Quickstart

Get started with Toolang by trying a shared agent first, then creating and running your own local agent.

Try a shared agent

The fastest way to try Toolang is to run a shared remote agent with uvx. This does not install Toolang permanently.

uvx is a shortcut for uv tool run.

uvx toolang run https://toolang.ai/dev.too

If you already have one supported API key configured, or a local Ollama server running, this should start the agent directly. See Configure models for more details.

When the agent starts, Toolang prints runtime information in the terminal. Look for the WebUI URL, open it in your browser, then click Chat to talk with the agent.

Install Toolang

Install Toolang when you want to create and customize local agents.

We recommend uv for installing Python command-line tools:

uv tool install toolang

If you prefer the standard Python tooling, pipx and pip also work. Use pipx for an isolated CLI-app installation, or pip inside a Python environment:

pipx install toolang
pip install toolang

If Python is not installed, we recommend installing it with uv.

After installation, check that Toolang is available:

toolang --help

Build your own agent

To create a local agent, say alice, run:

toolang new alice

Then start the agent:

toolang start alice

When the agent starts, Toolang prints runtime information in the terminal. Look for the WebUI URL, open it in your browser, then click Chat to talk with your agent.

A new agent works with the default setup, but you can extend it with caps such as skills, services, prompts, and more:

toolang alice skill add briceyan/review
toolang alice service add briceyan/github

Show the caps available to alice:

toolang alice caps

Caps can be added before or after the agent starts. Toolang detects state changes and uses the updated state in the next agent run.

To stop the agent, run either command:

toolang stop alice
toolang alice stop

The second form reads like “agent alice, stop,” which is convenient when working with the same agent repeatedly.

Configure models

Toolang works with many model providers, and more can be added through plugins. If you already have provider environment variables set, or a local model service running, Toolang can discover the available models automatically.

Before running Toolang, make sure the API key for your provider is available in your shell:

export OPENAI_API_KEY=...

Inspect configured providers and see how many models are accessible from the current environment:

toolang model providers

List available models with an optional filter:

toolang model list --filter "openai/*,ollama/*"

The same selector syntax can be passed to run or start to limit which models are available for that agent run:

toolang run alice --models "openai/gpt-5[openai]"

The first selected model is used by default. If more than one model is available, you can switch between them later in the WebUI.