In November 2023 I published a Laravel package on GitHub. It did one thing: connect Laravel applications with Ollama, the tool that lets you run AI models locally. No external APIs, no cloud costs, no data leaving your machine.
Today, that package has over 301,000 downloads, 612 stars on GitHub, 144 contributors, and 19 released versions. But the most interesting part is not the numbers — it's what happened because of them.
The Challenge
When I started, the landscape for AI in Laravel was nearly empty. If you wanted to integrate language models, your only real option was calling external APIs. Ollama had just launched as an open-source project that let you download and run models locally. But the API was raw HTTP, the documentation was geared toward Python users, and there was nothing for the PHP/Laravel ecosystem.
I saw the gap. Thousands of Laravel developers were interested in AI but had no tool that spoke their language: Facades, service providers, Composer packages.
Why It's Hard
The first version took a weekend to build. But the real challenge started the day after publishing it.
The Ollama API evolved fast. New endpoints for chat, vision, embeddings, function calling, thinking/reasoning models. Each Ollama update meant the package needed to catch up. At the same time, Laravel itself released version 11 and then 12, each with breaking changes.
Then there's the community dimension. 144 contributors means 144 different use cases, coding styles, and expectations. The hardest part? Saying no. Keeping a package focused when there are hundreds of valid feature requests pulling it in different directions.
Three decisions that defined the package's success
Make it feel native to Laravel. Not a generic PHP HTTP client, but a package that uses Facades, publishes config files with Artisan, and integrates via service providers.
Partner with Spatie. Using their spatie/laravel-package-tools as the package foundation gave immediate credibility and a battle-tested scaffolding.
Documentation-first development. Every feature shipped with a code example in the README. The example was written before the implementation. This meant the API was always designed for the developer experience.
Results
301K+ downloads, 612 GitHub stars, 144 contributors, 19 versions.
The package grew organically. No paid promotion, no Product Hunt launch. The growth came from developers finding it, trying it, and recommending it.
But the most valuable outcome is the conversations that start with "I saw your Ollama package" and end with "We need help building something with AI for our company." The package is proof of competence that no portfolio page can match.
What We'd Do Differently
I would invest more in automated testing from day one. I would create a version compatibility matrix earlier. I would build a dedicated documentation site sooner. And I would track the pipeline connection between package downloads and consulting inquiries from the start.
Takeaway
If you're a developer building a consulting or freelance business, publishing open-source code is the strongest signal you can send. It's code, not claims. It's adoption metrics, not testimonials. You build trust one commit at a time.
