goose

goose is for people who want everything local and nothing rented. Extensions are MCP servers, models come from 15-plus providers including Ollama, and it is open source with no vendor bill, so the daily experience is a desktop app or CLI.

The trait that decides it is the weekend: you will spend one tuning providers, extensions and recipes, and after that it is yours in a way the hosted tools never are. Block built it and gave it to the Linux Foundation. Pick it if you enjoy configuring a local agent. Pick Claude Code if you would rather use someone else's defaults.

goose's Docker sandbox is opt-in: pass --container and extensions execute inside an existing container, forget it and a bad command lands on a real filesystem with the user's permissions. A Linux Foundation home means no vendor to abandon it and no vendor obligated to fix it on a schedule, so a bug you hit waits for a volunteer.

The consequence: turn sandbox mode on before the first run and expect to read the issue tracker yourself. Sandbox by default would change the reliability score. What it does right: the source is open and extensions are plain MCP servers, so the fix you need is one you can write.

goose is simple in a good way. 1. Every extension is an MCP server declared in YAML, so the tool surface is enumerable from a config file rather than discovered at runtime. 2. Recipes make a task's prompt and extensions a versioned artifact, the closest thing here to a reproducible experiment, since two runs of one recipe differ only by model.

No benchmark is published, so capability is inferred from the design. The consequence is that goose is the easiest agent on this board to evaluate rigorously, and nobody has. The observation: it already outlived its vendor, Block, by moving to a foundation.

goose is the rare case with a boring money answer. Block released it in January 2025 and moved it to the Linux Foundation, so there is no cap table, no runway, and nothing to acquire. The moat is governance: a foundation cannot be bought, cannot pivot to a paid tier, and cannot lose a funding round.

The flip side is that nobody's quarterly number depends on shipping it, so cadence follows contributor interest rather than a roadmap with a revenue target. For a buyer that means stability without urgency. Position: long the codebase, with no counterparty, and budget engineering time instead of a subscription.

The demo is a desktop app running a recipe. There is no vendor, only the Linux Foundation, so there is no contract, no SSO, no audit log and no support line, and the security questionnaire has nobody to answer it. Sixty engineers means sixty keys, and the agent runs on laptops with an optional sandbox mode.

CI fit exists through the CLI, and onboarding is a brew install plus a provider key. Approved with conditions: central key issuance through our own model contract, sandbox mode required by policy, and one engineer named as the internal maintainer, because there is no external one.

goose is mine: Apache-2.0 Rust I can read and build. config.yaml declares extensions as MCP servers with cmd, args and envs, so my toolset is a file in git and a new machine is a checkout. Ollama is a provider, so it runs on my box with no vendor in the loop and nothing leaving the network.

Block gave it to the Linux Foundation; the foundation is the fork, already made, with governance attached. If it breaks I read the source and send a patch, which is the whole point. No grudging here. This is the shape every agent on this board should have.