Nemo Code — Free Claude Code CLI alternative (NVIDIA open models, Docker sandboxed)

[kevdogg102396-afk]

Hey everyone! Built something I think the community will find useful.

How it works: Claude Code CLI (Apache 2.0) + LiteLLM proxy + NVIDIA NIM free tier. Same tools, same framework, zero cost.

Features:

• One-command install with interactive setup wizard
• • Docker sandboxed mode (secure) or Local mode (full power)
• • 6 free NVIDIA models to choose from
• • Telegram bridge with conversation memory
• • MCP servers included (fetch, memory, filesystem)
• • Works on Windows (WSL), Mac, and Linux
    Repo: https://github.com/kevdogg102396-afk/free-claude-code

Would love to be considered for the Alternative Clients section of the awesome list. Happy to answer any questions!

[hesreallyhim]

I have a question.

[kevdogg102396-afk]

Hey! Sure thing — what's your question? Happy to help with anything about Nemo Code. 🙂

[dinosauria123]

Why did you close the repository ?

[kevdogg102396-afk]

Hey! Yeah sorry about that — I realized I was wrong about the licensing. The open models are fine to use but wrapping them in the Claude Code CLI as a stand alone product was a grey area I wasn't comfortable with anymore, so I pulled it down.

But honestly it led me to something way better. I built a proxy that lets you use YOUR Claude Code with any model. So you keep all your memories, CLAUDE.md, MCP servers, hooks, everything you've already set up — and just swap the brain.

It's called AnyModel. You can switch between Opus, Codex, Grok, Kimi K2.5, Gemini, whatever mid-session, same way you'd switch from Opus to Sonnet. Your 4 model slots each point wherever you want.

The best part — NVIDIA NIM has a ton of models completely free. Qwen 3.5, Kimi K2.5, DeepSeek, Nemotron, MiniMax. So you can run Claude Code's full workflow without spending a dime on API costs when you don't need the heavy hitters.

$5 on Gumroad: https://kevster527.gumroad.com/l/anymodel

28+ models, 9 providers, MCP tools to switch on the fly

[dinosauria123]

LiteLLM is dangerous now.

BerriAI/litellm#24512

[kevdogg102396-afk]

Yeah I saw that issue — that's exactly why AnyModel doesn't use LiteLLM. It's a pure Node.js proxy with zero dependencies. No third-party libraries in the middle handling your API keys or requests. Everything runs locally on your machine, nothing phones home, and you can read every line of the ~1,500 lines of source code yourself.

The whole point was keeping it simple and transparent. Just a direct translation layer between Anthropic and OpenAI formats, no bloated middleware.

Side note — if you still have Nemo Code, you're fine as long as you don't manually update LiteLLM. It was pinned to a safe version and locked from auto-updating.

[kevdogg102396-afk]

Hey just opened the repo back up — it's public again! Go grab it before someone sends me a letter lol. https://github.com/kevdogg102396-afk/nemo-code