The idea of open versus closed (or proprietary) models is familiar to anyone who has worked in software, and the definition is the same when it comes to AI. According to our partners at the Open Source Initiative (OSI), open source AI means anyone can look at how the model works, change it, use it, and share it freely without needing to ask for permission. OpenAI’s new open weight models don’t entirely meet this definition, but they are still in position to deliver the most impact for many businesses. 

A Morning Consult report from 2013 identified cost and privacy as the primary barriers to AI adoption for small businesses. The arrival of open models has the potential to address both concerns: they can be deployed offline, enabling greater data privacy, and they are free to use.  

Proprietary AI models often deliver stronger performance, but they require costly infrastructure and must run on an external provider’s servers, which requires a business to hand over its data. Open source models, by contrast, are improving quickly, and as they narrow the performance gap. Already, it is possible to run certain models locally, or even on a laptop or within a company’s own walls—options that can offer a powerful, self-managed alternative in the right circumstances.

Why should businesses care?

AI is becoming a lot more practical for Main Street businesses. What’s new isn’t just that open models can be a fixed-cost asset instead of a metered service. It’s that these self-hosted, affordable models are getting good—really good! Small businesses can now keep customer data in-house, avoid surprise price hikes, and spin up niche copilots like a “local operations assistant” or a “custom cake-order concierge” on their own terms, without needing to be AI experts or train models themselves.  

Open models also come with built-in transparency. Business owners can get a direct view of how the AI works, and the ability to store data in-house offers a surer way to meet compliance obligations. This is particularly important in an era where centralized data storage systems have been the targets of data breaches.

Having full control over the AI stack also means businesses can adapt and integrate models into the parts of their operations that matter most, such as loyalty programs, supply chain workflows, or customer support scripts, without worrying about vendor lock-in or shifting API terms. That kind of control turns AI into a strategic asset and helps businesses create more tailored, differentiated experiences.

Democratizing Access to AI Agents

Aside from this ongoing improvement, there is another cause for enthusiasm: open AI models aren’t just static prediction engines, they’re becoming active helpers. When paired with tools that allow them to take actions, these models turn into “AI agents” that can execute tasks automatically. Consider open source projects like Goose, an AI agent framework our team at Block released in January. 

Goose also runs on your own computer (not on Block’s servers) and can connect a language model to real-world business actions, from drafting emails to updating spreadsheets. In practical terms, this means a small business could have an AI that not only suggests answers, but actually logs into their inventory system, finds the relevant data, and helps complete a task, entirely on-site. 

Imagine automating your invoicing, email replies or appointment scheduling with an AI that operates like a diligent virtual assistant, and doing it without sending any data to the cloud or paying per-action fees. Open source models are the “brains” behind such solutions, and projects like Goose provide the “arms and legs” to carry out actions. The combination is powerful. Businesses and their customers could interact with AI that actually gets things done, such as finding products, placing orders, or handling bookings, rather than just chatting. They also wouldn’t have to pay extra for every transaction or worry about an external service outage in the middle of an operation.

None of this is hype or sci-fi, it’s the emerging reality of open source AI. To ensure this future reaches every corner deli and family clinic, the tech community and policymakers must continue investing in open models, accessible tools, and open standards. The U.S. government’s AI Action Plan explicitly highlights open source AI as vital for American innovation and even calls for convening stakeholders to drive open adoption among smaller businesses. That kind of support signals a growing consensus: if we want AI’s benefits to be broadly distributed, openness is key.