Nemo Code by ClawdWorks

Want these models inside your existing Claude Code install instead? Keep your memories, CLAUDE.md, MCP servers, hooks, and local workflow while swapping model backends with AnyModel.

Free AI coding agent. Zero subscription. One command.

Nemo Code gives you a Claude Code-style CLI experience - tools, file editing, shell commands, MCP servers, autocompact-style workflows, and optional Telegram access - powered by open/free model backends such as NVIDIA-hosted OSS models.

Built around:

• Claude Code CLI as a runtime dependency
• LiteLLM as the local Anthropic-compatible proxy
• NVIDIA NIM hosted model endpoints

Nemo Code does not bundle or redistribute Claude Code.

Quick Start

Mac / Linux / Git Bash

curl -fsSL https://raw.githubusercontent.com/kevdogg102396-afk/free-claude-code/master/install.sh | bash

Windows PowerShell

irm https://raw.githubusercontent.com/kevdogg102396-afk/free-claude-code/master/install.ps1 | iex

Then open a new terminal and run:

clawdworks

2026 Model Slots

Nemo maps Claude Code's model slots to current NVIDIA NIM model IDs. Switch in the TUI with /model.

Claude Code slot	NVIDIA model ID	Best for
Sonnet (default)	moonshotai/kimi-k2.6	Current top coding/default model
Opus	qwen/qwen3.5-397b-a17b	Large reasoning and complex tasks
Haiku	minimaxai/minimax-m2.7	Fast agentic coding and document work

Additional selectable install-time models:

• z-ai/glm5.1
• nvidia/nemotron-3-super-120b-a12b
• openai/gpt-oss-120b

Model availability can change on NVIDIA's side. If a model disappears or rate limits heavily, rerun clawdworks models or set NEMO_MODEL=<model-id>.

NEMO_MODEL controls the Sonnet/default route. Opus and Haiku stay mapped to the fixed large/fast fallback models above.

Install Modes

Docker (sandboxed, recommended)

Runs in a container and keeps work inside /workspace.

• Requires Docker
• Safer for personal machines
• Command: clawdworks
• Telegram bridge included: TELEGRAM_BOT_TOKEN=xxx clawdworks-telegram

Local (full power)

Runs directly on your machine.

• Requires Node.js 18+ and Python 3.8+
• Full filesystem access
• Browser automation and local MCP support if installed
• Command: clawdworks

Use local mode on a dedicated dev box or a machine where you are comfortable giving an agent broad access.

Commands

clawdworks
clawdworks run "fix the bug in app.js"
clawdworks models
clawdworks help

Docker Telegram:

TELEGRAM_BOT_TOKEN="your-token" clawdworks-telegram

Optional Telegram allowlist:

TELEGRAM_ALLOWED_CHAT_IDS="123456789,987654321" TELEGRAM_BOT_TOKEN="your-token" clawdworks-telegram

Telegram Bridge

The Docker path includes a one-liner Telegram bridge for phone access.

Features:

• Conversation memory for the last 20 exchanges
• Typing indicators during long runs
• No fixed Claude execution timeout
• Optional chat ID allowlist via TELEGRAM_ALLOWED_CHAT_IDS
• /sonnet, /opus, /haiku, and /model commands

How It Works

You -> Claude Code CLI -> LiteLLM proxy -> NVIDIA NIM -> open/free model

1. Claude Code sends Anthropic-style requests.
2. LiteLLM receives them on 127.0.0.1:4000.
3. LiteLLM maps Claude model IDs to NVIDIA NIM model IDs.
4. NVIDIA serves the selected model.

The current Claude Code aliases observed in May 2026 are:

• sonnet -> claude-sonnet-4-6
• opus -> claude-opus-4-7
• haiku -> claude-haiku-4-5-20251001

The proxy config maps those IDs to the NVIDIA models above.

Requirements

Docker mode:

• Docker Desktop or Docker Engine
• NVIDIA API key from build.nvidia.com

Local mode:

• Node.js 18+
• Python 3.8+
• NVIDIA API key from build.nvidia.com

Windows:

• Native PowerShell install is supported
• Git Bash is supported
• WSL can run the shell installer

Security And Disclaimers

Free models are not Claude

Open/free models available through NVIDIA NIM can be powerful, but they may be more vulnerable to prompt injection than Claude. Treat local agent access with care.

Risks include:

• Prompt injection from malicious files, websites, or pasted content
• Data exfiltration from files the agent can read
• Destructive command execution in local mode
• Accidental filesystem or OS damage

Recommendations:

1. Use Docker mode on personal machines.
2. Use local mode on dedicated development machines.
3. Avoid feeding untrusted content to a full-access agent.
4. Review commands before approving them.

Not affiliated with Anthropic

Nemo Code is built on top of the Claude Code CLI by Anthropic. The Claude Code CLI is proprietary software distributed under Anthropic's terms. Nemo Code is not affiliated with, endorsed by, or sponsored by Anthropic.

No warranty

Nemo Code is provided as-is, without warranty. Use it at your own risk.

Credits

• Claude Code CLI by Anthropic
• LiteLLM by BerriAI
• NVIDIA NIM
• ClawdWorks - Kevin Cline + Claude

License

The Nemo Code wrapper, proxy config, splash screen, install scripts, and Telegram bridge are MIT licensed. See LICENSE.

Nemo Code requires the Claude Code CLI (@anthropic-ai/claude-code), which is proprietary software by Anthropic PBC. Your use of the Claude Code CLI is subject to Anthropic's terms. Nemo Code installs the CLI as a runtime dependency via npm and does not bundle or redistribute it.