Goose

Useful in specific situations

Developer

Block

Version Reviewed

1.5.0

Last Updated

June 2026

Platform

macOS, Linux

Overview

I started evaluating Goose in early 2025 after hearing about it through the developer community. It is an open-source AI agent developed by Block (formerly Square) that operates in a sandboxed environment. It can edit code, execute terminal commands, and browse the web. The focus on safety and developer control is evident in its design.

I found Goose useful for projects where I need granular control over what the agent can access. The safety model is well thought out.

Personal Note

The safety model gave me confidence to let an agent operate on sensitive projects.

What Works Well

• The safety model is strong. Granular permission controls for file and command access.
• Open source with corporate backing from Block. This gives confidence in long-term maintenance.
• The sandboxed execution environment prevents accidental system modifications.
• Documentation is thorough with practical examples.
• Supports multiple LLM backends including local models.
• Extensible via MCP (Model Context Protocol).

Where It Works Less Well

• Setup is more involved than simpler CLI agents.
• The sandbox adds noticeable latency to operations.
• Code generation quality depends heavily on the backend model.
• The community is smaller than more established tools.
• I experienced occasional stability issues with the sandbox environment.

Use Cases

Teams that prioritize safety and need granular control over agent access. I found Goose useful for projects where I need to be careful about what the agent can modify.

Engineering Maturity

Medium. Handles common development tasks competently. The sandbox constraints can limit effectiveness on complex multi-service projects. The safety model is well designed for production environments.

Product Maturity

Medium. Core features are solid. The safety model is well implemented. The CLI interface is functional but not as polished as some alternatives.

Developer Experience

Installation requires setting up the Goose runtime and sandbox. Configuration is managed through a configuration file. The permission model requires some upfront setup but provides peace of mind.

Workflow Integration

The sandbox model means Goose operates in an isolated environment. File sharing between sandbox and host is configurable. Suitable for projects where safety and isolation matter.

Performance

The sandbox adds noticeable latency compared to agents that operate directly on the filesystem. Response times are acceptable for most tasks.

Documentation

Well-organized coverage of setup, configuration, safety features, and usage patterns. Examples reflect real use cases.

Pricing

Free and open source. Costs are limited to compute resources and model API fees.

Platform Support

macOS and Linux. Requires Docker or similar container runtime for the sandbox.

Verdict

Goose is a well-designed open-source agent with a strong safety model. I found it useful in situations where sandboxed execution and granular permission control are important. The sandbox overhead limits its appeal for fast-paced development.

Changelog