Gray vs Goose: which should you use?

Last updated

Short answer

Choose Goose if you want Block's agent with a desktop app, MCP extensions and runnable YAML recipes. Choose Gray if you want the smaller footprint — one static Rust binary, bash-only tools — with cron and always-on operation built in.

GrayGoose
What it isMinimal, always-on AI harness: one Rust binary, open source (MIT)Block's open-source (Apache-2.0) agent: CLI plus a desktop app
InterfaceTerminal REPL and headless modeCLI plus a desktop GUI
Extends withPlugins and skills; MCP servers through the official pluginExtensions built on MCP, plus shareable YAML recipes for repeatable tasks
ModelsAny OpenAI-compatible provider with your own API key; local models through Ollama; Claude Pro/Max and ChatGPT plans through official pluginsMost providers; can run fully offline against Ollama, LM Studio and other local servers
Runs 24/7Yes — the gateway runs Gray as a user service, with cron and app connectionsInteractive sessions; recipes automate repeatable runs
Security modelRuns the model's shell commands with your user's privileges; no sandbox or approval prompt by default. The official permissions plugin adds approvalsConfigurable permissions on tool use; local models can run fully offline
PriceFree (MIT); you pay your model provider. Hosted Voyager plan $77/month, not yet openFree (Apache-2.0); pay your model provider

Is Gray smaller than Goose?

Yes in shape: Gray is one static binary with bash as its only default tool; Goose ships a CLI and a desktop app with an extension system on top. Both are open source and work with your own provider keys.

Can both run fully offline?

Yes. Goose explicitly supports offline use with local models through Ollama, LM Studio and similar servers; Gray connects to the same local endpoints and GRAY_OFFLINE=1 disables its unsolicited network calls.

When should you choose Goose instead?

Choose Goose if you want a desktop app, its extension catalog, or shareable recipes for team workflows.

Sources

Competitor details are taken from the sources above as of the date at the top of this page. Products change quickly — if something here is out of date, open an issue on GitHub.