Workspace

A model that helps do the work, not just answer

An AI model in the platform can act as an agent: interpret a task, select suitable available tools and connect their results. It can work with your context, team materials and connected services. Specific actions depend on the model, settings and access rights.

Capabilities in combination

What matters is how the tools work together

In one chat, an employee can find a rule in a knowledge base, apply it to a task, calculate figures and prepare a document without manually moving context between windows.

‘Find our returns policy, compare it with this claim and draft a response in DOCX.’

The model uses accessible materials; an employee reviews the file before sending it.

Context

01.01

Knowledge bases

Put policies, instructions and project materials into personal or shared knowledge bases. Employees can ask questions in plain language.

‘What documents does our returns policy require?’

01.02

Platform MCP server

The platform has its own MCP server, letting you connect other agent systems and MCP-compatible applications to your company knowledge bases. An external agent can find bases and files, run semantic document search and read available materials.

For example, an external agent can check a draft document against a policy in your platform's knowledge base.

01.03

Access follows user permissions

An external agent can access only the knowledge bases and files permitted to the user who connects it. The MCP server is read-only: it cannot create, edit or delete materials.

Connections use the platform user's personal API key. Owners can find the MCP server address and guidance in the Access Credentials card on their dashboard home page.

01.04

Personal memory and notes

Memory belongs to the user account and can be used across chats. Ask the model to save durable work context or preferences; notes hold separate materials that you can revisit and develop with the model.

‘Remember that our weekly report must show overdue tasks separately.’

01.05

Chats and channels

Handle individual tasks in chat and discuss projects with colleagues in channels; keep the history and materials in the team workspace.

‘Summarize the decisions we made in the project channel.’

How an agent organizes work

The platform supports recurring workflows, parallel work and tracking action sequences. These capabilities help a model move from a request to execution using the tools available to it.

02.01

User skills

Create skills with instructions for recurring work, such as reports, document reviews or a preferred format. The model sees the names and descriptions of skills connected to the chat and can load their full instructions when needed.

A skill can hold the rules for a weekly report while the current data comes from your request. Instructions can be reused and shared with colleagues.

02.02

Parallel subagent work

At your request, the model can delegate independent parts of a task to several subagents. They work in parallel with available tools, and their results return to the main chat.

‘Have subagents review different sections of these materials in parallel, then combine their findings.’

02.03

A checklist inside the chat

The model can plan a task with multiple steps, show the current step and mark completed items. The list remains part of the work in that chat and helps you follow execution.

‘Plan the report preparation and mark each step as you complete it.’

02.04

Native tool calling

For models that support tool calling, the platform provides available tools and returns their results to the working conversation. The model can find materials, retrieve data, calculate figures and create documents using permitted functions.

02.05

Context tailored to the model

For built-in text models and owner-connected models, the compaction threshold accounts for the specified context window. In a long conversation, older history becomes a summary while recent messages remain in the working context.

The summary helps preserve work context, decisions and unfinished steps so the task can continue after compaction.

Actions

03.01

Documents and files

Discuss source data with a model and prepare a PDF report, DOCX document, XLSX spreadsheet or presentation in chat.

‘Prepare a monthly sales spreadsheet and a short management summary from these figures.’

03.02

Calendar and automations

Describe a task and its execution time in ordinary language: the model can create a one-time or recurring automation and run it using available tools. Meetings and reminders can also be created from chat.

‘Every Monday at 9 a.m., gather data from connected services and prepare a brief management summary.’

03.03

External systems

Connectors bring data from marketplaces, CRM, inventory, analytics and other systems into one work chat.

‘Compare this SKU's stock in Ozon and MoySklad.’

03.04

Search and computation

For current facts and precise arithmetic, the model can use web search and a code interpreter instead of relying only on memory.

‘Plot the uploaded data and explain the outliers.’

03.05

Confirmation for consequential actions

For selected higher-risk operations, the connector shows a dialog describing the action. The operation does not run without your approval. Routine searches, reading and everyday changes do not require approval at every step.

For example, before deleting data or a financial operation, you can review what will happen and approve or cancel it. The operations requiring confirmation depend on the connector.