AI Features Not Working

The Research AI Panel powers Paper Chat, Multi-Doc Chat, Outline Generation, the AI Writing Assistant, and other features across AcaTrove. This article explains what to do when AI features are unresponsive, slow, or producing unexpected results.

The AI Panel Does Not Respond

If you type a message and nothing happens after pressing Enter:

1. Check that the Research AI Panel is open. Click the AI icon in the sidebar to toggle it. The panel should appear on the right side of the screen.
2. Verify your internet connection. AI features require communication with AcaTrove's backend services, which in turn connect to external AI providers.
3. Refresh the page. A stale WebSocket connection or expired session can prevent the panel from sending messages.
4. If the panel shows a loading indicator that never resolves, the backend AI service may be temporarily unavailable. Wait a few minutes and try again.

"Something went wrong with the AI chat"

This error indicates that the AI service returned an error or timed out during processing.

Steps to resolve:

1. Try sending your message again. Transient failures with AI providers are occasional and often resolve on the next attempt.
2. Shorten your message or simplify your question. Extremely long or complex prompts can cause timeouts.
3. If you are using Paper Chat, confirm the document has finished processing. Recently uploaded documents need time for text extraction and embedding generation before they can be queried.

Responses Are Slow

AI response times depend on several factors:

• Document size: Larger documents require more processing for semantic search retrieval.
• Query complexity: Multi-part or comparative questions take longer than simple lookups.
• Server load: During peak usage, responses may take longer than usual.
• Network latency: A slow connection between your browser and AcaTrove adds to response times.

If responses consistently take more than 15 seconds, check your network speed. For institutional networks with proxy servers, direct connections typically perform better.

Responses Do Not Reference the Correct Document

The Research AI Panel uses a scope system to determine which documents the AI should draw from. If responses seem unrelated to your current document:

1. Check the scope indicator at the top of the panel. It should show the name of the publication, project, or lab you are working with.
2. If the scope says "General" or shows a different resource, the panel is not focused on your document. Navigate to the specific publication page, and the scope will update automatically.
3. For Multi-Doc Chat in projects or labs, ensure the project or lab contains the documents you expect. The AI can only reference documents that are attached to the current scope.

Conversation Context Resets Unexpectedly

Paper Chat maintains a six-hour conversation memory window. After six hours of inactivity, the AI loses the context of earlier messages. Your chat history and annotations remain visible, but the AI will not recall previous turns when generating new responses.

If you need to reference earlier discussion points after a context reset, copy the relevant details into your new message so the AI has that information available.

Outline Generation or Writing Assistant Issues

If the Outline Generation or AI Writing Assistant features fail:

1. Ensure you have provided the required inputs. Outline Generation needs a topic and a selected template at minimum.
2. Check that the description or prompt is substantive enough. Very short inputs (under 10 words) may not give the AI sufficient context.
3. If you see "Failed to generate," try rephrasing your input and submitting again.

Tips

• The AI panel is available on every page via the sidebar. You do not need to navigate to a dedicated AI section.
• When AI responses seem off-target, adding specific terminology from the paper or field helps the semantic search retrieve more relevant passages.
• For persistent AI issues, administrators can check service status at /admin/api-health, which monitors backend connectivity and response times.