NimbleEdge proposes cure for on-device AI development pain
The startup lays the groundwork for an ambitious plan to support on-device AI agents that don't require cloud connectivity with the release of a new open source SDK.
An edge computing service provider has its sights set on orchestrating agentic AI for mobile devices -- no cloud connectivity required.
NimbleEdge, founded in 2021 and headquartered in San Francisco, took the first steps toward this goal with the July 10 release of DeliteAI, an SDK that connects front-end app developers to on-device AI and machine learning frameworks without requiring Python expertise. DeliteAI was accompanied by the launch of an AI agent marketplace for mobile apps and the NimbleEdge Assistant, an on-device chatbot.
The company began by building on-device personalization and recommendation engines based on machine learning for its customers. While NimbleEdge doesn't disclose the names of those customers, its founders said the business supports mobile apps on millions of devices for food delivery, e-commerce and fantasy sports companies.
The rise of generative AI starting in 2022 worsened a disconnect that NimbleEdge developers were already struggling with, according to co-founder and CEO Varun Khare.
"There's this front-end ecosystem where developers are working in languages like Kotlin, Swift, React, and then there's this AI layer, which is being orchestrated in Python, and both sides don't know how to work with each other," Khare said. "So there needed to be a platform that can interface on the application side while also natively integrating AI."
DeliteAI is meant to be that AI development platform, connecting on-device runtimes such as ExecuTorch and Onnx with hooks based in Kotlin and Swift for front-end developers. Developers can optionally connect it to NimbleEdge's SaaS service to change those underlying Python scripts without having to update the front-end application.
Neeraj Poddar, co-founder and CTO of on-device AI development company NimbleEdge (left), and Varun Khare, co-founder and CEO.
The goal: On-device agentic AI, no cloud needed
The company's plans go much further than Python scripts: If its co-founders have their way, a forthcoming Android and iOS SDK will be the basis for on-device multi-agent workflows that automatically analyze mobile applications to personalize and coordinate them, while keeping all data and processing local.
"The oldest device that we deployed in production was a nine-year-old smartphone with 2 GB of RAM, but you still have to support all the [SDK] capabilities there," said Neeraj Poddar, co-founder and CTO at NimbleEdge.
Eventually, however, devices will support more processing power, and language models will get smaller but contain more knowledge than is available now, Khare said.
If you can do [on-device AI development] in a well-defined way, then the developers get to choose [whether] to use an on-device AI or the cloud, and they can switch between the two seamlessly.
Varun KhareCo-founder and CEO, NimbleEdge
"Then, we feel that we will be able to use multimodal models that can do vision, voice in multiple languages and text all on the device -- so a lot of the things that you need AI for, you need not go to the cloud," he said. "That's the exciting part of on-device AI. And if you can do that in a well-defined way, then the developers get to choose [whether] to use an on-device AI or the cloud, and they can switch between the two seamlessly."
In one analyst's view, on-device processors are already ahead of the software that can use them, and that software is typically built specifically for certain combinations of processors and operating systems -- potentially opening opportunities for NimbleEdge.
"In the PC world, Intel, Nvidia, AMD and Qualcomm all use different software interfaces to access their AI hardware, which means each platform requires bespoke development work," said Gabe Knuth, an analyst at Enterprise Strategy Group, now part of Omdia. "[NimbleEdge could offer] both a way to provide that abstraction layer and a way to develop and deploy AI workloads and agents to those devices."
According to its website, NimbleEdge is backed by venture capital investors Neotribe Ventures and Sistema Asia Capital. It counts Srinivas Narayanan, vice president of engineering at OpenAI, among its advisers. However, as a small company with fewer than 50 employees, according to its LinkedIn profile, NimbleEdge faces a steep uphill battle to adoption, said Steven Dickens, CEO and principal analyst at HyperFrame Research.
"It's always a challenge with companies of that size, trying to get into a [market such as mobile development] where there are some pretty big behemoths -- can it get above the noise?" Dickens said. "Will it become a feature of a bigger stack for Google, Samsung, T-Mobile or Ericsson?"
Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.