I tried a Claude Code alternative that's local, open source, and completely free - how it works

2 days ago 5

How to get started with Goose, a escaped open-source alternate to Claude Code

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways

Free AI tools Goose and Qwen3-coder whitethorn regenerate a pricey Claude Code plan.
Setup is straightforward but requires a almighty section machine.
Early tests amusement promise, though issues stay with accuracy and retries.

Jack Dorsey is the laminitis of Twitter (now X), Square (now Block), and Bluesky (still blue). Back successful July, helium posted a fairly cryptic connection connected X, saying "goose + qwen3-coder = wow".

Also: I've tested escaped vs. paid AI coding tools - here's which 1 I'd really use

Since then, involvement has grown successful some Goose and Qwen3-coder. Goose, developed by Dorsey's institution Block, is an open-source cause framework, akin to Claude Code. Qwen3-coder is simply a coding-centric ample connection exemplary akin to Sonnet-4.5. Both are free.

Together, suggests the internet, they tin harvester to make a afloat escaped rival to Claude Code. But tin they? Really? I decided to find out.

Also: I utilized Claude Code to vibe codification a Mac app successful 8 hours, but it was much enactment than magic

This is the archetypal of 3 articles that volition sermon the integration of Goose (the cause framework), Ollama (an LLM server), and Qwen3-coder (the LLM).

In this article, I'll amusement you however to get everything working. In the adjacent article, I'll springiness you a much in-depth knowing of the roles each of these 3 tools plays successful the AI cause coding process. And then, finally, I'll effort to usage these tools to physique a afloat powered iPad app arsenic an hold of the apps I've been gathering with Claude Code.

Okay, let's get started. I'm gathering this connected my Mac, but you tin instal each 3 tools connected your Windows oregon Linux machine, if that's however you roll.

Downloading the bundle

You'll request to commencement by downloading some Goose and Ollama. You'll aboriginal download the Qwen3-coder exemplary from wrong Ollama:

I primitively downloaded and installed Goose first. But I couldn't get it to speech to Ollama. Can you conjecture what I did wrong? Yep. I hadn't yet downloaded and acceptable up Ollama.

Installing Ollama and Qwen3-coder

My proposal is to instal Ollama first. As I mentioned, I'm utilizing MacOS, but you tin usage immoderate you prefer. You tin besides instal a command-line mentation of Ollama, but I similar the app version, truthful that's what we'll beryllium exploring:

Download Ollama. Then, double-click the installer. Once the exertion loads, you'll spot a chat-like interface. To the right, you'll spot the model. Mine defaulted to gpt-oss-20b.

Also: Gemini tin look done your emails and photos to 'help' you present - but should you fto it?

Click that, and a exemplary database volition popular up. I chose Qwen3-coder:30b, wherever 30b refers to the fig of exemplary parameters. This is simply a coding-optimized exemplary with astir 30 cardinal parameters:

Note that the exemplary won't download until it's forced to reply a prompt. I typed the connection "test," and the exemplary downloaded:

Note that this exemplary is 17GB, truthful marque definite you person capable retention space. This request highlights 1 of the large benefits of this full project. Your AI is local, moving connected your machine. You're not sending thing to the cloud.

Also: How to easy tally your favourite section AI models connected Linux with this useful app

Once you've installed Qwen3-coder, you request to marque the Ollama lawsuit disposable to different applications connected your computer. To instrumentality this step, prime Settings from the Ollama paper connected your paper bar:

Turn connected Expose Ollama to the network. I fto Ollama instal itself successful the .ollama directory. This attack hides the directory, truthful retrieve that you person a 17GB record buried successful there.

Finally, I acceptable my discourse magnitude to 32K. I person 128GB of RAM connected my machine, truthful if I commencement to tally retired of context, I'll boost it. But I wanted to spot however good this attack worked with a smaller discourse space.

Also, announcement that I did not motion successful to Ollama. You tin make an relationship and usage immoderate unreality services. But we're attempting to bash this wholly for escaped and wholly connected the section computer, truthful I'm avoiding signing successful whenever I can.

Also: Is your AI cause up to the task? 3 ways to find erstwhile to delegate

And that's it for Ollama and Qwen3-coder. You volition request to person Ollama launched and moving whenever you usage Goose, but you astir apt won't interact with it overmuch aft this.

Installing Goose

Next up, let's instal Goose. Go up and tally the installer. As with Ollama, determination are aggregate Goose implementations. I chose the MacOS Apple Silicon Desktop version:

Once you motorboat Goose for the archetypal time, you'll get this Welcome screen. You person respective configuration choices, but since we're going for an all-free setup, spell down to the Other Providers conception and click Go to Provider Settings:

Here, you'll spot a precise ample database of assorted cause tools and LLMs you tin run. Scroll down, find Ollama, and deed Configure:

Once you bash that step, you'll beryllium asked to Configure Ollama. This is wherever I got a spot confused because, silly me, I thought "Configure Ollama" meant I was actually configuring Ollama. Not truthful much. What you're doing (here, and for each the different providers) is configuring your connection, successful this lawsuit to Ollama:

You'll beryllium asked to take a model. Once again, take qwen3-coder:30b:

Once you've chosen some Ollama and qwen3-coder:30b, deed Select Model:

Congratulations. You've present installed and configured a section coding agent, moving connected your computer.

Taking Goose for a rotation

As with astir immoderate different chatbot, you'll privation to benignant a punctual into the punctual area. But first, it's not a atrocious thought to fto Goose cognize the directory you'll beryllium using. For my archetypal test, I acceptable Goose to enactment from a impermanent folder. You specify this astatine (1) by tapping the directory already shown:

Also enactment that the exemplary you're moving is indicated astatine (2). You tin acceptable Goose up to tally aggregate models, but we're conscionable moving with this 1 for now.

As a test, I utilized my standard trial challenge -- gathering a elemental WordPress plugin. In its archetypal run, Goose/Qwen3 failed. It generated a plugin, but it didn't work:

In my 2nd and 3rd tries, aft explaining what didn't enactment to Goose/Qwen3, it failed, and failed again.

Also: True agentic AI is years distant - here's wherefore and however we get there

By the 3rd try, it ran the randomization, but didn't wholly travel directions, which benignant of defeated the full intent of the archetypal plugin:

It took 5 rounds for Goose to get it right, and it was very, precise pleased with itself astir however close it expected itself to be:

First impressions

So what bash I deliberation astir this approach? I was disappointed it took Goose 5 tries to get my small trial to work. When I tested a clump of escaped chatbots with this assignment, each but Grok and a pre-Gemini 3 Gemini got my small trial close connected the archetypal try.

Also: How I trial an AI chatbot's coding quality - and you can, too

But a big quality betwixt chatbot coding and agentic coding is that agentic coding tools similar Claude Code and Goose enactment connected the existent root code. Therefore, repeated corrections bash amended the existent codebase.

When my workfellow Tiernan Ray tried Ollama connected his 16GB M1 Mac, helium recovered show was unbearable. But I'm moving this setup connected an M4 Max Mac Studio with 128GB of RAM. I adjacent had Chrome, Fusion, Final Cut, VS Code, Xcode, Wispr Flow, and Photoshop unfastened astatine the aforesaid time.

So far, I've lone tally a reasonably elemental programming test, but I recovered that wide show is rather good. I didn't spot a tangible quality successful turnaround from prompts betwixt the section lawsuit moving Goose connected my Mac Studio and local/cloud hybrid products similar Claude Code and OpenAI Codex that usage the AI companies' tremendous infrastructures.

Also: 4 caller roles volition pb the agentic AI gyration - here's what they require

But these are inactive archetypal impressions. I'll beryllium amended capable to archer you if I deliberation this escaped solution tin regenerate the spendy alternatives similar Claude Code's $100/mo Max program oregon OpenAI's $200/mo Pro program erstwhile I tally a large task done it. That investigation is inactive to come, truthful enactment tuned.

Have you tried moving a coding-focused LLM locally with tools similar Goose, Ollama, oregon Qwen? How did setup spell for you, and what hardware are you moving it on? If you've utilized unreality options similar Claude oregon OpenAI Codex, however does section show and output prime compare? Let america cognize successful the comments below.

You tin travel my day-to-day task updates connected societal media. Be definite to subscribe to my play update newsletter, and travel maine connected Twitter/X astatine @DavidGewirtz, connected Facebook astatine Facebook.com/DavidGewirtz, connected Instagram astatine Instagram.com/DavidGewirtz, connected Bluesky astatine @DavidGewirtz.com, and connected YouTube astatine YouTube.com/DavidGewirtzTV.

Read Entire Article