# goose

Local open-source agent for coding, research, writing and automation. Desktop, CLI and API connect a choice of models with MCP tools.

Canonical profile: https://agenten-kompass.mmeierhoff.chatgpt.site/en/agents/goose/
Editor: [Michael Meierhoff](https://agenten-kompass.mmeierhoff.chatgpt.site/en/about/)
Sources reviewed: 2026-10-10
Basis: provider documentation. Examples are editorial suggestions, not hands-on product tests.

## Overview

goose is part of the Agentic AI Foundation. Desktop, CLI and API use the local agent; models and connected services are selected separately.

- Provider: Agentic AI Foundation
- Product family: goose
- Profile type: Local general-purpose open-source agent
- Environment: macOS, Linux and Windows; desktop, CLI and API
- Connections: Files, shell, MCP extensions, skills and ACP agents
- Setup: Install goose, connect a model provider and configure extensions and tool permissions
- Pricing model: USD 0 software licence; model and service usage depends on the selected access

## Operation & oversight

Operational details for this profile have not yet been researched.

## Provider

Originally developed by Block, the project operates under the Agentic AI Foundation at the Linux Foundation.

- Operator: Agentic AI Foundation
- Organization: Open-source foundation
- Website: https://goose-docs.ai/
- Reviewed: 2026-10-10

- Provider source: https://goose-docs.ai/

- Provider source: https://goose-docs.ai/blog/2026/04/07/goose-moves-to-aaif/

## Prices and included usage

Pricing checked: 2026-10-10

goose is available under Apache 2.0. The software includes no fixed model quota; the selected provider determines ongoing costs.

### Open Source

USD 0 software licence; ongoing

Desktop, CLI and API; requires your own model access and service accounts.

Price source: https://goose-docs.ai/

Model tokens, existing subscriptions or local compute are funded separately. No uniform monthly or per-run price is publicly specified.

Pricing details: https://goose-docs.ai/

## Key features

### Work locally across interfaces

Desktop, CLI and API run the agent on your computer.

Source: https://goose-docs.ai/

### Connect MCP tools

Extensions connect browsers, databases and other services through MCP.

Source: https://goose-docs.ai/

### Choose a model provider

API providers, local models and compatible existing subscriptions can supply model access.

Source: https://goose-docs.ai/docs/quickstart/

### Reuse recipes

YAML recipes capture instructions, extensions and parameters for repeatable workflows.

Source: https://goose-docs.ai/docs/guides/recipes/session-recipes/

### Load skills as context

Agent Skills add reusable instructions and resources for specialised tasks.

Source: https://goose-docs.ai/docs/guides/context-engineering/using-skills/

### Add tool-call review

Adversary Mode evaluates configured tool calls against your rules before execution.

Source: https://goose-docs.ai/docs/guides/security/adversary-mode/

## Requirements and limits

### Set up model access and service costs

Requirement: The local agent requires configured model access. API usage, subscriptions and connected services have their own costs and limits.

Source: https://goose-docs.ai/docs/quickstart/

### ACP session restrictions

Limitation: With ACP providers, goose session resume and goose session fork currently do not support resuming or forking.

Source: https://goose-docs.ai/docs/guides/acp-providers/

### Adversary Mode fails open

Limitation: If the reviewer fails, the tool call is allowed. The mode reviews shell by default; additional tools must be explicitly included.

Source: https://goose-docs.ai/docs/guides/security/adversary-mode/

## Example workflows

### Prepare local data analysis

- Your input: A data directory, a question and approved analysis tools.
- Possible result: An analysis report with generated files and traceable commands.
- What to check: Check selected files, calculations and the model and tool access used.

### Save a repeatable workflow as a recipe

- Your input: A proven workflow with inputs, extensions and output format.
- Possible result: A parameterised recipe for later local runs.
- What to check: Check parameters, permissions and results with a small sample dataset.

## Context

Configure the model and extensions deliberately. ACP resume and fork are restricted; Adversary Mode permits actions when its reviewer fails.

Official product documentation reviewed. Workflows and assessments are editorial; hands-on product tests are outside the scope of this directory.

## Sources

- https://goose-docs.ai/

- https://goose-docs.ai/docs/quickstart/

- https://goose-docs.ai/docs/guides/recipes/session-recipes/

- https://goose-docs.ai/docs/guides/context-engineering/using-skills/

- https://goose-docs.ai/docs/guides/security/adversary-mode/

- https://goose-docs.ai/docs/guides/acp-providers/

- https://goose-docs.ai/blog/2026/04/07/goose-moves-to-aaif/

[Sources and method](https://agenten-kompass.mmeierhoff.chatgpt.site/en/sources/)