allclouds.pl / SAIE products / TWIN:DESK

TWIN:DESK - AI assistant and organisational knowledge base

TWIN:DESK is an AI assistant and knowledge base that answers from your organisation’s documents and cites the source. The browser application turns a basic chat into an everyday workspace for employees, teams and entire organisations - with institutional memory, data control and integration with administrative processes

Book a demonstrationGet the PDF brochure
×5AI engines in the Model Council
RAGanswers with source citations
MCPopen integration standard
3deployment options
Four pillars

More than a chat window

Before the user asks anything, TWIN:DESK suggests a summary of today's meetings, a daily plan based on calendars, email and tasks, or a review of recent emails

01
AI interactionchat as an intelligent entry point: model switching based on task confidentiality, context control, multimodality and proactive suggestions
02
Institutional knowledgedocuments, knowledge bases and case law as working material (RAG), with source citations
03
Operational worknotes, tasks, automations, reports and calendars in one place
04
OversightRBAC, audits, data classification, integration and model control
TWIN:DESK · New conversation
Good morning, AnnaWorkspace: Finance Department · status: confidential data
Suggestion: you have 3 meetings today. Prepare a summary and daily plan from your calendar, email and tasks?
Ask the assistant or drag a file…Prepare a summary and daily plan
Attachmentsconversation files
Councilmultiple models at once
Daily routineworking mode
Agent selectiondomain assistants
“Computer” modeautonomous tasks
Voice commandsmultimodal input
Suggestionsmeetings, tasks, summaries
Security statusworkspace confidentiality
Input panel: eight elements instead of an empty chat window
Domain assistants

Specialised assistants instead of a generic chatbot

A rigorously configured analyst is assigned to a specific department or case. The workspace receives its own knowledge bases, system tools, automation workflows (such as video analysis or invoice processing) and, optionally, image generation and analysis, memory and chat history

How a domain assistant is built
Identity and access · avatar, name, description, tags, base engine; · who on the team has access · Non-negotiable rules · required file search, internal · files as the source of truth, · no fabricated content · Answer standard · communication style and source citations · Assistant · domain assistant · processing flow · 1 · question assessment · 2 · retrieval · 3 · analysis · 4 · answer based on files · Own knowledge bases · System tools · Automation workflows · e.g. video analysis, invoice processing · Images, memory, chat history · optional · ASSISTANT CONFIGURATION · THE WORKSPACE RECEIVES
Working with company knowledge (RAG)

Company knowledge that AI does not confuse with the archive

The system understands document metadata, versions, status and validity periods - distinguishing public knowledge from confidential data, and archives from current procedures. Every answer cites a specific file passage. The Knowledge Gap Register captures unanswered questions, while the MCP module directs conversations to processes that perform real operations in connected systems

QuestionWho approves an invoice over PLN 50,000?
skipped · archived
Invoice handling procedure v2.1
status: archivedvalid until March 2026
selected · current
Invoice handling procedure v3.0
status: currentfrom 1 April 2026confidentiality: internalowner: Finance Department
The treasurer approves invoices over PLN 50,000 within 3 working days11 · Invoice handling procedure v3.0, § 4(2) · current version
Office work

AI in office work

An AI assistant in a company or public office is most useful in everyday office work. TWIN:DESK draws on the organisation’s knowledge base and cites the source of every answer:

Letters and documentsan editor with AI features, named versions and real-time collaboration
Email and calendarthe local agent reads Outlook email and the calendar without copying them to the central knowledge base, and prepares a daily plan
Meeting notesa recording becomes text with labelled speakers, and agreed actions become tasks with an owner and a deadline - new in the SAIE Pantera release
Summaries and translationslong documents condensed to their key points with a source citation, and translation on local models with a terminology glossary
Reports and figuresamounts are calculated by a decimal calculator, and every result has a calculation card
Mechanisms

Capabilities beyond an ordinary chat

Progressive disclosuresimple chat for beginners, a full terminal and automations for experts
Model Councilthe same question sent to up to 5 AI engines, producing one final answer
Confidentiality at uploadconfidential files permanently blocked from open models
Twin:Coina virtual budgeting unit instead of token price lists
Custom request rulesenforce language, block topics, hand a case over to an operator
Integrations

The integration layer classifies documents, extracts data, summarises and works with the knowledge base, while communication with source systems passes through controlled orchestration. Further use cases are introduced in stages without losing control of data

Integrations - through controlled orchestration
EZD RP · EZD PUW · e-Doręczenia (Polish e-delivery) · ePUAP · Polish Trusted Profile · KSeF 2.0 · Domain systems · Asseco · Comarch · Sygnity · Sputnik Software · Microsoft 365 · Outlook · Teams · Automations · n8n · Langflow · TWIN:DESK · integration layer · classification · documents · extraction · data and summaries · orchestration · controlled, in stages · PUBLIC-SECTOR SYSTEMS · ORGANISATIONAL SYSTEMS AND TOOLS
FEATURE CATALOGUE

58 TWIN:DESK features

The complete workspace feature list: access and profiles, chat and models, knowledge and data, agents and automations. Including 10 new in SAIE Pantera

Access management and profiling14
  1. Guided onboarding: a first sign-in guide matched to the user's role and permissions, progress visible to the administrator and a one-click restart
  2. Secure user sign-in with Microsoft SSO accounts and OAuth/LDAP protocols, integrated with the customer’s Active Directory
  3. Automatic management of user identities and permissions through integration with the SCIM 2.0 protocol
  4. Support for user roles, including at least a User role for working with chat, files and one’s own costs, and an Admin role for global configuration, models, audit and security
  5. Dynamic adjustment of feature visibility to the user’s role, the workspace, resource permissions and the system configuration
  6. Saving, searching, deleting and controlling the user’s standing preferences and the facts the system remembers
  7. Resolving conflicts in user memory, detecting contradictions in data and choosing the correct version of the profile
  8. Conversation with the user’s digital twin, including persistent memory, a multi-level cognitive profile and automatic synchronisation of knowledge from the work history
  9. Export of the digital twin’s memory and sharing it as a context source for an AI model
  10. Export of the full user configuration and the digital twin’s memory to a formatted .zip file for migration or backup
  11. Training the digital twin on dynamic context sources such as Google/Outlook calendars, Gmail/Exchange email and Slack/Teams messaging
  12. A transparent learning and summary log that lets the user review and edit the knowledge the digital twin acquires
  13. Isolation of user data under the principle of least privilege, with access only to data assigned to the relevant organisational group or project
  14. Creating workspaces, adding users with WRITE/READ roles and managing access to resources within a workspace
Chat and work with an AI model21
  1. Model and workspace configuration assistant: needs described in plain language turned into a system prompt, suggestions and parameters, with a before-and-after comparison
  2. Translation workspace on local models (CPU or GPU): segment editor, translation memory, terminology glossary and DOCX export with the layout intact
  3. Verified calculations: a decimal calculator computes amounts, every result has a calculation card, and a report with a mismatched figure is not published
  4. Starting new conversations in the correct context of the selected workspace
  5. Choice of LLM depending on data sensitivity, the nature of the task and applicable policies, with routing to the appropriate providers
  6. Switching the LLM during a session without losing the context built up so far
  7. Conversation history, attachments, knowledge source selection and conversation export in the main chat interface
  8. Quick drag-and-drop export of files generated by the agent directly to the user’s computer
  9. A Prompt Assistant that helps write effective instructions in line with the CRISPE standard, plus creating, editing and using ready-made prompts in chat
  10. Keeping conversation history with the option to export, share and tag conversations
  11. Voice chat with speech recognition and speech synthesis
  12. Image generation and editing with integrated visual models
  13. Consulting several models on one question at the same time to compare results and synthesise an answer
  14. A split view that makes it easier to compare results from different models side by side
  15. Structured answers conforming to a JSON schema for cloud and local models
  16. Running Python, SQL and other commands directly in the chat window, with the results shown in the user interface
  17. Choice of working mode: knowledge search, tasks carried out by a computer agent, or a conversation with the digital twin
  18. Personalising the digital twin, including giving it a name and an image, and configuring multilingual support for the interface and the model
  19. Advanced rollback of agent actions based on a DAG, with a visible rollback plan and user confirmation
  20. Learning from the user’s repeated decisions and proposing automations that run once approved
  21. A feedback loop for rating model answers and keeping a ratings register to optimise quality
Knowledge and data management11
  1. Live and uploaded recording transcription with speaker diarisation: text with labelled speakers and timestamps, AI analysis, findings and tasks, export to PDF, Word, TXT and JSON
  2. Agentic RAG and TWIN:INDEX: multi-step search with completeness checks and follow-up queries, a tree of long documents (PageIndex) and diagnostics for every round
  3. Named document versions: work stages saved under their own names, preview without changing the current content, and restore without losing history
  4. File upload, OCR, knowledge base management, and creation of workspaces and parent projects
  5. A central RAG knowledge base with mechanisms that reduce the risk of hallucinations
  6. A panel for filtering documents by metadata within a given workspace
  7. Advanced OCR that describes images and charts in a way that is useful for LLMs
  8. Use of personal knowledge sources such as local folders, Outlook email, the calendar and Microsoft Teams, without loading data into the central RAG knowledge base
  9. A live preview of the knowledge sources used in the conversation
  10. Knowledge gap management when context is missing, instead of generating uncertain answers or hallucinations
  11. A metadata-only graph layer that does not use raw prompts, full conversation content or users’ private memory
Agents and automation12
  1. Consolidated report builder driven by a published plan (DAG) with the status of every step, and an agentic loop with a plan to approve and a run history
  2. Sync of Outlook Calendar meetings into Asana tasks via the local Twin Desk Agent, with a summary and no duplicates
  3. The “AI usage” panel for local agents (e.g. Codex): tokens, calls and API equivalent for the current and three previous months, fed into the cost report
  4. Sharing data from Microsoft email and the calendar as context sources for AI
  5. A standalone rich text editor with AI features, versioning and real-time collaboration
  6. Scheduled running of scripts and prompts with a run history and status
  7. Planning blocks of working time and generating time reports based on system activity and the calendar
  8. Work-planning automations, including a morning priority triage and a daily summary of activities
  9. A task manager with Kanban boards, task statuses and due dates
  10. Defining if/then behavioural rules in natural language
  11. Triggering n8n workflows and integration with Langflow and flows
  12. Controlled delegation of tasks to other agents with approval and limit mechanisms
DEVELOPMENT ROADMAP

What is coming to TWIN:DESK in the next quarters

32 roadmap items through Q4 2027. The 2026/Q3 features already work in SAIE Pantera

Full SAIE development roadmap

2026/Q4IN PROGRESSSAIE Pantera extension

TD-535Automatic knowledge base

A knowledge base populated with the results of research on a specified topic

TD-536Proactive assistant

A local AI assistant that observes your work and proactively helps with relevant knowledge, suggestions and proposals for completing the task

2027/Q1SAIE TYGRYS

TD-425Knowledge Authority Layer

In large organizations, documents vary in importance, freshness, and scope, so AI responses must select the right sources based on the user's role and the context of the matter

TD-426Knowledge Graph RAG

Traditional context search loses the relationships between documents, concepts, and agent outputs, resulting in fragmented answers that are difficult to justify

TD-427AI Tool Studio

Building integrations and interface components usually requires development work, which delays the rollout of new capabilities for business users

TD-430Federated TwinDesk Cluster

Shared infrastructure should, when needed, allow individual components to be separated and used by other teams or in other situations - including emergencies - and easily reintegrated afterwards

Questions about TWIN:DESK

What if an employee does not know how to “talk” to AI?

TWIN:DESK includes a Prompt Assistant that helps formulate instructions in line with the CRISPE-CDF prompt engineering standard. For public offices we have prepared course 09 Official: a five-step working method and five questions before pasting a document. For public offices we have prepared course 09 Official: a five-step working method and five questions before pasting a document

MechanismsCourse 09 Official

How can you prevent specific files from leaking at the point of upload?

A document can be marked confidential immediately. The system then permanently blocks its use in conversations with external cloud AI models

Feature catalogue

Is TWIN:DESK just a text window?

No. It is a multimodal AI workspace supporting voice commands, attachment analysis and “computer mode”, where AI agents perform tasks on the user's behalf

More than a chat window

How does the interface adapt to different experience levels?

It reveals features progressively: beginners see a simple chat, while experts have full access to a terminal and workflow automation. At first sign-in a role-aware guide leads the way, and the administrator sees progress and can restart it

More than a chat window

Which Polish e-government systems does the platform integrate with?

EZD RP and EZD PUW, e-Doręczenia, KSeF 2.0 and ePUAP authentication

Integrations

How is TWIN:DESK different from another AI chat?

It is an everyday workspace that builds institutional memory: it securely handles internal documents and knowledge bases (RAG) and includes tools for automating tasks and processes

More than a chat windowCompany knowledge (RAG)

How does the Model Council improve work quality?

A question goes to up to five selected models in parallel. The platform collects and compares their responses, then prepares one final answer - with greater confidence in the result

Mechanisms

How can you control employees' AI usage costs?

The virtual accounting unit Twin:Coin hides complex token price lists and makes it easier to report and limit team costs

Feature catalogue

Does the assistant suggest actions itself, or does it need instructions?

It can work proactively: using email, task management and calendar integrations to prepare daily plans or meeting summaries

Mechanisms

What happens when AI lacks the data needed to answer?

Instead of inventing an answer, TWIN:DESK records the gap in the Knowledge Gap Register. Managers get a concrete indication of which procedures and instructions need updating

Company knowledge (RAG)

Can an assistant be limited to one department's processes?

Yes. Dedicated workspaces provide their own avatar, system-prompt guidelines and assigned tools, such as a code interpreter for the IT department

Domain assistants

What is a Cognitive Digital Twin?

A personal AI agent that learns how a user works and uses their knowledge, documents and activity history. It helps automate repetitive tasks and prepare recommendations - within the oversight boundaries set by the organisation

Mechanisms

What is the local agent in TWIN:DESK for?

It enables use of the user's personal knowledge sources: local folders, Outlook email, Microsoft Teams and calendars. AI works with them locally, without copying sensitive data into a central RAG knowledge base. The agent also moves meetings from Outlook Calendar into Asana tasks and shows, in the “AI usage” panel, the tokens and API equivalent of local tools such as Codex

Company knowledge (RAG)

How is TWIN:DESK different from ChatGPT Enterprise and Copilot?

ChatGPT Enterprise and Microsoft 365 Copilot run in the vendor’s cloud. TWIN:DESK can run as SaaS, on-premises or air-gapped, answers from your organisation’s documents with source citations, matches the model to the confidentiality of the task - including several models at once in the Model Council - and integrates with Polish systems such as EZD RP and KSeF 2.0

Deployment and securityChatGPT Enterprise, Copilot or your own AI

Does TWIN:DESK work with EZD RP and SharePoint?

Yes. TWIN:DESK connects to EZD RP and EZD PUW through its integration layer, and documents from SharePoint and Microsoft 365 can serve as a knowledge source for assistants - with user permissions preserved

Integrations

How much does TWIN:DESK cost?

The price depends on the number of users, the deployment option (SaaS, on-premises or air-gapped) and the scope of integrations. Model usage is accounted for in Twin:Coin, with limits for teams. The Offer Builder prepares an initial quote. Public offices whose leaders join the LIDER:AI Academy receive 5 free accounts for 2027. Public offices whose leaders join the LIDER:AI Academy receive 5 free accounts for 2027

Offer BuilderLIDER:AI Academy

Can TWIN:DESK turn a meeting recording into notes and tasks?

Yes. The transcription module takes a live or uploaded recording and turns it into text with labelled speakers and timestamps. AI analysis draws findings and tasks with an owner and a deadline from the conversation, and the result can be exported to PDF, Word, TXT and JSON or saved to the knowledge base

Video: recording transcription

Can TWIN:DESK translate a document without sending it to the cloud?

Yes. The translation workspace runs on local models (CPU or GPU). The translator works in a segment editor with a translation memory and the organisation's glossary, the DOCX file comes back with its layout intact, and the finished translation goes to the knowledge base in one click

Video: translation workspace

What does TWIN:DESK do when the documents it finds answer only part of the question?

It runs Agentic RAG. The mechanism judges what is missing from the material it has gathered, creates follow-up queries and searches again in another round, while TWIN:INDEX guides the search through the chapters of long documents. Diagnostics show every round, query and source, so the answer can be checked

Video: Agentic RAG

Are the amounts in reports calculated by the language model?

No. The model interprets the task while a dedicated decimal calculator computes the amounts. Every result has a calculation card with the arguments, the rounding rule and the engine version, and a report whose figure does not match the card will not be published

Video: verified calculations

Can a TWIN:DESK agent carry out a task from start to finish?

Yes, under human control. In the agentic loop the agent proposes a plan that can be checked and approved before it runs, then consults the knowledge base and calls tools itself, for example to create a Word document. The run history shows every step and the source of each decision. Tasks with a fixed course run from a published plan (DAG)

Video: agentic loop

How is an AI agent different from an AI assistant?

An AI assistant answers questions, prepares content and suggests what to do next, while a person decides on every step. An AI agent plans and carries out the next steps of a task itself, using tools and systems. In TWIN:DESK both work in one environment: domain assistants support everyday work, and the agent takes over a whole task, such as a consolidated report, following a plan approved by a person. More in the AI Glossary: AI agent

264 All questions · in 9 categories
NEXT

What else to explore

You are here:SAIETWIN:DESKPROXY:AICDF methodologyDeployment and securityDevelopment roadmap
TWIN:DESKPROXY:AI
Next step
PROXY:AI

AI control gateway: policies, Zero-Code Switch, WORM records and 57 features

Explore PROXY:AI
SAIEThree layers, one AI workspace - the problem, the solution and who it is for
CDF methodologyAI deployment step by step: initial assessment, phases F0-F6, oversight and CogOps
Deployment and securitySaaS, On-Premises, Air-Gap, platform architecture and seven ISO standards
Development roadmap112 roadmap items across six quarters, with the option to submit your own needs

https://www.allclouds.pl/en/produkty-saie/twin-desk/