Using AI agents

Interacting with AI in the chat

When you open an AI agent, the AI Chat tab opens. The agent also has the Knowledge base and Chat outputs tabs.

The model used by the agent is displayed to the right below the prompt field. Click it to choose a different model from the list of the models available in your workspace.

Above the prompt field, a set of ready-made prompts is displayed, for example, What can I ask you, Summarize the knowledge base, Show source documents, Find a document, Find tasks and deadlines, Compare documents, Find contradictions, and Show saved results. Click one of them to use it as your prompt. The same set is available via the built-in-prompts menu next to the + button.

Just enter your prompt in the Ask any question or describe what you need field and click the arrow button.

You can attach up to 5 documents to your prompt. Once a file is entered into the chat, its content will be passed to the model. Based on this, the model will provide a response.

To attach a document stored in the workspace:

1. Click the + button below the prompt field.
2. Select the Add files from storage option from the menu.
3. In the Select panel, choose the app to take the file from: Files, Rooms, Forms, or AI agents.
4. Choose the necessary documents and click the Select button.

You can also attach images to your chat messages to provide visual context for the AI. Use the Add files from device option of the same menu to select and upload an image from your device.

The + button below the prompt field opens a menu with the following options:

• Add files from storage – attach files stored in the workspace.
• Add files from device – attach files from your computer.
• Web search – the switcher that turns web search on or off for the current chat.
• Effort – the submenu where you set how much reasoning the model applies before it answers. The currently selected level is displayed next to the option name.

The following effort levels are available:

• No thinking – the model answers directly, without a reasoning step. This is the fastest option.
• Low – the model performs a short reasoning step before answering.
• High – the model reasons in more detail, which suits multi-step tasks such as document analysis or comparison.
• Maximum – the model applies the deepest reasoning available. Responses take the longest to generate.

The following buttons are displayed below each AI response:

• the icon – copies the response to the clipboard.
• the icon – asks the model to generate the response again.
• the icon – saves the response to a file. In the panel that opens, choose the location, enter the file name, and click the Save button.

It's also possible to enter a prompt, for example, Save as docx. In this case, the result will be stored in the Chat outputs tab.

You can also generate DOCX files, PDF forms, and PPTX presentations directly in the AI chat. The generated file is displayed inline within the message and is saved automatically to the Chat outputs folder. You can open the generated file for editing in a new tab.

For complex requests, the AI may display a Thinking block showing its reasoning process before presenting the final answer. This block appears automatically depending on the complexity of the request and is not available for all queries.

Click the Chat history button in the upper left corner to display all chats within this agent. The chats are grouped by date and can be found using the Search field above the list. The chats are related to the current user and cannot be shared.

You can click the icon next to the chat name to Open, Save, Rename, or Delete the chat.

When one of the existing chats is opened, you can use the New chat button to start a new chat.

Using web search

Web search allows the AI to look for information on the Internet and use it in its response. To make it available, first you need to connect a search provider in the workspace settings -> AI settings -> Web search.

In the chat, click the + button below the prompt field and turn Web search on or off. When web search is turned off, its tools are removed from the list of the tools offered to the model.

Just enter your search request in the Ask any question or describe what you need field and click the arrow button.

You can see the result in the chat. When you click on a site, it will open in a separate tab.

Web crawling is also enabled along with web search. This is an in-depth internet search that opens a specific website and retrieves all its contents. To activate it, you can simply enter any link into the chat, or it can be activated automatically after performing the web search. When you click on a site, it will open in a separate tab.

There's also a Knowledge base search. It searches for the most relevant pieces of content in documents from the Knowledge base tab and builds a response based on them.

Using MCP server tools

The ONLYOFFICE DocSpace MCP Server is enabled by default in the workspace settings -> AI settings -> MCP servers, so you can start using the MCP server right away. It allows AI agents to interact with different workspace elements, for example, you can ask AI to create a new room or list the files stored in a specified folder, etc.

For the full list of the tools provided by the ONLYOFFICE DocSpace MCP Server, refer to the API documentation.

On the MCP servers tab, click the Edit configuration button to set up MCP servers and enable their tools for use in the AI chat. Apart from the built-in server, you can connect external MCP servers; their tools are added to the common list and offered to the model. Individual tools and whole servers can be enabled and disabled.

If a tool or a whole server is disabled in the middle of a conversation, or the model calls a tool that does not exist, the chat does not hang: the model receives an unavailable result and the conversation continues.

When calling MCP server tools, the Confirmation dialog is displayed before the function is executed. It shows the name of the tool the AI is about to use, its source, and a short description of the action. Click the icon next to the description to show the arguments this function is called with.

Click the Allow button to call the function – its result will be sent to the chat. Click the Deny button to cancel the function call; the model will continue responding.

You can see the result of the function execution in the chat by clicking the icon.

Using knowledge base

To be able to add files to the knowledge base, first you need to add a provider in the Knowledge settings.

In the Knowledge base tab of the AI agent, you can upload files from the workspace or from your device.

E.g., if you want to create contracts using AI, you can upload the contract examples used in your organization, and AI will use information from them, such as your company name, etc.

When the Knowledge base tab is empty, the No files uploaded screen is displayed with a short description and two shortcuts, Upload from storage and Upload from device. The same options are available via the New button.

To upload a file from the workspace to the Knowledge base tab:

1. Click the New button in the upper left corner.
2. Select the Upload from storage option from the menu.
3. In the Select panel, choose the app to take the files from: Files, Rooms, Forms, or AI agents.
4. Choose the files you want to upload to the knowledge base and click the Select button.

To upload a file from your device to the Knowledge base tab:

1. Click the New button in the upper left corner.
2. Select the Upload from device option from the menu.
3. Choose the files you want to upload to the knowledge base in the file manager window and click the Open button.

This can also be done by dragging the necessary files from the file manager window to the working area.

When files are uploaded, they appear in the list with the preparing badge, indicating that the vectorization operation has begun.

If an error occurs, an error badge appears, allowing you to restart the operation. Badges are visible to all users. Only those who can add files can restart the operation.

Once the upload is complete, you can ask questions about the uploaded files without attaching them to a chat. These files can only be viewed, not edited.

Managing knowledge base files

The uploaded files are displayed in a list with the Name, Modified, and Size columns. Use the Search field above the list to find a certain file, the filter and sorting icons to the right of it to narrow down or reorder the list, and the view icon to switch between the list and tile views.

Click the icon next to a file to open the context menu with the following options:

• Select – select the file to perform group operations.
• Preview – open the file for viewing.
• Ask AI – send the file to the chat as context.
• Copy – copy the file to another location.
• Download – download the file, or choose another format in the submenu.
• Mark as favorite – add the file to the Favorite files list.
• File info – open the Info panel.
• Delete – remove the file from the knowledge base.

When one or more files are selected, the toolbar at the top of the page provides the Download, Download as, Copy, and Delete operations.

The Info panel contains the History and Details tabs. The Details tab displays the file properties: Owner, Type, File extension, Size, Modification date, Last modified by, Creation date, Versions, and Comments.

Chat outputs

The Chat outputs tab is the section for the files produced during the chat with AI: the documents, forms, and presentations generated by the AI agent are stored here automatically. You can also save chats and separate messages to this tab. The files stored here can be edited.

While the tab is empty, the No outputs yet screen is displayed with the Create chat shortcut that takes you to the AI Chat tab.

The saved files are displayed in a list with the Name, Modified, and Size columns. Use the Search field above the list to find a certain file, and the filter, sorting, and view icons to the right of it to narrow down, reorder, or switch between the list and tile views.

When one or more files are selected, the toolbar at the top of the page provides the Download, Download as, Move to, and Copy operations.

When a file is selected, the Info panel contains the Share, History, and Details tabs. Use the Create and copy option of the Share tab to create a shared link to the file and copy it to the clipboard.

Saving chats and messages

You can save chats and separate messages to a folder in the Files app, to a room in the Rooms app, or to an AI agent.

To save a chat to a file:

1. Click the Chat history button in the upper left corner.
2. Click the icon next to the necessary chat name.
3. Use the Save option.
4. In the Select panel, choose the app to save the file to: Files, Rooms, or AI agents, then navigate to the necessary location. For an AI agent, choose the agent, then its Result Storage or Knowledge folder. Use the path at the top of the panel to go back to the previous level and the Search field to find the necessary location.
5. Enter the file name in the File name field.
6. Click the Save button.

To save a message to a file:

1. Open the necessary chat.
2. Click the icon below the necessary message.
3. In the Select panel, choose the app to save the file to: Files, Rooms, or AI agents, then navigate to the necessary location. For an AI agent, choose the agent, then its Result Storage or Knowledge folder. Use the path at the top of the panel to go back to the previous level and the Search field to find the necessary location.
4. Enter the file name in the File name field.
5. Click the Save button.

Inviting other users

You can share AI agents with other users.

The following permission levels are available in AI agents:

• Agent manager (Paid) - Agent managers can manage the assigned AI agent, invite new users, and assign roles below their level. They can use the chat, upload knowledge files, work with the Chat outputs, and manage chat settings.
  Guests and Users cannot be assigned as Agent managers. Only Room admins and Full admins are suitable for the specified role.
• Content creator - Content creators can use the chat and work with files in the Chat outputs. They can upload files to the knowledge base but can't manage users or access agent settings.
  Guests cannot be assigned as Content creators. Only Room admins, Full admins, and Users are suitable for the specified role.
• Viewer - Viewers can only view files in the Chat outputs with outputs generated by others. They do not have access to the chat or the knowledge base.
  Unlike rooms, in agents, guests can only be assigned as Viewers.

To invite a user to an AI agent,

1. Open the necessary AI agent where you want to invite users.
2. Click the icon next to the AI agent name.
3. Select the Invite people option from the menu.
4. Invite external users (guests) via the link:
   1. Turn on the Invite via link switcher – the link will be copied to the clipboard. You can send the copied link in any convenient way.
      The Viewer permission level is only available for guests.
      This link is valid for 7 days only.
      
5. Alternatively, you can use the Add manually section to add email addresses manually one by one:
   1. Click the Choose from list option to select the existing workspace users. Check them in the list and click the Select button.
      • Use the Members tab to select internal users.
      • Use the Groups tab to select all users who belong to a certain group. It's also possible to choose the Everyone group to select all workspace users.
      • Use the Guests tab to select external users.

      You can also start typing the existing user or group name or email in the entry field and select the found users or groups. The existing contacts or groups that match the entered characters are displayed there as you type.

   2. To invite new external users (guests) by email, enter a valid email address and click Add below or press Enter. In the same way, add several email addresses if necessary.
   3. Change the permission level if necessary. The following permission levels are available: Agent manager, Content creator, Viewer.
   4. When ready, click the Send invitation button. Invitations will be sent to the specified email addresses.
      

It's also possible to invite members to the AI agent from the Info -> Contacts screen. Click the icon to add new members.

When a user accepts the invitation to the AI agent, they can view the agent in the AI agents app and participate in collaboration with access rights according to the permission level.