Open-source agent goose runs on Claude, ChatGPT or Gemini subscriptions, no API key needed

The open-source agent goose documents a way to run against an existing Claude, ChatGPT or Gemini subscription instead of a pay-per-token API key, removing a cost barrier that has made agent tooling awkward for anyone already paying monthly for those chat products.

According to a walkthrough published on dev.to, the capability rests on goose's support for ACP providers. ACP is the protocol goose uses to talk to a provider the user is already logged into, so the agent draws on the subscriber's existing plan rather than billing API tokens. The post is candid that it was assembled from the project's README rather than a hands-on test, so the steps should be treated as documented rather than independently verified.

What goose is

goose is a general-purpose agent that runs on the user's own machine. It ships as a native desktop app for macOS, Linux and Windows, as a CLI, and as an API, and it is written in Rust. The project supports more than 15 model providers, among them Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure and Bedrock, and connects to over 70 extensions through the Model Context Protocol. Code is not the only target: the README describes research, writing, automation and data analysis as in scope. The project is now run under the Linux Foundation's Agentic AI Foundation, with source code hosted in the aaif-goose/goose repository.

How the subscription route works

The dev.to walkthrough breaks setup into three steps.

First, install either the desktop app or the CLI via the project's official release script. The author's standard caution applies: read a piped script before running it on your machine.

Second, open the ACP providers guide in goose's documentation and connect the provider you already pay for. The exact flow differs per provider, and the author deliberately avoids guessing at command flags, pointing readers to the guide as the source of truth because this part of the project changes quickly.

Third, start small. Run a short, checkable task rather than a large job, then confirm two things: the task completes, and the provider's usage page shows consumption counted against your plan rather than a new API charge. If an API charge appears, the wrong provider entry was selected.

Caveats worth heeding

Subscription terms are set by each vendor, so the advice is to check that your plan permits this kind of use before relying on it. The walkthrough itself was written from the README, not a test run, so each step should be verified against the current documentation. And because extensions give an agent real access to your machine, the recommendation is to add them one at a time rather than enabling everything at once.

Why it matters

Per-token API billing has functioned as a quiet second tax on agent adoption. People and teams already paying for Claude, ChatGPT or Gemini each month have had to open a separate usage-based account to give an agent access to the same models, which pushes experimentation toward trivial tasks or deters it entirely. If subscription-backed access works as goose's README describes, an existing plan becomes the meter, and trying an agent stops being a budgeting decision.

The vendor flexibility matters too. Because goose supports many providers, users can move between Claude, Gemini and a local Ollama model from a single interface, which makes the subscription route a convenience rather than a lock-in. The open question is whether each provider's terms will accommodate this usage pattern over the long run, and that is the detail to watch before making it a dependency.