What is Goose?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
Goose is an open-source AI agent that can connect to different model providers and tools. Open-source software makes its source code—the human-readable instructions used to build it—available under terms that permit use, study, modification and sharing. Goose originated at Block and is now part of the Agentic AI Foundation.
The model is the trained component interpreting information and proposing responses. Goose supplies the surrounding agent environment, including ways to interact with tools and carry work across steps. A tool is a software function such as searching files or running a command. This lets you change the model without necessarily changing the environment.
Possible setups include a hosted DeepSeek model accessed through OpenRouter or a Gemma model running locally. Hosted means that the model runs on a provider’s computers. Local means the calculations run on the available machine. These examples demonstrate that the agent environment and the model can be selected separately.
An illustrative task is organizing a collection of notes. The same environment might read the files, ask a model to suggest groupings and present proposed changes. Switching models could alter the suggestions without necessarily changing the file tools or review interface. Changing the tools could alter what actions are possible even if the model stays the same.
A free harness does not make every connected service free. Model providers may charge for processing requests, and local models require suitable hardware. Permissions and information handling also depend on the complete configuration. A local application can still contact remote services.
The official project supports several ways of interacting with the agent, including a desktop application and command-line operation. A command-line interface is text-based control of software. These are interface options around the agent rather than different forms of intelligence.
Goose’s value in the glossary is as a concrete example of a reusable agent environment. Its openness can support inspection and adaptation, but does not make every generated result correct. The task, selected model, connected tools and verification process still determine whether the work is useful.
Source: Goose — Official Project Documentation.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy