AI Agent Admin Settings

    Admin Settings is the central control panel for configuring Ascendix Conversational Agent across your organization.

    Admin Settings is the central control panel for configuring Ascendix Conversational Agent across your organization. To access it, go to the App Launcher and search for Agent Admin Settings

    The panel is organized into six tabs: System Capabilities, General Settings, Provider, Prompts, Restrictions, Persona & Object Context, and LLM Usage.


    System Capabilities
    The System Capabilities tab is the main on/off control panel for all AI features in your org. Each toggle enables or disables a specific capability for all users — changes take effect immediately, no deployment or page refresh required.

    Capability What it controls
    Search Dynamic Search — find records through natural language
    Create / Update Records Record Management — create and update records through conversation
    Email Draft Email Draft — compose emails from a plain-language instruction
    Graph Create Graph Create — create and link networks of related records
    Memory Memory System and Feedback Loop — persistent user preferences and rating signals
    Activity Activity Tool — log calls, tasks, and events through natural language
    General Knowledge General Knowledge — open-ended advisory Q&A
    Plan Builder XX

    Notes

    • Additional capabilities will be added to this tab as the product expands.
    • Toggling a capability off removes it from the chat experience for all users in the org immediately.

    General Settings
    The General Settings tab contains organization-wide configuration options for Ascendix Conversational Agent. Additional controls will appear here as the product expands.

    Record Operations

    Controls how many records a single create or update request can touch. This limit applies across all users in the org and is separate from the feature toggles on the System Capabilities tab.

    Setting Description
    Max Create / Update Records per Request Sets the maximum number of records a single instruction can create or update in one pass

    Provider
    The Provider tab controls which external AI provider is active for the organization. Each provider can have an org-level shared API key configured here.

    Users can also set their own personal API key directly in the chat interface via the gear icon — personal keys take precedence over the org-level key when active.

    When no external provider is configured or active, the system falls back to Agentforce.

    OpenAI

    Setting Description
    API Key The org-level OpenAI API key. Once configured, displays as "API key configured" with a green checkmark
    Status Activates or deactivates OpenAI as the provider for the org. Toggle is immediate — no redeployment required
    Remove API Key Removes the configured org-level key

    Notes

    • Personal API keys set by individual users via the chat gear icon take precedence over the org key.
    • Changes to provider status take effect immediately.

    Prompts
    The Prompts tab lets administrators review the system prompts that drive each AI capability and configure which AI model runs each one. Prompt wording cannot be edited here — this tab is primarily for review and model assignment.

    For each prompt you can see:

    Field Description
    Prompt Dropdown to select which system prompt to view
    Version The current version of the selected prompt
    AI Model The model assigned to run this prompt — selectable per prompt
    Active Toggle to activate or deactivate the prompt
    Prompt Body The full system prompt text — read only

    Notes

    • Each capability in the system has its own underlying prompt — for example ActivityPlanSystem drives the Activity Tool, CsvFieldMapping drives the CSV import column mapping, and so on.

    Restrictions
    The Restrictions tab lets administrators block the AI from performing create, update, or delete operations on specific Salesforce objects or fields. This is the primary mechanism for protecting sensitive data from being touched by the AI — regardless of what a user asks.

    Restrictions deploy as custom metadata asynchronously after each save — there may be a short delay before a new restriction takes effect.

    Current restrictions

    Each restriction shows:

    Field Description
    Object API Name The Salesforce object the restriction applies to
    Restriction Type Block All — prevents any create, update, or delete on that object
    Reason A description of why the restriction exists — for admin reference
    Active Toggle to enable or disable the restriction without deleting it

    By default, the following objects are restricted out of the box for security reasons: Group, GroupMember, and User.

    Adding a restriction

    To add a new restriction, fill in the form at the bottom of the tab:

    Field Description
    Object API Name The API name of the Salesforce object to restrict (e.g. Account, Custom_Object__c)
    Field API Name Optional — restrict a specific field rather than the whole object
    Restriction Type Block All is the only available type
    Reason Optional description for admin reference

    Save to apply. The restriction deploys asynchronously — allow a short delay before testing.

    Notes

    • When a restriction is active, the AI will silently skip any operation that would violate it.
    • Restrictions apply org-wide to all users regardless of their Salesforce profile or permission set.
    • Deactivating a restriction disables it without deleting it — useful for temporary changes.

    Persona & Object Context
    The Persona & Object Context tab has two distinct sections that together shape how the AI understands and interacts with your organization's data and domain.

    Agent Persona

    Defines the AI's identity, tone, and domain focus. Changes deploy as custom metadata.

    Field Description
    Business Sphere Defines the industry and role context — e.g. "Commercial Real Estate broker Agent"
    Domain Context Terms, objects, and workflows the agent should assume the user cares about.
    Guidance Style, safety, and escalation rules — what the AI should avoid, when it should ask for confirmation before acting

    The Agent Persona is typically configured once during setup and refined over time as the team learns how the AI behaves in practice. It directly influences how the AI introduces itself, what vocabulary it prioritizes, and how it frames responses.

    Object Context

    Lets administrators add object-specific instructions that are injected into the AI's reasoning whenever it interacts with a particular Salesforce object — whether searching, creating, updating, or summarizing records on that object. Active rows load at runtime; inactive rows are ignored.

    Field Description
    Object API Name The Salesforce API name of the object (e.g. Account, Custom_Object__c)
    Label Display name — defaults to the object API name if left blank
    Instructions Field hints, disambiguation rules, preferred terminology — anything that helps the AI reason more accurately about that object
    Active Toggle to enable or disable the context row without deleting it

    Notes

    Instructions are injected at runtime only for interactions involving that specific object, keeping context focused and relevant.


    LLM Usage
    The LLM Usage tab provides a full log of every AI interaction in the org — including token consumption, flex credit usage, response latency, and success status. It's the primary tool for monitoring usage, managing costs, and auditing AI activity across the organization.

    Filters

    Filter Description
    Start date / End date Define the date range for the query
    Provider Filter by AI provider — defaults to all providers
    User Filter by a specific user — optional, defaults to all users

    Click Apply to load results for the selected filters.

    Summary metrics

    The top of the results shows aggregated totals for the selected period:

    Metric Description
    Input Tokens (Sum) Total input tokens consumed across all interactions
    Output Tokens (Sum) Total output tokens generated across all interactions
    Total Tokens (Sum) Combined input and output token count
    Flex Credits (Sum) Total flex credits consumed — the primary cost metric for Agentforce usage
    Calls (Count) Total number of AI interactions in the selected period

    Interaction log

    The table below the summary shows each individual AI interaction with full detail:

    Column Description
    Time Timestamp of the interaction
    User The Salesforce user who triggered it
    Provider The AI provider used (e.g. Agentforce, OpenAI)
    Model The specific model that ran the interaction
    Prompt The system prompt that was invoked
    Input / Output / Total Tokens Token breakdown for that interaction
    Flex Credits Flex credits consumed for that interaction
    Latency Response time in milliseconds

    Notes

    • Use this tab to track which features consume the most credits and identify heavy users or unexpected spikes.
    • Filtering by user is useful for cost allocation or investigating specific interactions.
    • The combination of Prompt and Model columns shows exactly which capability ran and on which model — useful for diagnosing performance issues or model misconfigurations.