigned conversations, allowing agents to pick them up directly
-
In the History section, Advanced Search filters now help you quickly find unassigned conversations
This improves visibility and ensures no conversation is left behind.
Ask KB Controls and AI Settings Improvements
Several enhancements give you more transparency and control over how answers are generated:
-
“Chunks only” and “Re‑ranking” options are now available in AI Settings
-
Chunks only lets you return raw source content
-
Re‑ranking intelligently prioritizes the most relevant results
-
-
The namespace name is now visible in Ask Knowledge Base v2
-
The Re‑ranking option is correctly displayed in the UI
-
AI action templates now include a settings summary, making it easier to review configurations at a glance
-
The
replyTotag is now visible in preview mode for AI Condition and Ask Knowledge Base actions
Usability Improvements
-
You can now create a new contact directly from the Contact List
-
Improved full‑text search on the History page and Non‑Real‑Time Conversations page
-
Global variables created in Design Studio are now correctly included in the variables list
Bug Fixes and Stability Improvements
This release also includes multiple fixes that improve reliability:
-
Fixed incorrect button colors when visibility control is enabled
-
Fixed iframe URL display issues in action playback
-
Fixed layout issues when changing iframe height
-
Fixed connector rendering by introducing a proper rendering queue
-
Fixed AI Agent duplication not working
-
Fixed sidebar items not activating correctly when query parameters are present
-
Fixed issues loading conversation history when certain query parameters are used
-
Fixed a bug where deleted groups still appeared in department load balancing
-
Fixed project subscription issues when the last_project object was not a project_user
These improvements move Tiledesk even closer to prompt‑driven automation, where AI is not just answering questions but operating reliably at scale. With better model choice, faster performance, and stronger operational controls, building intelligent workflows becomes simpler and more powerful with every release.
