Skip to content

Getting started

Install

On Linux or macOS, install the latest release into ~/.local/bin:

curl -fsSL https://smeltagent.dev/install.sh | sh

The script verifies the archive against the release SHA256SUMS. Set SMELT_VERSION=v0.6.0 to pin a release or SMELT_INSTALL_DIR to install somewhere else.

Grab the latest binary for Linux or macOS (x86_64 or aarch64), or Windows (x86_64), from GitHub Releases.

Linux and macOS:

tar xzf smelt-*.tar.gz
sudo mv smelt /usr/local/bin/

Windows PowerShell:

tar xzf smelt-x86_64-pc-windows-msvc.tar.gz
New-Item -ItemType Directory -Force "$HOME\.local\bin"
Move-Item .\smelt.exe "$HOME\.local\bin\smelt.exe"
# Add $HOME\.local\bin to your user PATH if it is not already present.

cargo install --locked --git https://github.com/leonardcser/smelt.git smelt-agent

Run

Just run smelt. The first launch opens a provider wizard. Subscription providers log in with OAuth and can be used immediately; API-key providers write a starter ~/.config/smelt/init.lua (%APPDATA%\smelt\init.lua on Windows).

You can also skip the wizard and configure everything from the command line:

smelt --model gpt-5.5 --api-base https://api.openai.com/v1 --api-key-env OPENAI_API_KEY
smelt --headless --model gpt-5.5 --api-base https://api.openai.com/v1 \
  --api-key-env OPENAI_API_KEY "summarize @README.md"
smelt --resume              # open the saved-session picker

See the CLI Reference for every flag.

Subscription providers

These use your existing subscription and require no API key. Run smelt auth once, pick the provider, then start smelt with no extra flags:

smelt auth                # one-time browser or device-code login
smelt

smelt auth                # device-code login
smelt

After login, smelt registers the kimi-code provider automatically. Start smelt with no flags to use it.

smelt auth                # device-code login
smelt

Every model your Copilot account exposes (Claude, GPT, Grok, …) is available immediately.

API-key providers

These need --model and an environment variable or config file:

export OPENAI_API_KEY=...
smelt --model gpt-5.5 \
      --api-base https://api.openai.com/v1 \
      --api-key-env OPENAI_API_KEY

OpenRouter and other OpenAI-compatible services follow the same shape; set --api-base to the provider's base URL and swap the model name.

export ANTHROPIC_API_KEY=...
smelt --model claude-opus-4-8 \
      --api-base https://api.anthropic.com/v1 \
      --api-key-env ANTHROPIC_API_KEY

ollama pull qwen3.6:27b
smelt --model qwen3.6:27b --api-base http://localhost:11434/v1

Any OpenAI-compatible server works (Ollama, vLLM, SGLang, llama.cpp).

Your first prompt

Type a request and press Enter. Point the agent at files with @, as in explain how @src/pricing.rs applies a discount:

The agent reads freely but asks before it edits files or runs commands. A dialog shows what it wants to do. Choose allow once, allow for the session or workspace, or deny (Esc).

A few keys to know:

  • Shift+Tab cycles the mode (Normal, Plan, Apply, Yolo). The status bar shows the current one.
  • Keep typing while it works: Enter queues a message, Ctrl+Enter steers.
  • Esc Esc cancels, or rewinds when idle. / opens the command picker and F1 shows help.

Save your config

Once you have a setup you like, write it to ~/.config/smelt/init.lua (%APPDATA%\smelt\init.lua on Windows) and run plain smelt from then on:

smelt.provider.register("ollama", {
  type = "openai-compatible",
  api_base = "http://localhost:11434/v1",
  models = { "qwen3.6:27b" },
})

smelt.provider.register("openai", {
  type = "openai",
  api_base = "https://api.openai.com/v1",
  api_key_env = "OPENAI_API_KEY",
  models = { "gpt-5.5" },
})

smelt.settings.vim = true
smelt.settings.file_icons = true -- needs a Nerd Font

Switch models at any time with /model. Config changes reload automatically.

Just ask the agent

You rarely need to edit config by hand. Ask: "switch to the ollama provider", "make the accent color orange" or "add a /standup command". The built-in customize skill edits your init.lua and reloads it.

Next