explaingit

pulsarosdevteam/pulsarcode

0PythonAudience · developerComplexity · 2/5ActiveLicenseSetup · moderate

TLDR

Launcher that points the Claude Code CLI at free NVIDIA NIM models like Kimi K2.6, DeepSeek V4 and Qwen 3 Coder instead of the paid Claude subscription.

Mindmap

mindmap
  root((pulsarcode))
    Inputs
      Free NVIDIA NIM API key
      Claude Code CLI
    Outputs
      Claude Code wired to NVIDIA models
      Local adapter on port 4000
    Use Cases
      Use Claude Code without a paid plan
      Try Kimi or DeepSeek for coding
      Pick a model from a TUI wizard
    Tech Stack
      Python
      NVIDIA NIM
      Claude Code
      Bash

Things people build with this

USE CASE 1

Run Claude Code against free NVIDIA-hosted models with no credit card

USE CASE 2

Switch between Kimi, Qwen, DeepSeek, Llama, and GPT-OSS from a TUI menu

USE CASE 3

Set up Claude Code on macOS or Linux with a single curl install

Tech stack

PythonNVIDIA NIMClaude CodeBash

Getting it running

Difficulty · moderate Time to first run · 30min

Needs a free NVIDIA NIM API key, Python 3.11+, and either Claude Code or Node.js already installed. Windows users go through WSL2.

AGPL-3.0: free to use and modify, but any networked service built on it must also share its source.

In plain English

pulsarcode is a small launcher that lets people run Anthropic's Claude Code, a popular command-line coding assistant, using free AI models hosted by NVIDIA instead of Anthropic's paid Claude models. The official Claude Code program is a free download but the Claude models it normally calls require a paid subscription. This project redirects those calls to NVIDIA's developer platform, where models like Kimi K2.6, Qwen 3 Coder, DeepSeek V4, Llama 3.3, and GPT-OSS 120B can be used with a free API key that does not require a credit card. Using it looks like this: you run a single command, and a wizard appears every time you launch. It first asks whether to keep or replace the stored NVIDIA NIM API key, then shows a menu of models grouped into a Recently Used section, a Coding Tier with the largest coding-tuned models, and a General Tier with mid-sized general models. You pick a model with arrow keys, press Enter, and the regular Claude Code interface opens with the chosen model wired in. The README shows the exact terminal screens you see during this flow. The project is for macOS 14 or newer or a recent Linux distribution, with Windows users needing WSL2. It requires Python 3.11 or newer and either Claude Code already installed or Node.js so the installer can pull Claude Code automatically. A free NVIDIA NIM key from build.nvidia.com is the other requirement, and the first-launch wizard walks you through generating one. Installation is a single curl-and-bash command that downloads a release tarball, sets up an isolated Python virtual environment, installs an adapter, and starts the launcher. The README spends a section explaining why the project exists. The authors point out that Claude Code's subscription, around 20 US dollars per month, is friction in high-income countries but multiple months of average salary in countries like Argentina, Vietnam, Nigeria, Tunisia, India, Indonesia, Egypt, Bangladesh, the Philippines, and Pakistan. They describe pulsarcode as a donation to developers who cannot or will not pay a monthly subscription for a frontier coding agent. Technically, the launcher uses an environment variable called ANTHROPIC_BASE_URL that Claude Code already supports for routing to alternative backends like AWS Bedrock or Google Vertex. pulsarcode points it at a small local adapter on 127.0.0.1 port 4000, which translates Anthropic's message format into NVIDIA's OpenAI-compatible chat completions and forwards them. The README states the project is AGPL-3.0, source-available, made by a Canadian federal corporation that earns nothing from it, and ships with no analytics or telemetry.

Copy-paste prompts

Prompt 1
Walk me through installing pulsarcode on Ubuntu 24 with Python 3.12
Prompt 2
Explain how pulsarcode uses ANTHROPIC_BASE_URL to route Claude Code calls to NVIDIA NIM
Prompt 3
Add a new model option to the pulsarcode TUI wizard, for example GLM-5.1
Prompt 4
Show me how the local adapter on 127.0.0.1 port 4000 translates Anthropic messages to NVIDIA chat completions
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.