kosmos-code 0.1.0

AI coding agent that runs in your terminal. Describe what you want to build and Kosmos creates it.

Kosmos CLI

AI coding agent that runs in your terminal. Describe what you want to build and Kosmos creates the project for you -- scaffolding, dependencies, code, and a live dev server preview.

Install

pip install kosmos-code

Or install from source:

git clone https://github.com/meryboth/kosmos-cli.git
cd kosmos-cli
pip install -e .

Quick start

# First run will ask you to configure your LLM provider
kosmos

# Or jump straight in with a prompt
kosmos "Create a todo app with React"

Configuration

Kosmos stores your config at ~/.kosmos/config.json. API keys are saved with restricted file permissions (600).

kosmos config          # show current provider, model, and masked API key
kosmos config setup    # interactive wizard to set provider + API key
kosmos config model    # change model
kosmos config reset    # remove all stored configuration

Supported providers

Provider	Models	Get a key
OpenAI	gpt-5.2, gpt-5-mini, gpt-5-nano, gpt-4.1, gpt-4.1-mini, o3, o4-mini	platform.openai.com/api-keys
Anthropic (via OpenRouter)	claude-opus-4.6, claude-sonnet-4.6, claude-haiku-4.5	openrouter.ai/keys
Google Gemini (via OpenRouter)	gemini-2.5-pro, gemini-2.5-flash	openrouter.ai/keys
Groq	llama-4-scout, gpt-oss-120b, llama-3.3-70b	console.groq.com/keys
OpenRouter	400+ models from all providers	openrouter.ai/keys

Smart model routing

Select Auto during setup (the default) and Kosmos picks the best model per task:

Tier	When	Example (OpenAI)
Light	Simple edits, installs, running tests	gpt-4.1-nano
Standard	Create a component, add a feature	gpt-5-mini
Heavy	Build full app, complex debugging	gpt-5.2

Tool-calling iterations (reading files, running commands) automatically use the light model to save tokens. Use --verbose to see model selection in real time.

Usage

kosmos                                     # interactive mode
kosmos "Create a landing page"             # single prompt
kosmos --verbose                           # show token usage and model routing
kosmos --dir ~/my-project "Fix the tests"  # work on an existing directory

When you run kosmos without --dir, you choose:

1. New project -- enter a name, pick a stack, start building
2. Continue existing project -- pick from ~/Documents/Kosmos/Projects

Supported stacks

Python, JavaScript (Node), React (Vite), Next.js, TypeScript (Node), Express API, FastAPI, or describe your own.

What can Kosmos do?

• Explore files -- list directories and read source code
• Write files -- create or overwrite any file in the project
• Run scripts -- execute .py, .js, .ts, .sh, .rb, .go, .php
• Run shell commands -- npm install, pip install, git init, etc.
• Start a dev server -- launches in background and returns the localhost URL

Project structure

kosmos-cli/
├── main.py            # CLI entry point, agent loop, retry logic
├── prompts.py         # LLM system prompt
├── call_function.py   # Tool dispatcher
├── config.py          # Config storage (~/.kosmos/config.json)
├── providers.py       # LLM provider definitions and client factory
├── cli_config.py      # kosmos config subcommands
├── router.py          # Smart model routing (task classification)
├── functions/
│   ├── get_files_info.py
│   ├── get_file_content.py
│   ├── write_file.py
│   ├── run_file.py
│   ├── run_command.py
│   └── start_server.py
├── pyproject.toml
└── LICENSE

Requirements

• Python >= 3.12
• An API key from any supported provider
• Node.js (optional, for JS/React/Next.js projects)

License

MIT