Workspaces
Organize users, documents, and AI settings into isolated workspaces for different teams or use cases.
Workspaces are organizational containers that let you group users, documents, and AI configurations into isolated environments. Each workspace has its own chat history, provider settings, document sources, and connectors.
Creating a Workspace
Only super admins can create workspaces.
Go to Workspaces
Go to Admin → Workspaces and click Add Workspace.
Enter Details
Enter a Name and optional Description.
Save
Click Save Workspace.
Managing Members
Workspace members are users who have access to chat within that workspace. Admins can add and remove members from their own workspaces. Super admins can manage members in all workspaces.
Adding Members
Go to Admin → Workspaces → select a workspace.
Click Add Members.
Select users from the list and click Add Members.
Removing Members
Go to Admin → Workspaces → select a workspace.
Find the member in the members list and click Remove next to their name.
Providers
Each workspace can override the global provider settings with its own API keys, model selection, and inference parameters. If a workspace setting is not configured, the global setting is used as a fallback. See Organization Settings for details on all available provider settings.
Configuring Provider Settings
Go to Admin → Workspaces → select a workspace.
Click Configure Provider Settings.
Configure the desired settings and click Save.
Available Overrides
| Setting | Description |
|---|---|
| Custom System Prompt | Additional instructions prepended to every conversation in this workspace. See Organization Settings for prompt guidelines. |
| Default Model | The AI model used for chat in this workspace |
| Default Image Model | The model used for image generation |
| Reasoning Effort | Thinking level for models that support extended reasoning (low, medium, high) |
| Thinking Budget | Maximum token budget for extended reasoning |
| Max Retries | Number of retry attempts when an API call fails |
| Retry Interval | Delay in seconds between retry attempts |
| API Keys | Workspace-level keys for OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, OpenRouter, and ElevenLabs |
Assistant Name & Avatar
Customize the assistant's identity per workspace. When enabled, a custom name and avatar image appear alongside the assistant's messages in chat.
Configuring Assistant Identity
Go to Admin → Workspaces → select a workspace → Edit.
Enter an Assistant Name (e.g., your brand name or assistant persona).
Enable the Display Assistant Name toggle to show the name in chat messages.
Optionally upload an Assistant Avatar image (PNG or JPEG, max 500 KB).
Click Save.
If no assistant name is set, the workspace falls back to the organization-level assistant name, then the global default.
Chat Suggestions
Configure suggestion buttons that appear on the chat welcome screen. Suggestions help users get started by providing common questions or prompts they can click to begin a conversation.
Managing Suggestions
Go to Admin → Workspaces → select a workspace.
In the Suggestions section, click Add Suggestion.
Enter the suggestion prompt text and click Save.
Suggestions appear as clickable buttons on the chat welcome screen. Clicking a suggestion sends the prompt immediately.
Contact Form
When the chatbot cannot resolve a user's question, it can offer to send a contact form email to a predefined recipient. This gives users a way to escalate their issue without leaving the chat.
Enabling the Contact Form
Go to Admin → Workspaces → select a workspace → Edit.
Enable the Contact Form toggle.
Enter the Recipient Email address where submissions should be sent.
Click Save.
When enabled, the chatbot will automatically offer to forward the conversation and the user's contact details to the configured recipient if it cannot answer a question. Each chat can only send one contact form to prevent duplicate submissions. The email includes a conversation summary, the user's contact information, and a link to the chat.
Web Fetch
When enabled, the chatbot can download and read documents shared via URL during a conversation. This is useful when users paste links to files (PDFs, Word documents, spreadsheets, etc.) and want the assistant to read and discuss their contents.
Enabling Web Fetch
Go to Admin → Workspaces → select a workspace → Edit.
Enable the Web Fetch toggle.
Click Save.
Supported Formats
PDF, DOCX, XLSX, PPTX, HTML, CSV, Markdown, and plain text. Files up to 20 MB are supported.
Google Drive Integration
When used together with a Google Drive connector, the chatbot automatically fetches the download URL first and then reads the file content, making it seamless to discuss Google Drive documents in chat.
Sandbox
The sandbox gives the assistant an isolated Linux environment where it can run code and produce files. Each workspace decides for itself whether to have one.
Turning it on for a workspace lets the assistant run code and hand the resulting files back in chat, using the bash and present_files tools. Files it saves are kept between conversations, and members of the workspace can browse, download and share them from Artifacts in the sidebar. Script-backed Skills need the sandbox too.
Enabling the Sandbox
Go to Admin → Workspaces → select a workspace → Edit.
Enable the Sandbox toggle.
Click Save.
The toggle is off by default, and it only appears when the sandbox is enabled for your organization. Turning it back off removes the two tools from that workspace and hides its Artifacts page; files already on the persistent volumes are left alone and reappear if you enable it again.
See Sandboxes for what the assistant can do inside one, how artifacts are browsed and published, and how to restrict its network access.
Output Format
Each workspace can be configured with a specific output format that determines how the AI presents its responses.
| Format | Description |
|---|---|
| Markdown | Standard markdown text responses (default) |
| OpenUI | Interactive generative UI with rich components such as charts, forms, tables, and more. See OpenUI for details. |
Voice Input
Each workspace can enable voice input (speech-to-text) so users can dictate messages instead of typing. When enabled, a microphone button appears in the chat input.
| Setting | Description |
|---|---|
| Voice Input | Enable or disable voice input for this workspace |
| Transcription Model | Model used for speech-to-text transcription. Leave blank to use the global/default transcription model. |
Knowledge Base
Workspaces can have document sources assigned to them. Documents belonging to those sources are used as context when answering questions in that workspace.
Assigning Document Sources
Go to Knowledge Base → select a document source.
In the Workspaces section, select the workspaces to assign it to.
Click Save. All documents ingested from the source will be available in the selected workspaces.
Documents cannot be assigned to workspaces individually. A document's access always comes from the document source it belongs to, including files you upload, which belong to a Local files source.
Conversation Compaction
Each workspace can override the global conversation compaction settings to control when long conversations are summarized.
Configuring Compaction
Go to Admin → Workspaces → select a workspace → Edit.
Configure the compaction settings and click Save.
| Setting | Description |
|---|---|
| Compaction Enabled | Override the global setting to enable or disable compaction for this workspace |
| Compaction Threshold | Token count at which compaction is triggered in this workspace. Overrides the global threshold. |
Connectors
Connectors extend AI capabilities with external tools. You can assign connectors to specific workspaces so their tools are available only where needed.
See Connectors for setup instructions.
Embeddable Chat
Each workspace supports an embeddable chat widget that can be integrated into external websites or applications. The widget can operate in three modes:
| Mode | Description |
|---|---|
| Authenticated | Users authenticate via a secure JWE token generated by your application |
| Guest Signup | Visitors provide their name and email before chatting |
| Anonymous | Visitors can chat immediately without any identification |
See Embeddable Chat for integration instructions.
Enabling Embeddable Chat
Go to Admin → Workspaces → select a workspace.
Click Configure Embedded Chat.
Select the desired Mode: Token for authenticated access, Guest Signup, or Anonymous. Choose Disabled to turn the widget off.
Configure key pairs for secure communication and click Save.