TeXRA

Appearance

Research Tools ​

You're deep in a manuscript and realise you need to cite "that attention paper from 2017" but can't remember the title. Or you want to verify an integral in your appendix. Or you need to pull twenty BibTeX entries from your Zotero library into a new project. TeXRA's research agents handle all of this without leaving your editor — in VS Code or the texra CLI.

What You Can Do ​

Find Papers ​

Ask a literature-search agent — search, chat, review, or presenter — to find papers. TeXRA queries arXiv (preprints) and Crossref (published works) automatically:

Find recent papers on transformer architectures for document understanding.
Focus on work from 2023-2024 that handles mathematical equations.

User story: A postdoc is writing the related-work section of a NeurIPS submission. She opens the search agent () and asks for papers on "efficient self-attention for long documents." In seconds she has a table of relevant arXiv preprints with titles, authors, and abstracts — ready to cite.

searchresults
efficient self-attention for long documents
  • Longformer: The Long-Document Transformer
    Beltagy, Peters, Cohan
    arXiv·arXiv:2004.05150
  • Efficient Attention: Attention with Linear Complexities
    Shen, Zhang, Zhao, Yi, Li
    arXiv·arXiv:1812.01243
  • Rethinking Attention with Performers
    Choromanski et al.
    Crossref·10.48550/arXiv.2009.14794
Every result is a real arXiv or Crossref lookup — never fabricated.

A typical search result set: each preprint carries its title, authors, an arXiv or Crossref source tag, and a one-click Cite — every row a real lookup, never fabricated.

Look Up Citations ​

Give a DOI or arXiv ID to get full bibliographic details:

Get the citation info for arxiv:2401.12345
Look up DOI 10.1038/nature12373

Behind the scenes this calls the arxiv_metadata and crossref_doi tools.

Download Paper Sources ​

For deeper analysis, ask to download the LaTeX source of a paper:

Download the source files for arxiv:2401.12345 so I can see how they made their figures.

This uses the download_arxiv_source tool and lands the files in your workspace.

Search the Web ​

For documentation, project pages, or general info:

Find the official PyTorch documentation for attention mechanisms.

The web_search tool prefers the active provider's native search (Anthropic, OpenAI, Gemini) and falls back to DuckDuckGo Instant Answers. web_fetch retrieves a specific URL and extracts its main content.

Manage References with Zotero ​

If you use Zotero with the Better BibTeX plugin, TeXRA can search, export, and add items to your library directly. Make sure Zotero is running while you use these features — check status on Dashboard → Integrations ().

Search my Zotero library for papers by Vaswani on attention mechanisms.
Export the selected Zotero items as a .bib file for my project.
Add this arXiv paper to my Zotero library.

User story: A PhD student is collecting references for a thesis chapter. She asks the search agent to find key papers on graph neural networks, then says "add these to my Zotero and export them to references.bib." The agent handles the lookup, adds entries to her Zotero library, and writes the BibTeX file — all in one conversation.

zoterotool calls · this run
  • zotero_searchgraph neural networksFinds matching items in your library
  • zotero_addarxiv:2401.12345Adds the paper to your Zotero library
  • zotero_exportreferences.bibWrites the selected items as BibTeX

How the search agent drives the zotero_* tools across one conversation — search → add → export — as the calls surface in the Progress view.

Default Bibliography Path

Set a default location for Zotero exports so agents always know where to save bibliography entries. The setting key is texra.bib.defaultPath — configure it in your .texra/config.json (CLI) or VS Code settings.

Verify Math with Wolfram ​

The research agent can call wolfram to run Wolfram Language code and check symbolic algebra, integrals, or limits before you commit them to the manuscript. Requires a local Wolfram Engine; status shows on Dashboard → Tools () → Computation ().

External Inquiry ​

The inquiry tool lets a TeXRA agent ask one question in an external chat (ChatGPT, Claude, Gemini) through a copy/paste flow, then resume with the answer. Dispatch is non-blocking: the agent's cycle continues while you fetch the answer, and resumes automatically once you paste it back (even after a reload). No API key required — it uses your existing subscription. In the CLI, the same flow appears as a terminal modal: paste the prepared question into your chat subscription, then paste the answer back into TeXRA.

External Inquiry No API key
Question for external chat copy out
Does the spectral-gap bound in Lemma 3 still hold when the operator is only essentially self-adjoint? Cite a standard reference.
ChatGPT Claude Gemini
Paste the answer paste back
Paste the chat's reply here…
Dispatch is non-blocking — the agent's cycle continues while you fetch the answer, and resumes automatically once you paste it back.

The inquiry copy-out / paste-back loop: TeXRA prepares a question, you copy it into ChatGPT, Claude, or Gemini, then paste the reply back to resume the run — no API key needed.

Which Agent to Use ​

Three research agents, each tuned for a different stage of the work — pick one from the Agent dropdown ():

Agent
search
Research
searchtool-use
researchtool-use
discusstool-use
searchdefault

Finding papers, literature reviews, fact-checking.

arxiv_metadatacrossref_doiweb_searchweb_fetchzotero
research

Computational verification, plus bash, file edits, and local LaTeX analysis.

wolframbashfile-edittexcountextract_figures
discuss

Brainstorming research directions with literature context.

web_searcharxiv_metadatainquiry

Each research agent and the tools it has enabled — check Dashboard → Agents () for the exact set.

Next Steps ​