Papilot

Papilot is an open-source, MIT-licensed local workspace from PapilotAI for writing and revising LaTeX papers with AI agents. Authors mark up passages and equations on the compiled PDF, send the notes to a coding agent, and review the typeset before-and-after changes. It is available as a macOS and Linux desktop app and as an npm CLI (version 0.8.7).

What It Is

Papilot is a PDF-centered revision tool for LaTeX manuscripts. Instead of editing source, you select a sentence or box an equation and describe what should change. SyncTeX maps the note to the exact LaTeX lines, and the agent edits a separate copy of the source. Papilot builds with your own latexmk and TeX distribution, so existing packages, BibTeX and TikZ figures keep working.

Review and Git Workflow

Each task runs in its own Git worktree, and the manuscript changes only when you accept an edit. Changes come back as typeset diffs where you can accept, reject, comment on, or undo each one. You can then commit to the manuscript as one commit, or apply edits to the workspace uncommitted. A Git panel supports staging, commits, pull (fast-forward only), push, and a commit graph showing task worktrees. If the same passage changed in the meantime, AI resolves conflicts and asks only when the intended content is unclear.

Agent Support

You choose an agent and model per task. Supported agents are Codex, Claude, Cursor, GitHub Copilot, Grok, Antigravity, Kimi, GLM, DeepSeek, Qwen Code and OpenCode. Papilot uses each agent's local CLI sign-in or API credentials you supply, and states it does not store credentials. Two tasks run at once while others queue.

Setup Path

Papilot requires Node.js 22.22+, Git, and a full TeX distribution with latexmk and synctex, plus at least one configured agent. Running papilot demo opens a bundled example paper at 127.0.0.1:7274. You can add your own paper from a LaTeX folder or an Overleaf ZIP. The server listens only on 127.0.0.1, and can be published over a tailnet with Tailscale Serve for phone or tablet review. Skills, custom skills and file or image attachments are also supported.