Multi-Document Chat

Multi-Document Chat lets you ask questions across all documents in a project, lab, collection, or your entire research library at once. Instead of reading each paper individually, you can synthesize information across your corpus in a single conversation.

How It Works

Multi-Document Chat uses Retrieval-Augmented Generation (RAG) to search across multiple documents simultaneously. When you ask a question, the system:

1. Converts your question into a vector embedding.
2. Searches across all document chunks within your selected scope.
3. Retrieves the most relevant passages from multiple documents.
4. Generates a synthesized response with source attribution for each claim.

Responses stream in real time, and source citations appear below each message so you can trace every statement back to its origin document.

Scopes

Multi-Document Chat operates at four scope levels, selectable from the Research AI Panel's scope dropdown:

Project -- Searches across all documents attached to a specific project. This is the most common scope for focused research tasks. Navigate to a project page and open the Research AI Panel to use it.

Lab -- Searches across all documents belonging to a lab. Useful when your research question spans multiple projects within the same lab.

Collection -- Targets a curated set of documents you have grouped into a paper collection. Useful for literature reviews where you have assembled a specific reading list.

Global (All My Research) -- Searches your entire document library. Use this for exploratory questions or when you are unsure which project or lab contains the relevant information.

Access Control

Multi-Document Chat respects AcaTrove's Attribute-Based Access Control (ABAC) system. The AI can only retrieve and cite documents you have permission to access. If a document is restricted by department, clearance level, or sharing settings, it will not appear in your search results.

This means that two users asking the same question in the same project may receive different answers if they have different access levels. The system ensures that sensitive research data is never exposed beyond its intended audience.

Conversations

Multi-Document Chat conversations are persisted on the server. This means:

• You can close the Research AI Panel, navigate to other pages, and return to find your conversation intact.
• Conversation history is scoped to the specific project, lab, or global context where it was created.
• Each conversation has a title based on the scope (e.g., "Project: Genomics Study" or "Global chat").
• The AI considers previous turns in the conversation when generating responses, enabling follow-up questions.

Source Diversity

The RAG system is designed to surface information from across your document library rather than relying heavily on a single paper. When multiple documents address your question, the response will draw from several sources and indicate which document each piece of information came from.

Example Questions

• "What sample sizes were used across all studies in this project?"
• "Compare the methodological approaches in my three most recent papers."
• "What gaps in the literature have been identified across this lab's publications?"
• "Which of my documents discuss machine learning applications in genomics?"
• "Summarize the key findings from all papers published in 2025."

Tips

• Start with a project scope for most research tasks. Only broaden to lab or global scope when your question genuinely spans multiple contexts.
• Check the document count badge in the scope selector. If a project has only one or two documents, Multi-Document Chat offers limited advantage over Paper Chat.
• Use follow-up questions to drill deeper. For example, start with "What are the main themes across these papers?" and then ask "Tell me more about Theme 2 and which papers support it."
• Source citations help you build proper reference lists. When the AI cites a document, note the title and section for your own writing.