Skip to main content

Admin Settings

The Admin Settings tab provides a centralized interface for managing all platform configuration values. Settings are organized into category tabs, each containing service groups with related configuration fields. Edits are staged across sections: use the SaveBar at the bottom to save all unsaved changes (Cmd/Ctrl+S) or Discard them. Leaving with unsaved changes prompts for confirmation. Failed changes remain staged for retry; uploads and command actions still run immediately.

Interface Overview​

Header Controls​

The settings page header provides:

  • Search -- Filter settings by name, description, or (non-sensitive) value. A counter to the left of the All Tabs checkbox shows how many settings match across every tab.
  • All Tabs -- A checkbox that flattens results from every settings tab into a single list (so a match on a tab you are not currently viewing still appears) and also reveals app-scoped settings that are hidden by default

Category Tabs​

Settings are organized into the following tabs, each with a distinctive icon:

TabCategories IncludedDescription
AI ConfigurationAI, KnowledgeAI model API keys, embedding settings, context configuration
External IntegrationsTools, SlackThird-party service connections and API credentials
FeaturesExperimental, NotebooksFeature flags and experimental feature toggles
SecurityAI ModerationContent moderation and access control rules
User ManagementUsers, ReferralsUser notification settings, seat defaults, credit enforcement
CommunicationsAdmin, FeedbackFeedback channels, email/Slack webhooks, What's New configuration
CustomizationBrandingLogo uploads, visual branding settings

Service Groups​

Within each tab, settings are grouped by service. Each service group card displays an icon, group name, and description. The following service groups are available:

AI Configuration Tab​

Service GroupSettingsDescription
OpenAI ServiceAPI key, default context, chunk size, format prompt template, image prompt, URL scanningOpenAI API integration
Anthropic ServiceAPI keyAnthropic API integration
Gemini ServiceAPI keyGoogle Gemini API integration
xAI ServiceAPI keyxAI API integration
Voyage ServiceAPI keyVoyage AI embedding integration
Embedding ServiceDefault embedding modelVector embedding configuration
Voice Session ServiceEnable/disable, AI voice, transcription modelVoice session configuration
Knowledge ManagementMax file size, vector threshold, max content lengthFile and vector storage settings

External Integrations Tab​

Service GroupSettingsDescription
Search ServiceSerper API keyWeb search integration
Slack IntegrationSigning secret, bot token, workspace configurationSlack bot settings
MCP ServerEnable/disable, GitHub client credentials, Atlassian client credentialsModel Context Protocol integration

Features Tab​

Service GroupSettingsDescription
Experimental FeaturesQuest Master, Mementos, Artifacts, Agents, Rapid Reply, Research Engine, Deep Research, Lattice, Knowledge Base Search, Help Chat, and moreFeature flag toggles
Notebook SettingsAuto-name notebookNotebook behavior configuration

Communications Tab​

Service GroupSettingsDescription
Feedback SystemEmail/Slack feedback toggles, webhook URLs, feedback email addressesFeedback routing configuration

Special Components​

Operations Model Setting​

Located in Models > Settings under Defaults, the Operations Model setting allows administrators to select which AI model is used for internal platform operations (such as auto-naming notebooks or generating summaries). The configuration includes:

  • Text Model -- The primary model for text operations
  • Image Model -- The model for image-related operations
  • Speech Model -- The model for speech-to-text operations

Operational Usage Billing​

Open Money > Pricing (the Growth and pricing page) and find Bill Operational Usage under Credits and pricing. With Enforce Credits on, automatic naming, summaries, tagging, knowledge-base query embeddings and file indexing outside data lakes debit credits for free-tier users paying for themselves. A user is free tier when they have no credit purchase and no active plan, including a plan granted through an organization seat.

Bill Operational Usage defaults to off. Turning it on also bills paying users and organization-funded usage. With Enforce Credits off, these operations do not debit credits. Data-lake ingestion and research retain their separate billing rules.

If a balance falls while the toggle is off, check whether the user has a purchase or active plan and which account pays for the operation. Tool and knowledge-base usage settles after the call, so it can take a free-tier balance slightly below zero. File indexing outside data lakes reserves credits first; if there are insufficient credits, the file shows an indexing failure. Add credits before retrying the upload.

Generation Models​

Located in Models > Settings under Generation, this card has one section per generation modality (Video and Music), with one switch per model in that modality's catalog. Each switch shows whether the server will accept generation jobs for that model:

  • A model that is on by default and has no admin override reads "Enabled by default". Once you flip a switch, your choice overrides the model's default in either direction
  • Models marked "Preview" are early-access provider models whose behavior or pricing may still change
  • Models marked "Retired" are no longer served by their provider. Their switch is locked off, and an override saved before the retirement does not turn them back on
  • Turning a model off makes new jobs for it be refused and removes it from the public model listings (GET /api/v1/video-models, GET /api/v1/music-models); a disabled video or music model also disappears from Studio
  • The in-chat music generation tool does not honour these switches yet; it always uses its built-in default music model
  • Models whose provider is not configured on the deployment (for example, the non-production test models on production) cannot generate regardless of the switch

Video and music are stored as separate settings (videoGeneration and musicGeneration), so a change in one section never affects the other.

Agent Video Clips Per Run​

Located in the AI category. Sets how many video clips one agent-mode run may start, counted across the run and the subagents it delegates to in the same process. The default is 2. Set it to 0 to remove video generation from agent mode. The tool is offered only when video_generation is in the Agent Orchestration Defaults allowed tools. A deployment that saved that list before this release keeps its stored list, so an admin must add video_generation to it. A platform admin can override it for a single organization from the setting's scoped overrides. When a run reaches its limit, the agent is told so and no further clip is started; each clip still holds and settles its own credits. Runs started with an API key never get the video tool.

Python Artifact Data Origins​

Located in the Experimental category, and used only while Artifacts are enabled. Python artifacts run in an isolated frame that can reach only the Python package CDN. This setting lists the HTTPS origins their code may also read data from, one per line, such as https://data.example.com. It is empty by default.

  • Enter an origin only: https://, a public hostname and an optional port, with no path, query, wildcard or credentials
  • IP addresses, localhost, and private or internal names are refused, as are this app's own hosts and their subdomains
  • At most 10 origins can be listed; entries are lowercased and duplicates are dropped on save
  • If the stored list is ever invalid, no origin is allowed until it is saved again
  • The model is told which origins it may read from, and the Python viewer shows the list to the user
  • The data server must allow cross-origin reads (for example Access-Control-Allow-Origin: *). Requests carry no cookies or credentials from this app
  • A change applies to pages loaded up to ten minutes after it is saved; a page that has already run Python keeps the earlier list until it is reloaded

Python code can send what it has to a listed origin as well as read from it, so list only hosts you trust with artifact content. HTML and React artifacts stay offline whatever this list holds.

Logo Upload​

Located in the Branding category, the Logo Upload component provides:

  • Light Mode Logo -- Upload a custom logo for light mode display
  • Dark Mode Logo -- Upload a separate logo for dark mode display
  • Use Both Logos -- Toggle to enable separate light/dark logos
  • Images are automatically compressed to a maximum of 200x200 pixels

What's New Configuration​

Located in the Admin category, this component manages the automated "What's New" notifications that inform users about platform updates. Configuration includes model selection, template editing, and preview capabilities.

MCP Server Configuration​

The MCP tab provides a dedicated form for creating MCP (Model Context Protocol) server configurations:

  • Server Name -- The identifier for the MCP server
  • Enabled -- Toggle to enable or disable the server
  • Environment Variables -- Key-value pairs for server-specific configuration, with the ability to add or remove variables dynamically

Setting Input Types​

Individual settings use one of the following input types based on their data type:

TypeControlBehavior
StringText input fieldFree-text entry, with sensitive values masked
NumberNumeric input fieldAccepts numeric values only
BooleanToggle switchOn/off toggle for feature flags and enable/disable settings
SelectDropdownConstrained to predefined options

Sensitive settings (such as API keys) are marked as such and their values are hidden in the UI.

Best Practices​

  • API Keys -- Always use the sensitive setting fields for API keys; these values are masked in the UI for security
  • Feature Flags -- Test experimental features in staging (dev stage) before enabling them in production
  • Search -- Use the search bar to quickly locate specific settings rather than browsing through all tabs
  • All Tabs -- Enable this to see search matches from every tab in one list (useful when a setting might live on a tab you are not viewing) and to reveal app-scoped settings that are hidden by default