Local-first AI paper workspace

Write papers where your code, figures, and experiments already live.

DraftPilot brings LaTeX writing, local PDF builds, review notes, rebuttal TODOs, VM codebases, and AI agents into one focused research workspace — running on your machine, against real files on disk.

Start setup See the workspace
draftpilot launch --paper ./paper --codebase ./repo --open CLI
  • Local-first
  • Any agent
  • No accounts
  • Any codebase
DraftPilot workspace showing LaTeX editing, change review controls, linked codebases, the paper agent, and a PDF preview.
Workflow

Three moves. One workspace.

The paper, the evidence, and the agent — under one roof. Here's the arc from typing to shipping.

01

Write like Overleaf

Source editor, PDF preview, SyncTeX jumps, autocomplete, comments, logs, and local latexmk builds.

main.tex
\section{Method}
\begin{figure}
  \includegraphics{fig1.pdf}
\end{figure}
SyncTeX ✓page 3
02

Work like a researcher

Figures, supplements, posters, slides, rebuttal TODOs, and per-project deadlines all live in the project folder.

✓Baseline experiments
✓Write ablations
Rebuttal draft
Poster QA
CVPR · 12 days
03

Ask agents with context

Codex or another local agent inspects the paper, codebase, errors, notes, and experiment outputs together.

youWhy does NeuMan diverge on scene 3?
agent Reading method.tex + ml-hugs/eval.py
The moat

Capabilities that make research handoffs disappear.

Bring any agent. Attach any codebase. Ship every deliverable from the same folder. Cloud editors can't catch up — because cloud editors can't reach your code.

Bring your own agent

DraftPilot is agent-neutral. Codex, Claude, local Ollama models, or a custom shell script — each one gets the same paper and codebase context.

C
CodexOpenAI · gpt-5
connected
A
ClaudeAnthropic · opus 4.7
connected
L
Localqwen-coder · 32b
connected
+
Custom agentany shell script

Attach any number of codebases

Point at one repo, or five. Local paths, remote VM paths, monorepos. Agents reason across the manuscript and the implementation — without you copy-pasting a thing.

ml-hugslocal · main
gaussian-splattinglocal · dev
datasetsvm · gpu-1

Zero context handoffs

Paper, code, PDFs, logs, comments, and agent chat all in one workspace. No shuttling snippets.

Live diff review of every AI edit

Agent edits appear inline as a colored diff. Accept, reject, or edit — all from the same editor, without leaving the file.

method.tex+2 −1
- The reconstruction assumes static geometry.
+ We reconstruct dynamic scenes with separate
+ human, object, and scene branches~\cite{ye2024}.
✓ Accept Reject

Auto slides & poster

Turn any paper into a conference deck or landscape poster with baposter.

One-click arXiv package

Strips line numbers & conf metadata, bundles the current .bbl, converts SVG → PDF.

Fix Build, then verify

Agent repairs LaTeX errors, re-runs the build, tells you if the paper is actually clean.

VM-native launch

SSH tunnel to the remote paper & code. Local URL, real remote paths, one command.

Deadlines & comments

Per-project deadlines, numbered comments, TODOs, rebuttal checklists — plus SyncTeX and autocomplete.

Why local-first

What the cloud can't reach.

DraftPilot doesn't try to replace every editor. It's for the moment when the paper, code, logs, and experiments need to stay connected — and stay on your machine.

Capability Cloud editors DraftPilot local-first
Familiar editor + PDF pane ✓Yes ✓Same feel, local files
Local file access ×Uploads only ✓Full disk access
Bring your own AI agent ×Locked to vendor ✓Codex, Claude, local, custom
Codebase context for agents ×Paper only ✓Any repo, any depth
VM experiments over SSH ×Not possible ✓SSH tunnel, real paths
Poster & slide generation ~Manual ✓Built-in, baposter
arXiv preprint packaging ~Hand-clean tarball ✓One click
No accounts, no upload ×Account required ✓Local-only by default
Read the mode guide no accounts · no database · telemetry opt-out
Download · v0.1.7

Install once. Point it at a paper. Keep writing.

Stable single-binary CLI for macOS and Linux, plus a Windows preview. Self-updating, checksum-verified.

macOS Apple Silicon · arm64
Stable 
$curl -fsSL https://jerrinbright.com/draftpilot/install.sh | bash
45 MB SHA-256 verified Self-updating
Linux glibc distros · x64
Recommended 
$curl -fsSL https://jerrinbright.com/draftpilot/install.sh | bash
50 MB SHA-256 verified Works on VMs
Windows PowerShell 5+ · x64
Preview 
>powershell -ExecutionPolicy Bypass -Command "irm https://jerrinbright.com/draftpilot/install.ps1 | iex"
45 MB SHA-256 verified Preview channel
Release notes Manifest · JSON SHA-256 checksums
v0.1.7 shipped · SHA-256 verified · auto-update via draftpilot update
Get started

Your next paper deserves better plumbing.

Install commands, LaTeX prerequisites, local setup, VM setup, updates, and storage details — all in one onboarding guide.

Open onboarding Quickstart
$ draftpilot launch --paper ./paper --codebase ./repo --open CLI