Captain Nemo

Your own coding agent, running on NVIDIA Nemotron.

Captain Nemo is a small, inspectable workbench in the spirit of Claude Code and Codex. It streams replies, shows every file change as a diff you approve, and never calls any model but Nemotron. It runs on your machine with your own NVIDIA API key.

Version 0.1.0, 251 KB. Runs on Windows, macOS, and Linux with Node.js 22.19 or newer. No GPU needed: NVIDIA hosts the models and offers free trial access.

Captain Nemo badge: the captain with his octopus and the Nautilus
Captain Nemo in coding mode: the model read greet.js and proposed an edit, shown as a diff adding a return statement, with Approve and Decline buttons
A real session on Nemotron 3 Super. The agent listed the workspace, read the file, and proposed one targeted change. Nothing is written until you approve it.

What it does

The parts of a modern agent harness that matter, without the parts that phone home.

Streams as it thinks

Replies arrive token by token, with Nemotron's reasoning in a collapsible thought-process block.

Diffs you approve

Every edit is proposed as a unified diff. Approve it, decline it, or switch to Auto-edit when you trust the plan.

Real tools

Reads, searches, targeted edits, full writes, and shell commands in your project folder, each shown as a card you can expand.

Nemotron only

Four allow-listed model IDs and one fixed endpoint. No fallback provider, no router, no surprise vendors.

Private by default

Binds to localhost. Your key stays in memory unless you choose to remember it. History is a plain file you own.

Two modes, one place

A coding agent for your repo and a general assistant for everything else, in light and dark teal.

Get started

About two minutes from download to first reply.

  1. Create a free key at NVIDIA Build.
  2. Download the zip and unpack it anywhere.
  3. Install, build, and start with the commands shown here. The start script already handles machines where security software re-signs HTTPS.
  4. Open http://127.0.0.1:4317, paste your key in Settings, and point it at a project folder.
cd captainnemo
npm ci
npm run build
npm start

Models

Switch between turns to compare. Availability and limits depend on your NVIDIA account.

ModelBest forID
Nemotron 3 SuperA balanced starting pointnemotron-3-super-120b-a12b
Nemotron 3 UltraComplex coding and reasoningnemotron-3-ultra-550b-a55b
Nemotron 3.5 LightningFast everyday tasksnemotron-3.5-lightning-30b-a3b
Nemotron 3 NanoComparing the earlier generationnemotron-nano-3-30b-a3b

Limits, stated plainly

It is a personal, local tool. Know these before you rely on it.
  • No live web search or browsing in this version.
  • Approved shell commands are not sandboxed. They run with your user permissions, so review them.
  • Project content the model reads is sent to NVIDIA's API. Free access is subject to NVIDIA's trial terms and rate limits.
  • Not affiliated with or endorsed by NVIDIA.