Introduction

When revising papers in DeepSeek Harness (dsh), a common problem is that PDF annotations, LaTeX source editing, compilation results, and git history are scattered across different steps. PaperLab (maple-pwn/paperlab) places these steps in the same workspace: select text on the rendered PDF to write annotations, and then the dsh agent modifies the source code based on the annotations, checks compilation, and commits to git. PaperLab’s AI execution engine runs entirely within DeepSeek Harness; the workspace never directly calls model APIs.

What is it

PaperLab is an Overleaf-style paper iteration and revision workspace, with the AI engine being a DeepSeek Harness plugin. The repository is maintained by maple-pwn, licensed under MIT © 2026 PaperLab contributors.

It solves a very specific problem: enabling “read annotations, revise paper, view real PDF, and rollback versions” to be completed within the same workflow, while delegating model execution, tool calls, and version management to the dsh agent workflow.

Core Features

Interface and Annotations

  • Overleaf-style three-column layout with draggable panels.
  • Select any text on the PDF to write feedback.
  • Precise placement of yellow highlights based on page coordinates.
  • Real PDF rendering for consistent typesetting of formulas and figures.
  • Source code editing with automatic saving and compilation.

AI Revision

AI revision is based on the dsh agent, the process is:

Read Annotations → Unique Match Edit → Compilation Check → Git Commit

The workspace provides a wide-screen trace, live-streaming the thinking process, tool calls, and outputs, complete with timing and LIVE status. You can also switch models, thinking intensity, and orchestration modes within the workspace.

Version Management and Engineering Capabilities

  • Each project is a git repository, AI revisions automatically create checkpoints.
  • Supports history browsing, arbitrary diffs, and one-click revert.
  • One-click export: source files (with annotations) / PDF / git history.
  • Auto-detects XeLaTeX, respects latexmkrc, and provides a multi-compilation engine fallback chain.

Installation and Usage

First, confirm the prerequisites:

Node.js ≥ 20
pnpm ≥ 10
Python ≥ 3.11
git
DeepSeek Harness (dsh)
LaTeX optional

Install the plugin to the paperlab profile:

dsh plugin --profile paperlab add github:maple-pwn/paperlab

The repository root is a dsh bundle compliant with official standards, containing:

dsh.bundle
cordis.patch.yml
lib/

Note: Installing a plugin means running third-party code. Please read the source code first and test in an environment where you do not save keys.

Typical Usage

First, clone and enter the repository:

git clone https://github.com/maple-pwn/paperlab.git
cd paperlab

Then start the workspace. You can run the startup script:

./start.sh

You can also run the Make target:

make start

start.sh automatically completes: install dependencies → build → install plugin to dsh profile → start dsh → start workspace.

After starting, open the browser workspace:

http://127.0.0.1:8210

Model credentials belong to dsh. Configure them in the “Models” page of the dsh Web UI:

http://127.0.0.1:3080

PaperLab does not touch any API key.

If you need to verify the build or test, you can run:

make test

Use Cases and Notes

PaperLab is suitable for scenarios like:

  • Need to iterate on LaTeX papers in the browser.
  • Want to make annotations on the real PDF rather than just editing the source code.
  • Want AI modifications to include compilation checks, git commits, and rollback records.
  • Want model credentials and model calls to stay on the dsh side.

Usage notes:

  • PaperLab’s AI execution engine runs entirely within DeepSeek Harness; the workspace never directly calls model APIs.
  • When the plugin runs, it operates within the permission context of the current dsh process; you should check the source code, license, and runtime environment before installing.
  • Model credentials are configured in the dsh “Models” page; do not put keys into the PaperLab project.
  • LaTeX is an optional prerequisite; actual compilation capabilities still depend on the local environment.

Related Resources

GitHub Repository:

https://github.com/maple-pwn/paperlab