AI Assistant Administration

The AI assistant administration page provides fine-grained control over AcaTrove's conversational AI behavior. Administrators can customize system prompts, set guardrails, configure citation requirements, and tailor the assistant's personality to match institutional expectations. Admin role is required.

Accessing AI Assistant Admin

Navigate to /admin/ai-assistant from the admin sidebar.

AI assistant administration page with prompt and behavior settings

System Prompt Customization

The system prompt defines the AI assistant's baseline behavior. While AcaTrove provides a carefully designed default prompt, administrators can customize it:

• Institutional context -- Add information about your institution, departments, or research focus areas so the assistant can provide more relevant responses.
• Tone and style -- Adjust whether the assistant communicates formally, semi-formally, or casually.
• Domain emphasis -- Instruct the assistant to prioritize certain research domains or methodologies relevant to your workspace.

Edits to the system prompt take effect for all new conversations. Existing conversations retain the prompt that was active when they were started.

Citation Requirements

Configure how the AI assistant handles citations in its responses:

• Always cite sources -- The assistant includes citations for all factual claims, drawing from the user's document library.
• Cite when available -- The assistant cites when relevant documents are found but does not refuse to answer when citations are unavailable.
• No citation requirement -- The assistant provides answers without citation obligations.

The default setting is "Cite when available," which balances helpfulness with academic rigor.

Guardrails

Set boundaries for what the AI assistant will and will not do:

• Prohibited topics -- Specify topics the assistant should decline to discuss.
• Response length limits -- Set maximum response lengths to keep conversations focused.
• Academic integrity mode -- When enabled, the assistant explicitly notes when it is generating original text (as opposed to summarizing or quoting sources) and includes transparency disclaimers.

Conversation Settings

• Memory duration -- How long the assistant remembers previous messages in a conversation (default: 6 hours). Longer memory produces more coherent extended conversations but uses more tokens.
• Maximum conversation length -- The number of turns before a conversation is automatically archived and a new one begins.
• Welcome message -- Customize the message users see when they open a new AI assistant conversation.

Usage Monitoring

The AI assistant admin page includes a usage summary showing:

• Total conversations started per day, week, and month.
• Average conversation length.
• Most common topics and questions.
• User satisfaction ratings (if feedback collection is enabled).

This data helps you understand how your workspace uses the assistant and whether configuration changes are needed.

Tips

• Review the usage data monthly to identify common questions that might indicate a need for better documentation or training.
• Enable academic integrity mode for workspaces where transparency about AI-generated content is important.
• Test system prompt changes in a personal conversation before deploying them workspace-wide.