Analytics cookies only (Google Analytics, PostHog). No ad networks, no cross-site tracking. See our privacy policy.
Go
Self-host goose
updated Nov 2026deployed by usprices checked · Jul 2026Install verified on Ubuntu 26.04 · Nov 2026 · how we test
We earn commissions when you shop through the links below. Full disclosure →
A general-purpose AI agent written in Rust that runs on your own machine, with a desktop app, a command-line interface and an API. It connects to more than 15 model providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock and others) and to extensions through the Model Context Protocol. It has no web interface of its own; `goose serve` exposes an ACP server over HTTP and WebSocket. The CLI is a tool you run in a terminal, and the container image the project publishes is that CLI with git and curl added, so it needs a model provider configured before it can do any work.
Pick goose when you want a coding and automation agent you run yourself in a terminal, written in Rust and not tied to one model vendor. It works with more than 15 providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock and others), reaches tools through the Model Context Protocol, and has a desktop app and an API alongside the CLI. The project is Apache-2.0, now under the Agentic AI Foundation at the Linux Foundation; the repository moved from `block/goose` to `aaif-goose/goose`, and the old address redirects. This entry is the command-line tool: it has no web interface to open (`goose serve` exposes an ACP server over HTTP and WebSocket). Our snippet pins the `ghcr.io/aaif-goose/goose:1.57.0` image the project publishes (built from the v1.57.0 tag, published 2026-10-28, amd64 and arm64; Debian with git and curl added) and proves it with `--version` and `info`, which need no model or API key. The project's other install route is a script that downloads a release binary. We did not find an upstream RAM minimum; the 256 MB floor is an estimate. Among the catalog's coding agents, OpenHands is a self-hosted control center for coding agents, OpenCode is another terminal agent that also has a desktop app and a web mode, and PR-Agent reviews pull requests; goose is the choice when you want a general-purpose agent in Rust under Linux Foundation governance that is happy in a terminal or its desktop app; OpenCode is the one that also serves its sessions in a browser.
What you need
Any VPS with at least 256 MB of RAM
A domain you control — most self-hosted setups need HTTPS in front of them
Paid link — we earn a commission if you shop through it.
Install
Run these commands on your server:
# goose — open-source AI agent CLI, no web UI — pinned to the 1.57.0 image the project publishes to GitHub Container Registry (the Dockerfile in the repository, built from the v1.57.0 tag)docker pull ghcr.io/aaif-goose/goose:1.57.0docker run --rm ghcr.io/aaif-goose/goose:1.57.0 --versiondocker run --rm ghcr.io/aaif-goose/goose:1.57.0 info# → no URL: goose is a command-line tool. Run it with docker run -it --rm -v goose_config:/home/goose/.config/goose ghcr.io/aaif-goose/goose:1.57.0 configure to pick a model provider and add your API key
goose's documentation flags these before you rely on it:
non-negotiableThe image does nothing until you pick a model. Run docker run -it --rm -v goose_config:/home/goose/.config/goose ghcr.io/aaif-goose/goose:1.57.0 configure to choose a provider and enter its API key; goose has no bundled model.
non-negotiableThe documented default for GOOSE_MODE is auto, which lets the agent run tools without asking you. Set it to approve or smart_approve if you want a prompt before commands run, and keep the container away from directories you do not want edited.
non-negotiableIn a container there is no desktop keyring, so the docs say goose falls back to a plain-text secrets.yaml under ~/.config/goose. Keep that volume private, or pass the provider's environment variable (for example OPENAI_API_KEY), which takes precedence.
non-negotiableThe container runs as the non-root user goose with its home at /home/goose and only the CLI inside, so it sees only what you mount. To work on a project, mount it, for example with -v "$PWD":/work -w /work, and make sure user 1000 can write to it.
non-negotiableThe published image is not covered in the install docs; the project's primary install route is the release binary script, which downloads whatever version GOOSE_VERSION names. The latest image tag follows the main branch, so pin a version tag as we do.