Artificial intelligence has largely been introduced to the public through the cloud. You open an application, enter a prompt, and that information is sent to infrastructure operated by another company. The model runs somewhere else, a response comes back, and the provider determines the pricing, availability, limitations, and capabilities surrounding that experience.

Open models offer a different path. They make it possible for artificial intelligence to exist beyond a single company's infrastructure, allowing developers, organizations, and individuals to run and build upon models using hardware they control.

As AI moves from something we occasionally interact with into technology embedded throughout our devices, workplaces, and daily lives, that distinction is becoming increasingly important.

Understanding open models

When people refer to an open model, they are often describing an open-weight model. Model weights are the learned numerical parameters that determine how a trained model behaves. Making those weights available allows others to download and run the model independently and, depending on its license, modify, fine-tune, optimize, or build applications around it.

Open weights are not necessarily the same as fully open-source AI. Some models provide access to their weights while keeping their training data, development methods, or other components private. Fully open-source AI generally requires broader access to the information and components necessary to study, modify, and build upon the underlying system.

Both approaches, however, represent a meaningful departure from AI delivered exclusively through proprietary services. When model weights are available, access to the underlying technology is no longer entirely dependent on maintaining a connection to the company that developed it.

That independence creates new possibilities. Models can be stored, operated on private infrastructure, optimized for specialized applications, or deployed directly on consumer devices. Developers can build without relying on a single inference provider, organizations can retain greater control over their systems, and individuals can use AI without necessarily sending their information to a remote server.

In many ways, this follows a longer tradition in computing. Open-source software helped establish the foundations of the modern technology industry, from operating systems and programming languages to databases and web infrastructure. By allowing developers to build upon shared technology, open ecosystems helped make innovation more accessible to organizations that otherwise could not afford to create everything themselves.

AI is now approaching a similar moment. Training advanced models can require enormous investments in data, infrastructure, computing power, and expertise. Making models available allows a much broader ecosystem to participate without having to reproduce that investment from the beginning.

Instead of every company needing to become an AI research laboratory, more companies can focus on what they build with AI.

Why openness matters

One of the most important benefits of open models is choice.

When an application depends entirely on a proprietary model provider, it also depends on that provider's pricing, usage limits, infrastructure, and product decisions. Models can be discontinued, APIs can change, and capabilities that were previously available can become restricted or more expensive. For businesses building around AI, these dependencies can create significant long-term uncertainty.

Open models provide an alternative. Organizations can select models based on their particular requirements, deploy them on infrastructure they control, optimize them for specific workloads, and maintain access to versions that their products depend on.

For individuals, the implications extend further through the emergence of local AI.

As open models become smaller and more efficient, increasingly capable systems can run directly on personal computers and smartphones. Conversations, documents, personal information, and other sensitive data can be processed locally without inherently requiring transmission to an external inference provider.

This changes the relationship between people and the technology they use. AI can remain available without an internet connection, users can have greater control over how their information is processed, and access does not have to depend on a subscription or a fee for every request.

Local AI still carries costs. Hardware, electricity, storage, and software development all require resources. But running a model on a device someone already owns creates a fundamentally different economic model from purchasing access to intelligence through a remote API.

Openness also encourages competition beyond model development itself. When capable models become widely available, companies can differentiate through inference efficiency, hardware optimization, memory, interfaces, tools, security, and the experiences they build around those models.

The result is an ecosystem where innovation is not limited to the organizations with the resources to train the largest models.

The debate over open AI

The future of open models is not without uncertainty.

As AI capabilities advance, policymakers, researchers, and technology companies are increasingly examining the benefits and risks of making powerful models widely available. Unlike an API-controlled system, publicly released model weights cannot easily be recalled. They can be modified, redistributed, and potentially adapted for purposes their original developers never intended.

These concerns deserve serious consideration. Greater access can support independent research, transparency, competition, and defensive security work, but it can also introduce risks involving cybersecurity, public safety, and misuse. The challenge is determining how to address those risks without unnecessarily restricting the broader ecosystem.

In recent years, that discussion has also become intertwined with international competition over AI development. Governments are considering how access to advanced models affects economic competitiveness, national security, and technological independence. At the same time, major technology companies and open-source organizations have argued that widely available model weights are important for maintaining innovation and reducing dependence on a small number of providers.

The debate is often presented as a choice between open and closed AI, or between innovation and safety. In reality, the questions are more complex. Different models have different capabilities and risks, and the appropriate safeguards for one system may not be necessary for another.

A productive approach requires distinguishing between demonstrated risks associated with particular capabilities and broad restrictions that could prevent developers, researchers, and individuals from accessing useful technology.

Protecting people and preserving an open AI ecosystem should not be treated as mutually exclusive goals.

Building a future with more choice

There will continue to be an important role for cloud-based artificial intelligence. The largest frontier models require substantial computing infrastructure and can offer capabilities that smaller local systems cannot currently match. For many applications, accessing those models remotely will remain the most practical option.

But the future of AI does not need to be exclusively cloud-based, just as it does not need to be exclusively local.

There should be room for large models operating in data centers and smaller models running directly on personal devices. Businesses should be able to use hosted APIs when convenience and scale are priorities, while maintaining the option to operate models independently when privacy, control, or customization matters more.

Individuals should have that same flexibility. Some tasks may benefit from the capabilities of remote systems, while others involve personal information that people would prefer to keep entirely on their own devices.

At Oscilla, that belief is central to what we're building.

We believe people should be able to choose their models, run AI on hardware they already own, and use intelligent software without automatically transferring their information to another company's infrastructure.

Our work focuses on making open models accessible through local-first experiences across Apple devices. By bringing models, memory, files, tools, and other capabilities together in a single environment, we want to make personal AI more useful without requiring people to surrender control over how it operates.

That vision does not depend on every model being open or every AI experience happening locally. It depends on preserving the ability to choose.

As models become more efficient, consumer hardware becomes more capable, and open research continues to advance, more intelligence will be able to move from remote data centers onto the devices people already own.

Keeping that possibility accessible matters because the future of AI will be shaped by more than which company develops the most capable model.

It will also be shaped by who gets to use it, who gets to build with it, and who remains in control.