Inside your dedicated work platform instance

When you log in to your team's work platform, the main screen opens with the AI chat input field — "How can I help you today?" — as its central element. This is where all interaction with the AI takes place within your isolated infrastructure.

At the top of the screen is a "Select model" dropdown menu.

Important: Model access is strictly managed by the platform administrator. By default, a new user has no models assigned. To start working with AI, the administrator must assign specific models either to the user personally or to a group the user belongs to.

Side Navigation Panel

On the left side of the screen is a slide-out panel containing the main working tools:

• New chat: Creates a blank dialogue session with a clean context window.

Recommendation: Always separate chats by topic — do not mix different subjects in one conversation. The AI model analyses the entire history of the current chat. To maintain quality, start a new chat for each new topic.

• Search: Lets you search across the history of all your dialogues.
• Workspace: Section for creating knowledge bases. A base can be private or public. For public bases, access can be granted to specific users (group-level access is not available here).

Profile and Settings

At the very bottom of the side panel is the current profile icon (e.g. "Test User"). Clicking it opens a menu where you can:

• Sign out of the account.
• Go to the archive (if chats were previously moved there).
• Open the "Settings" window.

The settings window is divided into several tabs:

• General: Change the system language and visual theme.
• Interface: Visually customise the platform to individual preferences.
• Data management: Manage uploaded files — delete outdated documents to free space on the VM.
• Account: Profile information; change username and update password.
• (Administrators only): The administrator's profile menu also contains an extra button to access the platform control panel. It is hidden for regular users.

Each user of your team's work platform interacts in chat with the AI model they choose from their available list. This list is configured by the administrator of your specific instance.

The manager, as the primary administrator, can appoint additional administrators whose responsibility is to directly manage the platform — including granting and configuring access rights to specific AI models for each individual user.

Visual Model Markers

For easier navigation in the model dropdown, the interface uses clear markers. The most important things to watch are the vision icon and each model's description.

Built-in catalog and external providers. Some models are available immediately from the AI-COMFORT catalog, while the organisation can also connect its own OpenAI-compatible APIs such as OpenRouter, GigaChat, YandexGPT, or even a self-hosted model on its own infrastructure.

Why this matters: AI-COMFORT does not lock the customer into a single model supplier. You can use the platform's ready catalog with ruble billing, or connect your own quotas, your provider's free models, Russian services, or local models for your team. Selected external models also receive the corporate work profile: system prompt, reasoning, work-ready temperature, tools, and other platform settings.

Eye icon (👁️) — Vision support. This marker indicates a multimodal AI: the model can "see" and analyse attached images.

Use cases: Invaluable for analysing complex infographics or digitising documents. Simply upload a scanned file, ask the model to convert the image to text (specifying formatting if needed), and the model will return a ready digital version instead of a raw scan.

The message input field includes a range of additional tools for extended interaction with the AI model. The toolbar is split into a left section (attachments and settings) and a right section (voice input).

1. Adding Materials (Plus icon)

Clicking the + icon on the left side of the input field opens a five-option menu for loading context:

• Upload files: Attach local documents for analysis.
• Capture: Take and attach a screenshot.
• Attach web pages: Load data via URL. The system scans the page, extracts text, and makes it available for analysis in chat.
• Attach knowledge: Connect a previously created knowledge base.
• Reference chats: Attach the history of previous dialogues — useful for revisiting past conclusions or deeply analysing old conversations.

2. Integration Management (Diamond icon)

The second icon on the left opens a menu for managing additional capabilities:

• Web search: Disabled by default. When activated, the AI searches the internet for keywords in the query, collects up-to-date information, and bases its answer on those current results.
• Tools: Enables or disables file-creation modules. All modules are active by default. The AI automatically detects which tools are currently enabled and knows which file formats it can generate. Disabling certain options removes the ability to create files in those formats.

3. Tool Reference (Wrench icon)

Next to those is an icon showing a wrench and a number (e.g. 12). Clicking it opens an "Available tools" information window. You cannot enable or disable modules here — it is a reference panel listing all active tools (PDF, DOCX, Excel, presentations, etc.) with a description of what each one generates.

4. Voice Input (Microphone icon)

On the right side of the input field is a microphone icon. It lets you dictate a query by voice: the audio is automatically transcribed to text and instantly inserted into the input box for sending.

Directly in chat mode, the platform can automatically generate charts and diagrams. When the relevant tool is active (enabled by default), the AI visualises data using built-in rendering based on the Mermaid markup language.

The user can request a specific chart type (e.g. flowchart, sequence diagram, Gantt chart or mind map), or let the AI model choose the optimal visual format on its own.

Contextual Analysis

A key advantage is that when building a diagram, the model analyses not only the most recent message but the entire conversation history. While it is recommended to include maximum detail in the drawing request, deep understanding of the full thread allows the AI to create highly accurate, complex, and relevant diagrams — making it an invaluable tool for data analysis, business-process planning, and any complex analytics.

Managing the Generated Diagram

The generated diagram appears directly in the chat window. In the top-right corner of the diagram block there are action icons:

• Download: Save the diagram as an image file to your device.
• Refresh: Re-render the current diagram.
• Copy: Copy the raw Mermaid source code — useful for pasting, storing, or further manual editing in third-party Mermaid-compatible tools.

Directly in dialogue mode you can create ready-made documents. The platform supports generating files in 12 different formats, including text documents (PDF, DOCX, ODT), spreadsheets, and presentations.

Control via Text Requests

No separate buttons or menu switches are needed to create a file. Simply describe the task in plain text — for example, ask the AI to draft a contract template in PDF format. The model recognises the command, activates the appropriate tool, and returns a download link for the finished document.

Flexible Editing and Format Conversion

If the result needs changes or doesn't fully match expectations, refinement happens right in the same chat. At any point you can send a follow-up message asking to:

• Regenerate the file with updated details;
• Change the structure or content of the document;
• Produce the same document in a different format (e.g. convert the generated PDF to DOCX, or request several formats at once).

The model instantly adapts the result and generates new files, provided the document-creation tools are active in the current integration settings.

Beyond text and document generation, the chat mode includes a powerful feature for writing and instantly executing code scripts. The AI model generates code based on the request and embeds it in its response as an interactive block.

To test the code, there is no need to download files or use external editors — simply click the "Run" button in the block's top menu and the script executes directly inside the platform interface.

Pyodide Execution Environment

Code runs in a secure, isolated environment directly in the browser using Pyodide (a Python interpreter compiled to WebAssembly). Because computations happen client-side, the environment has certain resource limits and is not intended for heavy server-side programmes. However, it is ideal for quick calculations, algorithms, and data processing.

Why This Matters: Mathematical Accuracy

The primary purpose of built-in scripts is to ensure 100% mathematical accuracy. The architecture of any language model means it can make errors in complex mental arithmetic and lose logical thread in multi-step formulas.

Recommendation: Instead of asking the AI to compute complex values directly, ask it to write a script for the calculation.

How It Works in Practice

1. 1State the task in the query (e.g. calculate the cost of goods).
2. 2The model analyses the full conversation context, extracts the necessary input data, and substitutes it into the code variables.
3. 3The model outputs the finished script.
4. 4After clicking "Run", the system executes the code according to mathematical rules and returns a precisely calculated result (STDOUT/STDERR) in chat.

Beyond numerical calculations, Pyodide also allows generating small graphical models and visualisations (e.g. charts) when the task requires a visual representation of data.

Directly in the chat interface you can create full HTML pages. The AI can deliver results in two formats: generate a ready HTML file for download, or output the source code directly in the dialogue window as an interactive text block.

Preview Mode

When code is output directly in chat, it can be tested immediately without any external browser or editor. The block's top menu includes a "Preview" button.

Clicking it splits the interface in two: the dialogue and source code stay on the left, and a fully rendered visualisation of the generated HTML page appears in the right panel.

Key Use Cases

This feature significantly speeds up web markup work and is well suited for:

• Quickly creating single-page sites (landing pages).
• Designing and visually planning individual interface components.
• Preparing interactive promotional materials.
• Rapid debugging, testing, and refining existing HTML code in real time.

The platform integrates support for vision AI models (marked with the eye icon) that can recognise and deeply analyse not only text but also complex graphic files. Vision is available both in the built-in AI-COMFORT catalog and in connected external providers whenever a specific model supports it.

Key Use Cases

• Document digitisation: Quickly extract text from scanned files, charts, or document photos for further editing.
• Complex graphics analysis: Detailed analysis of objects, composition, and context in any uploaded image.
• Automatic description generation: Create detailed text descriptions, metadata, or narratives based on an uploaded picture.

Visual analysis saves significant time: there is no longer any need to manually describe complex graphical content to give the AI a task. Simply attach an image in the chat window and ask a question — the model will examine the details and return an accurate text result.

The platform allows creating high-quality images directly in the chat interface. For this purpose a specialised paid model — Nano Banana 2 (Gemini 3.1 Pro) — is integrated, known for its excellent cost-to-quality ratio.

How to Work with the Model

Interaction happens in the usual dialogue mode, which allows flexible control over visual content creation:

• Generation from description: Create images from scratch based on a text prompt.
• Using references: The model can analyse attached images and generate new ones based on the visual context.
• Flexible format control: Specify the desired aspect ratio directly in the request or select it below the generated image (e.g. 16:9, 1:1, or 9:16).

Pricing and Access

This feature is paid. Billing is per generated image (approximately 10 RUB per generation).

Note: Because using the model incurs additional costs, access to image generation is restricted by default. The platform administrator assigns this model only to users or groups who genuinely need it for their work (e.g. for creating promotional materials).

One of the platform's key features is creating and managing knowledge bases. This tool lets you accumulate large volumes of data and use them as intelligent context for AI model interactions.

Creating and Configuring Access

A new knowledge base is created in the "Workspace" section. When setting it up, the following parameters are defined:

• Name and description: Identify the purpose and contents of the base.
• Access type: The base can be private (creator only) or public (accessible to colleagues inside the platform).
• Access management: For public bases, specific users can be granted access. The base will automatically appear in the relevant menu section for those users.

Supported Formats and Vectorisation

The platform supports uploading and vectorising virtually all popular document formats, turning scattered files into a single structured database.

Supported formats:

• Text documents: PDF, DOC, DOCX, ODT, RTF, TXT, EPUB.
• Spreadsheets: XLS, XLSX, ODS, CSV.
• Presentations: PPT, PPTX, ODP.
• Structured text and code: MD (Markdown), XML, HTML.

Working with scans (Built-in OCR): If a PDF consisting of image scans is uploaded, the platform uses its built-in optical recognition to extract text from images, convert it to digital form, and store it in the knowledge base. Despite minor digitisation inaccuracies, AI models compensate effectively through their understanding of the document's overall context.

Storage Technology: RAG and Vector Search

All information in knowledge bases is stored in vectorised form using RAG (Retrieval-Augmented Generation) technology — a modern data format with two key advantages:

1. 1Scalability: The system instantly locates the right fragments even in massive archives spanning thousands of pages.
2. 2Resource efficiency: When querying the knowledge base, the AI does not read all documents in full. It finds only relevant sections by keywords and semantic meaning, making processing fast and minimising token usage.

Using in Chat

To interact with a base, click the "+" (Add) icon in the chat window and select "Attach knowledge".

Once the base is connected, you can give the AI model any task involving that information:

• Search for specific facts, dates, or conditions in the archives.
• Analyse data and prepare summary reports based on documents.
• Use the base as a persistent context foundation for responses.

Use cases range from quick searches through internal regulations to deep analysis of complex technical documentation.

Interaction on the platform is not limited to a dialogue with a single AI model. The system provides flexible options for managing which algorithms process requests within a single session.

Switching Models in the Current Chat

In an active dialogue you can change the AI model at any time via the top menu. When a new model is selected, all subsequent messages are processed by it. This allows easy switching between algorithms without losing the accumulated context of the current conversation.

Running Multiple Models in Parallel (Multi-chat)

Multiple different models can be connected to a single dialogue simultaneously. To do this, use the "+" (Add model) button next to the current model name in the top bar.

How parallel mode works:

• When a message is sent, the query is automatically and simultaneously sent to all selected models.
• Responses are generated in parallel and displayed side by side on the same screen (in a split-window format).

Use Cases

This feature is invaluable for deep analysis of important information. It lets you simultaneously obtain different "perspectives", compare the reasoning logic or creativity of several models in a single window — without creating parallel chats and manually duplicating queries.