Paper Chat

Paper Chat lets you upload a research document and have a natural-language conversation with its contents. Ask questions, request summaries, and extract key findings -- all backed by citations that point to specific passages in the original text.

Getting Started

1. Navigate to a publication in AcaTrove, or upload a new PDF via the publications page.
2. Open the Research AI Panel from the sidebar.
3. The scope automatically sets to the current publication. The panel shows "Ask me about this publication."
4. Type your question and press Enter.

Free-Tier Limit

Hosted Free accounts currently get 5 Paper Chat requests per week. Actual chat requests consume the quota. Uploading a paper for later chat or creating an empty conversation shell does not.

Paper Chat processes the document's full text using semantic chunking and vector embeddings. This means the AI does not simply search for keywords -- it understands the meaning of your question and retrieves the most relevant passages.

What You Can Ask

Paper Chat handles a wide range of research questions:

• Summaries: "Summarize the main findings of this paper."
• Methods: "What statistical methods were used in the analysis?"
• Specific details: "What was the sample size in Experiment 2?"
• Critical analysis: "What limitations do the authors acknowledge?"
• Comparisons: "How does the theoretical framework here differ from Smith et al. 2023?"
• Definitions: "What do the authors mean by 'ecological validity' in this context?"

Citation-Backed Responses

Every Paper Chat response includes source citations. These citations specify:

• The section of the paper where the information was found
• The page number (when available)
• A preview snippet of the original text

This design ensures you can verify any AI-generated claim against the source material. Citations appear below each assistant message and can be expanded to see the full passage.

Annotation Types

Paper Chat supports nine types of annotations that you can create while reading and chatting:

| Type | Purpose | |------|---------| | Highlight | Mark important passages for later reference | | Note | Add your own comments and observations | | Insight | AI-generated analysis of selected text | | Question | Flag areas that need clarification | | Summary | Create condensed versions of sections | | Definition | Capture term definitions and explanations | | Citation | Annotate specific references and their significance | | Methodology | Tag methodological details for comparison | | Result | Mark key findings and data points |

Annotations are stored per conversation and persist across sessions.

Conversation History

Paper Chat maintains a six-hour conversation memory window. Within that window, you can ask follow-up questions that build on earlier exchanges. The AI considers your recent conversation turns when generating responses, enabling natural multi-turn dialogues.

After six hours of inactivity, the conversation context resets. Your annotations and chat history remain accessible, but the AI will no longer have the earlier turns in its active context.

Supported Formats

Paper Chat works with PDF documents uploaded to AcaTrove. The system uses PyMuPDF for text extraction, with a fallback parser for scanned or image-heavy PDFs. For best results, upload text-based PDFs rather than scanned images.

Tips

• Start with a broad question ("What is this paper about?") to orient yourself, then drill down into specifics.
• Use the annotation system to build a structured reading. Highlight key findings, tag methodologies, and add questions as you go.
• If the AI's response seems incomplete, rephrase your question with more specific terms from the paper. The semantic search works best with precise language.
• Paper Chat is available through the Research AI Panel on any publication page. You do not need to navigate to a separate tool.