PaperManager

Paper Detail

Click any paper card to open the Paper Detail view (/paper/:id). It has four main panels.

Metadata & Abstract

The left panel shows:

Title, year, DOI, venue
Citation count (from Semantic Scholar)
Metadata source badge
Authors — click an author chip to open their People profile
Topics — click a topic chip to filter the library to that topic
Abstract (original)
AI summary (generated by Claude Opus during upload)

Click Edit to open a modal where you can update any field.

PDF Viewer

Available when a PDF was uploaded (not for URL-only papers). The PDF is streamed from Google Drive and rendered inline in the browser.

!!! tip For URL-ingested papers you can still attach a PDF later via the Upload PDF button in the metadata panel.

Figures

Figures are extracted from the PDF and stored on Google Drive. The figures panel shows:

Figure number and page number
Caption (generated by Docling, Ollama, or Claude Vision)
Click any figure for full-size view
Ask about this figure — opens a vision chat powered by Claude; ask anything about the figure image

Extracting Figures

Click Extract Figures and choose a caption method:

Method	Description
Docling	Structural PDF parser — fast, no AI cost
Ollama	Local LLM captions (default)
Claude Vision	Highest quality; uses Anthropic API

The default method can be set in Settings.

Notes

A Markdown editor is attached to each paper.

Special Syntax

Syntax	Effect
`@PersonName`	Links to a Person node; creates it if not found
`#TopicName`	Links to a Topic node; creates it if not found

When you save a note, all @mentions and #topics become graph relationships:

(Note)-[:MENTIONS]->(Person)
(Note)-[:MENTIONS]->(Topic)

This means you can later query “all notes that mention @Jan” or “all papers with notes about #transformers” from the Cypher editor or knowledge chat.

Preview

Toggle Preview mode to render the note as formatted Markdown.

Chat with Paper

Ask natural language questions about the paper’s full text. The raw_text extracted from the PDF is included in the context (truncated to model limits).

Model Options

Model	Provider	Best for
Claude Opus 4.6	Anthropic (personal key)	Highest quality answers
Claude Work	Anthropic Foundry gateway	Enterprise environments
Ollama (llama3.2:3b)	Local	Fully offline

Example Questions

What is the main contribution of this paper?
How does this method compare to previous approaches?
What are the key limitations mentioned by the authors?
Can you explain the loss function used in equation 3?

References & Citations

Outgoing References

Papers that this paper cites. Extracted on-demand and shown in a review list before being saved.

Extraction pipeline (tries in order):

Semantic Scholar /references API (requires DOI)
Regex on the REFERENCES section of raw_text
Claude Haiku on the last 30% of the document (when earlier strategies return < 3 results)

Each saved reference creates a Paper stub node (title + DOI), tagged from-references, and linked with a CITES relationship. If you later import the full paper by DOI, the stub is enriched rather than duplicated.

Incoming Citations

Papers in your library that cite this paper. Maintained automatically as you import papers.

Managing Tags, Topics, and People

From the Paper Detail view you can:

Add / remove tags — type a tag name or choose from the list
Add / remove topics — link or unlink research topics
Link people with roles — connect a person with a workflow role

INVOLVES Roles

Role	Meaning
`authored_by`	Author of the paper
`shared_by`	Person who shared the paper with you
`working_on`	Person is working on this topic/paper
`collaborating`	You are collaborating with this person on this
`feedback_needed`	You need feedback from this person
`supervisor`	Supervisor for this work

Projects

Assign the paper to one or more projects directly from the Paper Detail page. You can also do this during the upload flow.