Research Chat Mode

Research Chat is the primary mode of the Research AI Panel. It provides a conversational AI assistant that uses Retrieval-Augmented Generation (RAG) to answer your questions using your own research documents as context.

How It Works

When you type a question in Research Chat, the system follows a multi-step process:

1. Intent classification -- The AI determines whether your prompt is a question (ask) or an editing request (edit). A badge appears showing which intent was detected.
2. Context gathering -- Based on your current scope, the system retrieves relevant document content, page context, and conversation history.
3. Response generation -- The AI generates a streaming response grounded in your documents, with source citations where applicable.

Asking Questions

Type your question in the text area at the bottom of the panel and press Enter or click Send. Responses stream in real time. You can also use the microphone button to dictate questions via voice input.

Example questions:

• "What methods were used in this study?"
• "Summarize the key findings across all documents in this project."
• "How does this paper's approach compare to the literature?"
• "What are the main limitations discussed in this publication?"

Citations and Sources

When the AI references content from your documents, it attaches source citations. These appear below each assistant message as a collapsible sources list. Each citation includes:

• The source document title
• The relevant section or page
• A preview snippet of the referenced content

Click a source to verify the original text. This ensures you can always trace an AI response back to the underlying data.

Conversation History

Research Chat maintains conversation history within each scope:

• Single-document scopes (publication, document): History is stored in-memory for the current session. It clears when you change scopes.
• Multi-document scopes (project, lab, global): Conversations are persisted server-side. You can close the panel, navigate away, and return to find your conversation intact.

The AI considers the last several turns of conversation when generating responses, enabling multi-turn dialogues where you can ask follow-up questions naturally.

Intent Routing

The panel uses automatic intent routing to distinguish between questions and edit requests. When you ask a question, it routes to the appropriate chat backend. When it detects an edit intent (such as "rewrite this paragraph" or "improve the clarity"), it routes to the document editing system instead.

If the intent classifier routes your prompt as an edit but you intended to ask a question, click the "Ask instead" button that appears next to the intent badge.

Scope-Dependent Behavior

The behavior of Research Chat changes based on your selected scope:

• Publication scope: Uses Paper Chat to answer questions about a single uploaded document with citation-backed responses.
• Project scope: Uses Multi-Document Chat to search across all documents in the project via RAG.
• Lab scope: Searches across all documents belonging to the lab.
• Global scope: Queries your entire research library.
• Document, Job, Grant scopes: Uses page context to answer questions about the current item.

See Scope Selection for guidance on choosing the right scope.

Tips

• Be specific in your questions. "What sample size was used in Study 2?" will produce a more useful answer than "Tell me about the methods."
• For multi-document scopes, the AI will tell you which documents it drew from. Use this to discover connections across your research.
• If a response seems incomplete, ask a follow-up. The conversation context helps the AI refine its answers.