Install goose and connect a model
Code Snippets
curl -fsSL https://github.com/aaif-goose/goose/releases/download/v1.57.0/download_cli.sh \
| GOOSE_VERSION=v1.57.0 CONFIGURE=false bashactive_provider: openai
providers:
openai:
enabled: true
model: gpt-4.1-mini
configured: true
GOOSE_TELEMETRY_ENABLED: false
extensions:
developer:
type: builtin
name: developer
enabled: true
bundled: true
timeout: 300
todo:
type: platform
name: todo
enabled: true
bundled: truegoose info --checkInstall goose and connect a model. In this video you'll install a pinned goose, connect it to a model, and give it a first task. goose is an open source AI agent that started at Block, and now belongs to the Agentic AI Foundation. It runs on your machine. A model behind an OpenAI compatible endpoint decides what to do, and goose does it, with the tools that its extensions give it. The developer extension reads and writes files, and runs shell commands. In this lesson you will install goose at a pinned version, connect it to a model with a config file and an environment variable, and run a first task. Before you start, you need curl, bzip2 and jq, and a model endpoint that speaks the OpenAI API, with its key. The first command checks the tools. goose comes as a desktop app and as a command line tool. I use the command line, because recipes, scripts, and CI all run there. The agent is the same in both. I install version 1.57.0, released on the twenty-eighth of October 2026. The installer accepts a version, so I pin it. Agent tools change fast, and a CI job should not change under you. The script downloads the goose program for this platform and puts it in the local bin folder. The configure flag set to false skips the interactive setup, because I configure goose with a file. goose keeps three things on disk: a configuration file, a database of sessions, and logs. The info command shows where. The configuration file does not exist yet. To connect a model, goose needs a provider, a model name, and a key. The key never goes into a file. goose reads it from the environment variable OPENAI_API_KEY, which I loaded before recording. The file only names the provider and the model. This is the configuration. The first lines select the OpenAI provider, which works with any OpenAI-compatible endpoint, and the model name sonnet. I turn off usage statistics, so that goose does not stop to ask about them. Then come the extensions. An extension gives goose tools. Developer is the one that reads and writes files and runs shell commands. Todo gives goose a checklist for longer tasks. Both are built in, so nothing is downloaded. Now I check the connection. The check flag makes a real request to the provider, to prove that the key and the model work. Both extensions are enabled. Now a first task. The run command executes one prompt and exits. I ask a harmless question about this folder. goose uses a tool from the developer extension, then answers. The wording of the answer will differ from take to take. goose sends the prompt to the model together with the list of tools. The model asks for a tool, goose runs it, and the model writes the answer from the result. Look at the top of the output. goose prints the provider and the model it is using, and a session ID made of the date and a counter. That header is the quickest way to confirm that a run used the settings you expect. goose also prints every tool call it makes, so you can see what it did, not only what it said. The session is saved. The session list command shows it, with the folder it ran in. That is the setup. goose is installed at a pinned version, connected to a model through an environment variable, with two extensions. In this lesson you learned to pin goose to one version, to keep the key in the environment while the file names the provider and the model, and to read the run header and the tool calls to see what goose used, and what it did.