Cisco announced today at Partner Summit that it will make its proprietary models available through Model Foundry. In other words, domain-specific AI models that it develops in-house are now centrally available. That may not seem like such a big step, but it certainly is. We explain why in this article.
Ask any decision-maker at an organization what they think of AI, and the answer will typically be a variation on the following: “We need to do something with it and recognize its value, but given the unpredictable nature and performance of AI models, we don’t dare to fully commit to it.”
Model Foundry, in combination with other components, is Cisco’s response to comments like these. This new component within Cisco Cloud Control is designed to give organizations, through partners, the confidence that the models will perform well. Even as they gain more autonomy, the models will reportedly perform reliably. Moreover, this will happen in a trusted and secure environment.
What is Cisco Model Foundry?
To put it simply, Model Foundry is a platform service that Cisco will offer within Cloud Control. Initially, it’s a service for partners. However, Cisco is also expanding Model Foundry’s availability to platforms such as AWS and Microsoft. This gives developers and enterprises in general access to what Model Foundry has to offer. And that’s quite a lot.
You can think of Model Foundry as the place where all the AI models that Cisco develops in-house become available. These are smaller models (SLMs). They have been trained on a very specific dataset and are therefore well-suited for very specific tasks.
What’s included in Cisco Model Foundry?
At launch, Model Foundry includes four models:
The Deep Network model (now at version 1.4) focuses on NetOps, such as troubleshooting. The Decision Model is a decision layer for AI agents that manage the network. Through this model, agents know, among other things, which tools to use sequentially. Furthermore, they also know whether the evidence and diagnosis of a network problem match.
The Time Series Model is specifically designed for data that machines generate continuously, for example, every second. This is the type of data that generic AI models struggle with. Finally, there is the Antares-3B model. It is the most powerful, yet still relatively small, model in the Antares series, focused on identifying vulnerabilities in software.
The goal is to turn Model Foundry into a catalog featuring a wide variety of models from which partners can choose.
Why is Model Foundry important?
Model Foundry is first and foremost a way for Cisco to offer its first-party models to partners and, later, to other parties as well. That wasn’t always possible before. For example, there were models that were more or less hidden within Cisco’s own tooling. They couldn’t be offered separately at all. In this regard, Model Foundry is primarily a way to get more out of the investments made in Cisco’s own models. As a result, they now reach many more customers.
In addition to deriving more value from its own models and, ideally, generating more revenue, there is, however, another, much more important reason why Model Foundry is a good idea. That is because such a service fills a significant gap in the offerings of, in this case, partners. They can provide their customers with excellent support by offering the models within Model Foundry. More broadly, support is also available within Cisco Cloud Control.
Partners can offer the models “as-is,” but they can also fine-tune them slightly for each client to make them an even better fit. They can also build an inferencing service around them. According to Cisco, all they need is an API key to use Model Foundry.
Performance and trust come together
Ultimately, Model Foundry is a key part of a larger whole, which should lead to better performance as well as greater trust in AI. For many organizations, this begins with the quality of the services provided by their service providers (i.e., the Cisco partners). If we assume, for the sake of simplicity, that partners know exactly what their customers need, then that knowledge, combined with the focus of the models in Model Foundry, has the potential to be an excellent combination.
Many Cisco partners will also continue to gain that knowledge. In fact, that is more or less what the new Cisco 360 Partner Program calls for. It is much more focused on vertical expertise, in security, networking, or another part of Cisco’s portfolio.
In this way, Model Foundry brings several elements together. First, there’s the power of the small models, which are designed to deliver strong results when AI is deployed for specific purposes. Additionally, Model Foundry is offered through Cisco Cloud Control. That platform acts asan AI harness and ensures that partners can deploy the models with confidence.
Finally, partners need to help their customers as effectively as possible. They should now be able to do so without worrying about the peripheral issues. At least, that’s the story on paper. It’s a compelling story, with the potential for a major impact on the rollout of AI by partners to their customers. Partners will, however, need to adapt and prepare to deliver this service to their customers. We’ll find out how that goes in due time. Cisco Model Foundry is launching today in Controlled Availability for partners.