Paper Import

Source: server/catgo/routers/paper.py, server/catgo/models/paper.py

The paper import module ingests scientific papers into CatGo's session store so CatBot can reference them while constructing workflows. Two ingestion paths are supported: direct PDF upload, and DOI resolution via CrossRef metadata.

This module is the data ingestion layer. The downstream "tell CatBot about this paper and let it build a workflow" flow happens through normal CatBot chat turns — there is no automated paper-to-workflow extractor.

What This Module Does

• PDF upload — Accept a PDF, extract its text content, store it in a TTL-managed session
• DOI resolution — Resolve a DOI through CrossRef to get title, authors, abstract, and metadata
• Session storage — Hold paper text + metadata in memory keyed by session ID, with automatic cleanup after a TTL window
• Text retrieval — Return the extracted text for chat context augmentation

What This Module Does Not Do

• Automatic parameter extraction (functionals, k-points, cutoffs) from paper PDFs — this is a CatBot prompt-driven task, not a backend feature
• Automatic workflow generation from paper methods — users describe the paper's method in chat and let CatBot construct the workflow

Server API

All endpoints live under the /paper prefix:

Endpoint	Method	Description
/paper/upload	POST	Upload a PDF; returns a session_id and parsed metadata
/paper/{session_id}	GET	Retrieve session info (title, authors, page count, expiry)
/paper/{session_id}/text	GET	Get the extracted text body of the paper
/paper/resolve-doi	POST	Resolve a DOI via CrossRef; returns metadata without storing
/paper/{session_id}	DELETE	Manually clean up a session before TTL expiry

Data Model

PaperSessionInfo

• session_id — UUID for the in-memory session
• title — Paper title (from PDF metadata or first heading)
• authors — Author list (when extractable)
• created_at — Session creation timestamp
• expires_at — TTL expiry timestamp

DOIResolveResponse

Returns CrossRef metadata: title, authors, journal, publication year, abstract, and the resolved DOI URL.

Typical Flow

1. User drops a PDF into CatBot or pastes a DOI
2. Frontend posts to /paper/upload or /paper/resolve-doi and receives a session_id
3. The chat context includes the paper's text/metadata in subsequent LLM calls (see context.ts)
4. User asks CatBot to build a workflow based on the paper's method — CatBot drafts the workflow via the workflow tools
5. Session expires after TTL or is manually deleted

Related

• Paper Import Tutorial — Step-by-step user guide
• Chat System — How paper context flows into LLM calls
• Workflow Tools — The tools CatBot uses to construct workflows