TWIN:DESK

1. Secure user login with Microsoft SSO accounts and OAuth/LDAP protocols, integrated with the customer's Active Directory
2. Automated identity and user-permission management through SCIM 2.0 integration
3. User-role support, including at least a User role for chat, files and personal costs and an Admin role for global configuration, models, auditing and security
4. Dynamic adjustment of function visibility to the user's role, workspace, resource permissions and system configuration
5. Saving, searching, deleting and controlling persistent user preferences and facts remembered by the system
6. Resolving user-memory conflicts, detecting contradictory data and selecting the correct profile version
7. Conversation with the user's digital twin, including persistent memory, a multi-level cognitive profile and automatic knowledge synchronization from work history
8. Exporting digital-twin memory and making it available as a context source for the AI model
9. Exporting the user's complete configuration and digital-twin memory to a formatted .zip file for migration or backup
10. Training the digital twin from dynamic context sources such as Google/Outlook calendars, Gmail/Exchange mail and Slack/Teams messengers
11. Maintaining a transparent learning and summary journal that lets the user review and edit knowledge acquired by the digital twin
12. Isolating user data according to least-privilege principles, with access limited to data assigned to the appropriate organizational group or project
13. Creating workspaces, adding users with WRITE/READ roles and managing access to resources within a workspace

1. Starting new conversations in the correct context of the selected workspace
2. Selecting an LLM according to data sensitivity, task type and applicable policies, with routing to the appropriate providers
3. Changing the LLM used during a session without losing the established context
4. Handling conversation history, attachments, knowledge-source selection and conversation exports in the main chat interface
5. Quickly exporting agent-generated files by dragging and dropping them directly onto the user's computer
6. A Prompt Assistant that supports effective instructions following the CRISPE standard and lets users create, edit and use ready prompts in chat
7. Preserving conversation history with options to export, share and tag conversations
8. Voice chat with speech recognition and voice synthesis
9. Image generation and editing with integrated visual models
10. Consulting several models with one question to compare results and synthesize an answer
11. A split view for parallel comparison of results generated by different models
12. Structured responses compliant with a JSON schema for cloud and local models
13. Executing Python, SQL and other commands directly in the chat window with results presented in the user interface
14. Selecting a work mode: knowledge search, tasks performed by a computer agent or conversation with a digital twin
15. Personalizing the digital twin, including its name, image and multilingual interface and model configuration
16. Advanced DAG-based rollback of agent actions with a visible reversal plan and user confirmation
17. Learning from recurring user decisions and proposing automation after approval
18. A feedback loop for rating model responses and maintaining a rating register to optimize quality

1. File uploads, OCR, knowledge-base management, workspace creation and parent-project management
2. A central RAG database with mechanisms that reduce hallucination risk
3. A document-filtering panel based on metadata within a workspace
4. Advanced OCR that describes images and charts in a form useful to LLMs
5. Use of personal knowledge sources such as local folders, Outlook mail, calendars and Microsoft Teams without loading data into the central RAG database
6. A live view of knowledge sources used in a conversation
7. Knowledge-gap management when context is missing instead of generating uncertain or hallucinated answers
8. A metadata-only graph layer that does not use raw prompts, complete conversation content or private user memory

1. Making Microsoft mail and calendar data available as context sources for AI
2. An independent rich-text editor with AI functions, versioning and real-time collaboration
3. Scheduled execution of scripts and prompts with execution history and status
4. Planning work-time blocks and generating time reports from system activity and calendars
5. Work-planning automation, including morning priority triage and daily activity summaries
6. A task manager with Kanban boards, task statuses and due dates
7. Defining if/then behavioral rules in natural language
8. Calling n8n workflows and integrating with Langflow and flows
9. Controlled task delegation to other agents with approval mechanisms and limits

2026/Q3IN PROGRESS

2026/Q4