Skip to content
AppDuo
ReviewsComparisonsBest ofCollections
Source-backed analysis13

Review basis

Research-based review: the AppDuo Editorial Team compiled it from official pages, documentation, app stores, press and user feedback, without hands-on testing. Every claim links to a source with the date it was checked.

13 reference sources

Information updated Nov 11, 2026

How we review
goose logo

goose Review

Open-source AI agent that runs on your machine and works with many model providers

Editorial Team7.3/10Write the first review

By Official AppDuo editorial team · Fact-checked

  • macOS
  • Windows
  • Linux
  • API
Open source · Free
Visit goose
Compare side by side
On this page

Decide at a glance

Our verdict

goose suits developers who want a free agent where they pick the model and control the tools connected via MCP. The trade-off is that you manage API keys and permissions yourself and accept the prompt-injection risk that Block's own security team demonstrated.

Best for: Developers who want to switch between many models, cloud or local

Skip it if: Buyers who need a packaged product with official customer support and a service contract

Pros

  • Free and open source under Apache 2.0; costs arise only at the model provider
  • Works with 15+ model providers and 70+ MCP extensions, including local models
  • Desktop app (macOS, Linux, Windows), CLI and API; the desktop source ships 16 interface languages (English, Vietnamese, Japanese, Korean…)

Cons

  • Prompt-injection risk: Block reported its red team ran an infostealer via a poisoned recipe; the docs also say detection is a safeguard, not a guarantee
  • Approved commands run with your full permissions, so you must vet extensions and tool permissions yourself
All pros and cons · Who is it for?

More pros

  • Usage metrics are opt-in and exclude conversations and code
  • Governed under the Linux Foundation's Agentic AI Foundation and actively released (v1.59.0 on 2026-11-10)

More cons

  • No paid plan with official support; help relies on docs and community (400+ open GitHub issues at check time)

Best for

  • Teams that need an open-source, vendor-neutral agent under Apache 2.0
  • People who want to automate non-coding work through MCP extensions

Skip it if

  • People who do not want to set up a model provider and tool permissions themselves

Experience & key features

goose is an open-source AI agent that runs on your machine. According to its README it is a general-purpose agent: used for coding but also research, writing, automation and data analysis. It ships as a desktop app for macOS, Linux and Windows, a CLI and an API. Block released goose in early 2025 and contributed it to the Linux Foundation's Agentic AI Foundation when the foundation was announced on 2025-12-09.

How it works and key features

goose connects to 15+ model providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure, Bedrock and more) with API keys, or reuses existing Claude, ChatGPT or Gemini subscriptions through the ACP protocol. Tools extend through 70+ MCP extensions. Notable features include recipes (portable YAML workflows), subagents for parallel tasks, MCP Apps that render interactive UIs in the app, and security settings such as tool permissions, prompt-injection detection and sandbox mode.

Pricing and value

goose itself is free, but the docs state that not every model provider has a free tier. Groq and Google Gemini offer free access, local models run through Ollama or LM Studio, and OpenRouter is pay-per-use. ChatGPT Codex needs a ChatGPT Plus/Pro subscription and Claude ACP needs a Claude Code subscription. The docs also note you may need to upgrade your model for better performance.

More analysis

Security and risk

In January 2026 Block published Operation Pale Fire, an internal red-team exercise aimed at goose. According to Block and The Register, the attackers hid instructions in calendar invites and recipes (including invisible Unicode characters); in the third campaign an employee ran a poisoned recipe and an infostealer ran on their laptop. goose has since added showing recipe instructions on load, stripping invisible characters and an initial prompt-injection detection. The goose docs also state that these checks are a safeguard, not a guarantee.

Who it is for

goose fits people who are comfortable with a command line, API keys and permission settings and want to pick their own model rather than be tied to one vendor. Buyers who want a packaged tool with official support should compare commercial coding agents or other open-source options such as OpenCode, Cline and OpenHands.

Screenshots

Banner image from the official goose homepage
Banner image from the official goose homepage
Subagents illustration from the official goose docs
Subagents illustration from the official goose docs
MCP Apps illustration from the official goose docs
MCP Apps illustration from the official goose docs

Plans & pricing

Prices last verified Nov 11, 2026

goose (open source)

Personal

Free

Per the official docs, goose is free and open source (Apache 2.0). Costs sit with the model provider: some offer free tiers (Groq, Google Gemini), local models via Ollama or LM Studio carry no service fee, and OpenRouter is pay-per-use. Some connections need a separate subscription, such as ChatGPT Plus/Pro or Claude Code.

Free plan limits
The pages opened show no goose-imposed limits; limits depend on the model provider you choose. The docs note you may need to upgrade your model for better performance.
Free trial
No free trial
Cancellation & refunds

Not applicable: goose has no paid plan. Refunds for model providers are set by each provider.

Privacy

Per the official docs, goose collects only anonymous usage metrics and only if you opt in: OS, version, provider and model, extension usage counts and session metrics, but not conversations, code, tool arguments or personal data. Conversations and prompts may still go to the LLM provider you choose, under that provider's policies. The optional ML prompt-injection detection sends tool-call content and recent messages to the configured endpoint.

Official pricing page

goose alternatives

Claude Code logo7.4/10

Claude Code

Anthropic's coding agent across terminal, IDE, web and CI

Paid
Cline logo7.3/10

Cline

Open-source coding agent that runs on the AI models you choose

Open source
All alternatives

Controversies & incidents

Issues documented by the press, regulators, courts or the vendor itself. Every item is sourced; unproven claims are labelled as allegations. Last researched: Nov 11, 2026

  • Medium impactConfirmedSecurity ·

    Block's red team ran an infostealer through a poisoned goose recipe

    According to Block, its internal Operation Pale Fire exercise used a prompt injection hidden in a recipe to run an infostealer on an employee's machine. Block says it added warnings and hidden-character filtering.

    Read the full account

    Per Block's engineering post (2026-01-15), its offensive security team hid prompt injections in Google Calendar invites using zero-width Unicode characters and used goose recipes carrying hidden instructions. The first two campaigns did not get the infostealer to run (the second failed because of a typo in the injection). The third posed as a bug report about right-to-left text; an employee ran the poisoned recipe while debugging and the infostealer ran. Block's detection and response team identified and contained the simulated threat. The Register (2026-01-12) quotes CISO James Nettesheim: 'it's possible to hide what that recipe is actually doing' and 'we have to assume that prompt injection is possible'. This was an internal exercise, and the sources opened report no harm to outside users. goose gained recipe-instruction display, invisible-character filtering and prompt-injection detection; per The Register, an approach where one LLM checks another is still being tested internally and is not yet in open-source goose.

    Vendor response

    Block published the findings, described the fixes and stated that goose along with other LLM coding tools 'can not effectively isolate untrusted context'.

    Sources:Block Engineering: how we red teamed our own AI agentThe Register (2026-01-12)

Frequently asked questions

Is goose free?

Yes. The official docs call goose a free and open-source AI agent. You may still pay a model provider unless you use a free tier or a local model.

Who develops goose?

Block created goose and contributed it to the Linux Foundation's Agentic AI Foundation, announced on 2025-12-09. The repo now lives under the aaif-goose organization on GitHub.

Can goose use local models?

Yes. The docs list Ollama, LM Studio, Lemonade and several other local options. You must download a model first, and models without tool calling can only chat.

Does goose send my data anywhere?

Anonymous usage metrics are sent only if you opt in and exclude conversations and code. Your working content still reaches the LLM provider you choose, under its policies.

Does goose support Vietnamese?

Yes, per the source code: the desktop app ships a Vietnamese catalog alongside 15 other languages, such as Japanese, Korean, German, French and Spanish.

Review evidence & sources13

Scores

Ease of use
6.5
Why this score

Desktop app and a multi-language interface, but you must pick a model provider, create API keys and configure permissions; for the native Windows CLI the install page recommends Git Bash.

Features
8.0
Why this score

Recipes, subagents, MCP Apps, 70+ MCP extensions, ACP for Zed/JetBrains/VS Code and security layers such as tool permissions, prompt-injection detection and adversary mode.

Value for money
8.5
Why this score

Free, open-source software; free-tier or local models work. Actual cost depends on the model provider, and the docs note you may need to upgrade your model for better performance.

Performance
7.0
Why this score

The README says it is built in Rust for performance and portability, but output quality depends on the model you choose (the docs say goose works best with Claude 4 models) and no independent benchmarks appear in the sources opened.

Support
6.5
Why this score

Docs, a Discord community and GitHub; no official support channel with service commitments was found. The repo has 400+ open issues.

Privacy & security
7.5
Why this score

Runs locally and usage metrics are opt-in and exclude conversations. Deductions: data still goes to the LLM provider, and Block's Operation Pale Fire red-team exercise showed prompt injection can lead to malware execution if an untrusted recipe is run.

Sources

Specs & compatibility
Developer
Agentic AI Foundation (Linux Foundation), originally Block
Headquarters
United States
Launched
2025
Category
DevelopmentAI

Capabilities

  • Open source
  • AI features
  • Public API
  • Integrations
  • Works offline

Platforms

  • macOS
  • Windows
  • Linux
  • API

Interface languages

  • English
  • Vietnamese
  • Japanese
  • Chinese
  • Spanish
  • French
  • Korean
  • German
  • Portuguese
  • Russian
  • Indonesian

Integrations

  • MCP extensions (70+)
  • Agent Client Protocol (Zed, JetBrains, VS Code)
  • Anthropic
  • OpenAI
  • Google Gemini
  • Ollama
  • OpenRouter
  • Azure
  • Amazon Bedrock

Support

  • Help center
  • Community forum

Change history

  1. Feature.Latest release at check time: v1.59.0.
  2. Feature.Block published Operation Pale Fire and its fixes: recipe instructions shown, zero-width characters stripped, an initial prompt-injection detection.
  3. Company.The Linux Foundation formed the Agentic AI Foundation; goose (from Block), MCP (Anthropic) and AGENTS.md (OpenAI) are the founding projects.
Full change history (5)
  1. Launch.Block released goose as open source in early 2025, per the Linux Foundation announcement.
  2. Update.The goose repository was created on GitHub.
About the author

Reviews & comments

Written by readers with a verified email address.

No reviews yet. Have you used goose? Share your real experience.

Reader reviews are counted separately and do not affect the Editorial Team’s score.

No comments yet. Be the first to share your experience.

Comments are readers’ opinions, not content from the AppDuo Editorial Team.