TeXRA

Appearance

File management ​

A scientific research project encompasses diverse assets: manuscripts, derivations, appendices, computational scripts, figures, bibliographies, and auxiliary style files. This guide explains how TeXRA organizes, scopes, and manages these files during agent execution.

File categories ​

TeXRA organizes files into three categories, each with its own accepted file types and read/write role:

InputRead + edit

Primary documents the agent reads and rewrites.

.tex.txt.md
ContextRead-only

Reference material the agent sees but never modifies.

.bib.bbl.sty.clsreference papers
MediaVisual / audio

Figures and recordings handed to vision/audio models.

imagesPDF figuresaudio

Three categories: Input is read + edited, Context is read-only, Media carries visual and audio content.

Input files ​

These are the primary files the agent reads and edits. They typically include:

  • LaTeX documents (.tex)
  • Text files (.txt)
  • Markdown files (.md)

Input files can be a single main document or several related files (such as chapters of a book or sections of a paper).

Context files ​

Read-only files the agent sees but does not modify. They can be:

  • Bibliographies (.bib/.bbl)
  • Style and macro files (.sty/.cls)
  • Reference papers or previous versions
  • Any document the output should match for formatting

Figure & media files ​

Files containing visual or audio content, such as images, PDFs used as figures, or audio recordings. Read the Working with figures guide for supported types, UI controls, and automatic extraction.

Pasting images from clipboard ​

You can paste images from your clipboard directly into the instruction text area. This is useful for including screenshots, diagrams, or other visual content:

  1. Copy an image to your clipboard (for example, take a screenshot or copy from another application)
  2. Place the cursor in the instruction text area where you want to reference the image
  3. Paste (Ctrl/Cmd+V). The image is then:
    • Saved to the workspace storage
    • Referenced in the text as [pasted_<timestamp>_<rand>.ext]
    • Added to the Media Files list
    • Made available to the AI model

Pasted images appear in the file selector like any other media file, so you can review or remove them before sending. If the selected model cannot read images, TeXRA warns you so you can switch to a vision-capable model.

Clipboard paste accepts many image formats (JPEG, PNG, GIF, WebP, BMP, SVG, TIFF, HEIC, HEIF, AVIF, PSD), but the formats that can be processed depend on what the selected AI model supports. Most vision models support common formats like JPEG, PNG, GIF, and WebP.

Note: Pasted images are stored in workspace storage and removed automatically after 3 days to save space.

File selection interface ​

The TeXRA interface has a section for each file category (Input , Context , Media ):

draft.tex — texra-paper
\documentclass{article}
\usepackage{preamble} % from Context
\begin{document}
\input{sections/intro}
\includegraphics{figure1.pdf} % from Media
\bibliography{references} % from Context
\end{document}

Each category holds an ordered list of files with three quick actions in the header.

File list controls ​

Each category (Input, Context, and Media) manages a list of files; a single-file workflow is a list of length one. The header has three buttons:

  • Add opened files (): Append every editor tab whose extension matches the category to the list.
  • Clear all files (): Empty the list.
  • Add files (): Open a file picker to append files.

You can also drag and drop files from the OS file manager (or from VS Code's Explorer) onto a category to add them.

Inside the list:

  • Each file row has a small trash icon to remove that file.
  • Drag a row to reorder files. Input files are sent to the agent in this order.

The list is the only mode; there is no separate "single file" view. To work with one file, keep the list at length one.

This works well for:

  • Processing several chapters of a book
  • Working with documents split across several files
  • Batch processing similar documents
  • Including several reference materials

Output files ​

For document tasks that edit documents, output filenames are the selected input filenames in the same order. Agents that create fixed new files declare those names in their YAML with task.outputs.

File path handling ​

TeXRA handles file paths as follows:

Relative vs. absolute paths ​

  • Display: Files are displayed with paths relative to the workspace root
  • Processing: TeXRA resolves paths to their absolute form when needed
  • Output: Document task outputs are saved in run storage. Use Accept or Pack when you want to copy reviewed outputs back into the workspace.

File discovery rules ​

TeXRA uses built-in file extensions and exclusions when discovering inputs, context, edited files, and media. Read the file discovery section of the configuration guide.

Output file naming ​

TeXRA stores document task outputs in the run folder under run storage. Within that folder, the revised document keeps your input filename in every revision (the first revision in r0/, the second in r1/, and so on):

r{round}/<input-filename>

Subdirectories are preserved, so chapters/intro.tex lands at r0/chapters/intro.tex. The only fixed name is the raw model response, r{round}/output.xml.

For example:

  • Input: paper.tex
  • Agent: polish
  • Model: anthropic/claude-sonnet-5-5
  • Output: r0/paper.tex

When the agent's task has a second (reflection) revision, you may also see:

  • Revision 2: r1/paper.tex

File management commands ​

TeXRA provides several commands for managing generated files, available from the main interface or the ProgressBoard:

Pack ​

The "Pack" button () snapshots the run folder into a structured history folder:

  1. Creates a timestamped directory in the "History" folder
  2. Copies all relevant output files, logs, and mirrored dependencies
  3. Preserves the relationship between input and output files

Use this to keep a clean workspace while preserving previous outputs.

Clean ​

The "Clean" button () removes output files for the selected run:

  1. Identifies the run folder for the current run
  2. Removes generated artifacts from the run folder
  3. Leaves original input files untouched

Use this to remove generated artifacts from run storage after reviewing the results.

Opening generated files ​

Document task outputs are listed in the ProgressBoard under Generated Files. Select a file name to preview it in VS Code. Files open using VS Code's default viewer, so PDFs and images display correctly while .tex documents open in the editor.

To browse the whole run folder, use the

Open in run storage toolbar

button. This reveals the run folder with generated files, compile logs, mirrored LaTeX dependencies, and intermediate artifacts. (From a terminal, texra history show <id> lists the same stored artifacts; see the card below.)

Run storage ​

Every run gets an isolated run folder under TeXRA's workspace storage directory:

text
executions/<executionId>/

Document task outputs are written there first, not directly over your workspace files. Three commands then move that run's artifacts to three different places:

Run storageisolated
executions/<id>/
r0/output.tex
r0/output.log
r1/output.tex
written here first — never over your workspace
Accept Workspacecopies reviewed outputs back in
Pack History/archives the whole run, timestamped
Clean Deletedremoves the run folder; inputs untouched

Accept copies reviewed outputs into the workspace, Pack archives the run into History, and Clean deletes the run folder. Your input files are never touched.

The CLI follows the same storage-first lifecycle: it reads and writes the same executions/<executionId>/ run store, and --output plays the role of Accept:

texra run
$texra run polish --input paper.tex
…/executions/9f3a6c81d24e/r1/paper.tex
outputs land in the run's storage folder first
$texra run polish --input paper.tex --output paper.polished.tex
paper.polished.tex
--output copies the final artifact next to your input
$texra history show 9f3a6c81d24e
… Execution / Status / Agent / Model details …
Files (2):
4128r0/paper.tex
4310r1/paper.tex

Three beats of the same lifecycle: storage-first output, --output as Accept, and history show to browse a run's stored files.

The folder also stores intermediate artifacts such as optional debug JSON files written when texra.debug.saveModelIO is enabled.

Working with LaTeX projects ​

For LaTeX projects with many files and dependencies:

LaTeX directory structure ​

TeXRA works well with standard LaTeX project structures:

project/
├── main.tex            # Main document
├── chapters/           # Chapter files
│   ├── intro.tex       # Introduction
│   └── methodology.tex # Methodology
├── figures/            # Figure directory
│   ├── diagram.pdf     # Figure file
│   └── graph.png       # Another figure
└── bibliography.bib    # Bibliography

Input file detection ​

TeXRA detects appropriate input files based on:

  1. File extension (.tex, .txt, .md)
  2. Location in project hierarchy
  3. Content structure

LaTeX Workshop integration ​

If a LaTeX workspace does not have the LaTeX Workshop extension installed, TeXRA offers to install it. TeXRA does not rewrite editor or LaTeX Workshop preferences during startup. Use TeXRA's LaTeX settings page to inspect and apply the available workspace recommendations explicitly.

Cross-computer syncing ​

If you work on several computers, use a cloud storage service like Dropbox to sync the History folder, which holds packed versions of your documents.

To keep your local directory structure while syncing this folder, use symbolic links. Store the actual folder in Dropbox and create a symbolic link in your local project directory. For example:

bash
ln -s /path/to/Dropbox/texra-projects/ProjectName/History /path/to/local/ProjectName/History

Replace /path/to/Dropbox and /path/to/local with your actual Dropbox and local project paths.

Best practices ​

Organizing input files ​

  • Main document first: When using several input files, list the main document first
  • Logical order: Arrange chapter files in reading order
  • Consistency: Keep file naming conventions consistent

Managing output files ​

  • Regular cleanup: Use the "Clean" command to remove unneeded outputs
  • Version control: Use "Pack" to preserve important milestones
  • Diff review: Use LaTeXdiff to review changes before accepting them

Reference materials ​

  • Relevant examples: Include only directly relevant reference files
  • Context limits: Keep model context limits in mind when adding references
  • Format consistency: Use reference materials with similar formatting styles

Next steps ​

Now that you know how to manage files in TeXRA, you may want to read about: