01
No install day
No Homebrew, no dpkg, no missing libnotify. Sign in, pick a provider, start a session.
Goose, without the setup.
Goose Cloud is a managed home for Block’s open‑source AI agent — the desktop, CLI, and API you already know, running on our machines instead of yours. Bring your own OpenAI, Anthropic, or Gemini key. We handle the rest.
14‑day pilot · cancel any time · your keys never route through us
- Ships with Goose v1.47.0
- 40+ LLM providers
- 70+ MCP extensions
- Apache 2.0 under the hood
Article OneThe Case for Hosting
You wanted an agent, not another installer.
Goose is fantastic. It runs locally, works with any LLM, and reaches into your machine through MCP. But local means your laptop — with your Homebrew, your GPU driver, your Node version, your VPN.
Goose Cloud runs the same Rust binary in a per‑user sandbox on our infrastructure, exposes it through a browser (and, soon, a slim desktop shell), and keeps your sessions warm between visits. You still bring your own model keys — nothing routes through us that you don’t configure.
It’s the same agent. Different machine.
The best local tools are the ones that work everywhere you are. Goose Cloud is the shortest path between Goose and a working session.
02
Warm sessions
Your workspace, extensions, and recipes are already loaded. Close the tab, come back tomorrow, keep going.
03
Your keys, your models
OpenAI, Anthropic, Gemini, DeepSeek, xAI, Groq, or your Ollama endpoint — you configure it, we don’t proxy it.
Reading files. Running code. Fixing pipelines. The agent, at three different jobs, on the same afternoon.
Article TwoThe Workflow
Four steps from “I heard about Goose” to “ship it.”
No install day, no dotfiles, no dependency archaeology. The distance between curiosity and a working session is one short form.
-
I.
Get a workspace
Sign up. We provision a per‑user Goose v1.47.0 instance, a private volume, and a URL. About 20 seconds.
-
II.
Bring your model key
Paste an OpenAI, Anthropic, Gemini, or Ollama endpoint into the same provider picker you’d see locally — the one with 40+ providers.
-
III.
Enable the extensions you want
GitHub, Slack, Postgres, filesystem, browser, shell, memory, developer, orchestrator — the whole MCP catalog, one toggle each.
-
IV.
Start a session
Chat, run code, review PRs, refactor a repo, script an internal tool. Your sessions persist. Your logs stay yours.
Article ThreeInside the Agent
The screens you actually use, on our machine.
Every screenshot below is the real Goose desktop v1.47.0, captured out of the running app. Same UI, same providers, same local‑model flow — hosted for you.
The welcome screen is the same on the first visit to your workspace. Pick a hosted local model, or connect a provider you already have credentials for.
Anthropic, Azure AI Foundry, ChatGPT Codex, Databricks, Gemini, Hugging Face, Ollama, OpenAI, OpenRouter, xAI, DeepSeek, Groq, Bedrock, Vertex AI, and more.
The provider form is unchanged. Paste your OpenAI, Anthropic, or Gemini key straight in. We never see it in plaintext; we never route your requests.
Prefer to stay off cloud LLMs? Pick a hosted GGUF. We spin it up on a GPU node, quantized to fit your workspace, and Goose talks to it exactly like a local Ollama.
SidebarTwo Little Facts
It’s literally in the cloud.
Managed containers. Isolated per user. Snapshot‑backed volumes. No noisy neighbors, no shared filesystem.
Recipes travel with you.
Your Goose recipes, extensions, and sessions live in your workspace. Move to another device, sign in, keep cooking.
Article FourThe Price
One workspace. One price. No metering.
You bring your model keys, so what you pay us is just for the machine and the plumbing. No token math on our side.
Solo developer
$49.95/month
The full Goose desktop experience, hosted per user. Cancel any time.
Start your 14‑day pilot →14‑day pilot. You bring the model key.
- Dedicated Goose v1.47.0 instance
- Bring your own model key (any of 40+ providers)
- All 70+ MCP extensions available
- Persistent workspace volume + session history
- Access via the web app + CLI + API
- Optional hosted local GGUF models
- Automatic upgrades to the next Goose release
- Community support · office hours
Article FiveQuestions Asked So Far
The FAQ, kept short.
Is this the same Goose that Block open‑sourced?
Yes. Same binary, same version. We track the upstream release train from block/goose and roll workspaces forward automatically.
Do I need to bring my own model key?
Yes. Goose Cloud hosts the runtime — the LLM calls go directly from your workspace to whichever provider you configured (OpenAI, Anthropic, Gemini, etc.). Your key sits encrypted in your workspace’s secret store.
What about local models?
We can host quantized GGUFs on GPU nodes and expose them to your Goose instance like a local Ollama. Optional add‑on, off by default.
Do you see my prompts or code?
No. Model traffic flows directly from your workspace container to the provider you configured. We run the container; we don’t proxy or read the traffic. Standard infra logging only.
How is this different from Cursor / Claude Code / Codex?
Those are closed products with their own agent loops. Goose is open source (Apache 2.0), model‑agnostic, and MCP‑native. Goose Cloud is the shortest way to run it without owning the infra.
Can I still run Goose locally?
Absolutely. Everything you set up in the cloud (recipes, extensions, provider config) exports back to a local install.
Team plan?
In the works. Solo is the whole product today.
Take the agent. Leave the setup.
Start hosting — $49.95/mo →Workspace live in about 20 seconds. Cancel any time.