Feature Releases

Risk Analysis Report: Updates to Auditing Logic & Risk Evaluation

• Where is this: AI Lifecycle Optimiser — Risk & Governance (Available to Tool Owners and Admins in workspaces where the feature is permitted)

• Instruction Disclaimer Integration: The revised report now accounts for text edit (i.e. disclaimers additions) made in the User Instruction field when assessing risk compliance.

• OWASP Grounding Framework: Integrated the OWASP (Open Worldwide Application Security Project) Top 10 for LLMs into core grounding data to enable more robust security and vulnerability evaluations during tool audits.

• Updated File Handling Policy: Refined risk recommendations regarding file uploads. Rather than discouraging File Chat usage, the assessment engine now guides tool creators and end users on safe document upload practices. 

Transition to File Chat 2.0 – Phase 1 Release
We have completed the transition from our legacy CKS File Chat system to File Chat 2.0 (Beta). This transition shifts file processing from legacy retrieval methods to direct-to-LLM native context ingestion, delivering faster response times, higher AI comprehension, and expanded capacity. 

• Beyond tool level usage, you can now chat with files directly at the base AI Assistant chat level without launching a Capability Tool, as well as upload additional documents mid-conversation during active chat sessions.

• System limits have expanded significantly across the platform. You can now upload up to 20 files per container with an individual file size limit of 32 MB per file (up from 10 files and 5 MB). Supported formats include standard documents (PDF, DOCX, TXT, MD), data sheets (XLSX, PPTX, CSV, JSON), and images (PNG, JPG, JPEG, WEBP).

Supported models across your workspace now carry the [With Uploads] tag:

• OpenAI | GPT-4o | [With Uploads]

• OpenAI | GPT-4.1 | [With Uploads]

Action Required: Transition Checklist

Because File Chat 2.0 uses direct model processing, previous file indexes cannot be fully carried over. Please complete the steps below for your role:

For Tool Creators

• Re-upload Tool Knowledge: Access the Tool Knowledge File Guide to verify your tools, re-upload necessary files, and test output accuracy. 

For All Users

• Start Fresh Sessions: Previous chats using legacy file uploads will no longer process files. Launch a new chat session to use new file chat models.

• Re-upload User Container Files: Re-attach your documents directly within your new chat window. Refer to the File Chat in Conversation Guide for step-by-step instructions.

• Review Phase 1 Guidelines: Read the File Chat 2.0 User Guide to understand current file limits, mid-chat uploads, and model-switching behavior.
  
  

Need Help?

If you experience unexpected behaviour, consult our Step-by-Step Bug Reporting Guide to submit a report.

(Note: Wider model support, tool chaining with files, and enhanced model-switching warnings are actively in development for upcoming patches.)

Usage Logs: Advanced Token Visibility

• Enhanced Insights: We have expanded token tracking to include consumptions from the Tool Wizard and AI Lifecycle Optimiser. This provides Workspace Admins with a view of usage patterns associated with advanced tool building and auditing.

• Privacy Note: This comprehensive report remains exclusive to Workspace Admins to maintain organisational security and data privacy.

Web Search Optimisation 

• Web Search is now available both with base chats and alongside chats with tools, enabling users to retrieve up-to-date information that extends beyond the Large Language Model (LLM)'s data cutoff. Note that web search success is query-dependent. 

Table Export in Chat Conversations

• Direct Export: Users can now export tables generated within a chat as standalone files (XLSX or CSV).

• How to use: Simply hover over any table within a conversation to reveal the export action buttons.

Agentic UI/UX Enhancements (Phase 1)

• Interface Optimisation: The Chat Cover Page and Input Message fields have been redesigned to support "Agentic" workflows.

• New Navigation: This update introduces the Agents and Modes menus—a centralised hub for accessing specialised tools like Web Search and ToolSense.

• Permissions: Visibility of the Chat Page and specific modes (like ToolSense) are governed by toolkit permissions and workspace-level settings.

Transcription Mode (Phase 1)

• Dual-Method Conversion: Users can now convert voice-to-text using two distinct methods based on workspace configuration:

  1. Live Transcription (Real-Time STT): Audio is transcribed immediately as the user speaks.

  2. (Batch) Transcription Mode (Post-Processed): High-accuracy transcription performed after the recording is finalised and uploaded.

• For further enquiries or activation, please contact your Account Manager.

Usage Logs: Token Counting

• For Admins: A new 'Token Usage Report' provides insights into which LLMs are being utilised and by which users.

• For All Users: A 'Context Token Count' and 'Total Tokens Used' artefact now appear beneath the chat box, allowing real-time monitoring of session consumption.

Tool Wizard & AI Lifecycle Optimiser

• Tool Wizard: A powerful new module designed to streamline the end-to-end development of custom AI tools.

• AI Lifecycle Optimiser: A dedicated auditing tool for Tool Owners and Admins to evaluate performance across Risk & Governance, Transparency, and Cost.

  • Update: Chat history and logs are now fully enabled for both modules.

  • Access Control: These must be enabled by a System Admin at the workspace level and can then be toggled by Facilitators for specific toolkits.

Voice Chat Integration (Phase 1)

• Real-Time Speech-to-Speech: Introduced the initial API integration for real-time voice conversations, accessible via the Chat UI Cover Page.

  • Constraints: Currently supports OpenAI models only.
    To ensure performance efficiency, access is restricted to selected tools and requires System Admin activation.

Role-Based Tool Creation Limits

• Workspace Governance: Workspace owners can now set a maximum limit on how many tools a participant can create or own within a specific toolkit.

  • Dynamic Enforcement: Note that limits are role-specific. For example, a user restricted as a "Participant" in one toolkit may still have full "Facilitator" creation rights in another.

  • Action: Contact your Account Manager to implement these workspace-level restrictions.