SWIM.AI Open Sources Its Edge Intelligence Dev Stack

SWIM.AI announced this week that it has open sourced the stateful data processing engine at the core of its namesake app development platform under an Apache 2.0 license. It is now freely available to developers.

Swim is an intelligence platform designed to make it simple to build and deploy stateful, distributed, real-time edge applications, without requiring a database, message broker or application server. It can run locally at the edge or in the cloud to analyze, visualize, collaborate and act on streaming data as it is created.

"Swim is vertically integrated, but we eliminated the stack," the company's founder and chief architect, Chris Sachs, told Pure AI. "It's not an assembly of open source pieces or middleware. We built it from first principles. It does its own persistence its own peer-to-peer multiplex messaging, it has its own application management built in. It has everything you need in a 2MB stack that runs everywhere, from tiny edge devices to massive server clusters."

The Swim platform comprises three components:

  • The Swim platform and runtime, which provides the foundation for building and running stateful, scalable, real-time applications without requiring a database, message broker or app server.
  • A UI framework, SDKs and client, which allows developers to build new streaming visualizations and embedding streaming UI components into existing applications in days.
  • A "streaming-first" API (the company says it's the world's first), which integrates data streams and analytical insights with existing UIs or external applications.

Those streaming APIs are powered by the company's own WARP streaming protocol, which was developed as a multiplexed streaming upgrade to HTTP that enables the creation of bidirectional streaming links among distributed application services. "With WARP, developers can create meshes of application services which continuously share streaming insights with each other at network real-time," the company said in a statement.

If you're working in artificial intelligence, machine learning and/or deep learning, this is a dev platform worth noticing for several reasons. The concept at the center of the platform is "edge intelligence," explained CTO Simon Crosby.

"Edge applications are driven by data," Crosby told Pure AI. "When data show up, you compute, you operate in the graph in the flow of the data, you are time-limited and you are naturally distributed. That's fundamentally different from the abstractions we've used before, which are highly centralized, memory-limited and data centric. There's so much data streaming from so many things, that this notion of centralized learning on stuff is never going to work. What we have to do is learn on the fly."

Swim provides developers with a way to define entities and relationships (schema); build stateful, distributed digital "twin" models of the real-world from streaming data that collaborate to analyze, learn, predict and respond on the fly; and continuously stream real-time insights to UIs and applications.

To be clear, a "digital twin" is a stateful distributed object created from data streams, on the fly. Digital twins are persistent, active objects that process their own data. The Swim developer model also includes "lanes," which are object members; "links," which are the relationships among web; and "plane," which is a collection of actor definitions.

Sachs put the SWIM approach into a machine learning context.

"One of the great things about these stateful digital twins is they can train their own models," he said. "An ongoing issue with AI is, even though it sounds like data automation heaven, the training it's actually a very human-labor intensive process. But when you can look at full-fidelity data sets, and particularly when you operate in real time on streaming data, you can sort of turn on causality. You can use the fact that you know the state of some part of the world is now, and you can make a guess about what's going to happen next, and then you can wait and see what actually happens next, and then measure the error between what you thought would happen and what actually happened. Then you can take your training sample, back propagate it through neural networks, and you can actually self-train models that learn environments."

Sachs offered a real-world use case: managing the traffic signals in the City of Palo Alto, Calif. "Every second they're making thousands of predictions about their future states," he said. "This shows why it's so important to be stateful on the edge. CPUs work on nanosecond timescales, and networks operate on milliseconds or tens of milliseconds. That's a millionfold time dilation. When we train these traffic neural networks, we can do an entire training and prediction cycle in 17 milliseconds, less than half the time it takes to get a network packet to the nearest router."

Within a few months, the company has completed more than a trillion training operations in Palo Alto -- which is the size the training set of all of Google's images, Sachs said -- all of which is self-training and self-labeling, with no human effort involved. The company has implemented the same neural network in industrial settings.

"The assumptions have fundamentally changed," Sachs added. "We've gone from very little bandwidth and a few desktop machines to billions of heterogeneous connected devices with ubiquitous networks. But while the assumptions have changed, the way we build software really hasn't. That's why we created Swim."

Swim is available in a free Community version and a fee-based Enterprise version. The Swim community portal includes all of the documentation and support resources developers need to start building streaming applications today.

About the Author

John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS.  He can be reached at jwaters@converge360.com.