PeakVision AI bets on local enterprise agents with FengKan AIOS

PeakVision AI is pitching a different way to bring artificial intelligence into businesses: run more of the work locally, then let agents carry out tasks across an enterprise environment. The Hangzhou startup’s approach combines locally deployed computing, its FengKan AIOS platform and services designed around the job a user wants completed, rather than a collection of separate apps. ¹

That is a practical response to three familiar enterprise concerns: sensitive data leaving the company, the cost of repeated cloud inference and the complexity of deploying AI in the first place. PeakVision AI co-founder and CTO Liangyu Nie told TechNode that the company is building around those constraints, with a mix of hardware, subscriptions and customised deployments. ¹

An AI stack built around tasks

PeakVision AI says FengKan AIOS is intended to let users describe a goal in a chat interface while agents handle the steps in a local computing environment. The company’s own architecture materials describe an enterprise stack spanning agent orchestration, knowledge retrieval, model dispatch, security and inference. ²

The design is not entirely isolated from the wider ecosystem. TechNode reports that FengKan AIOS supports third-party hardware, while PeakVision is selectively developing a Skills ecosystem. Proprietary customer workflows are expected to remain private, with more general-purpose Skills potentially shared by partners and forward-deployed engineers. ¹

The company also works with hardware partners. TechNode says AMD is supporting an FDE academy run by Shanghai Sixunited Intelligent Technology, while PeakVision is working with chipmakers including ChangXin Memory Technologies to improve performance and reduce memory costs. ¹

Targeting smaller businesses

PeakVision AI is positioning itself against enterprise platforms aimed primarily at large regulated organisations. Its stated target is small and medium-sized businesses that face similar data, cost and deployment problems but cannot afford large specialist teams. The company says its lower-cost hardware and repeatable industry solutions are meant to make local AI more practical for those customers. ¹

The reported use cases range from manufacturing and financial analysis to anti-corruption investigations and fraud tracing. PeakVision AI told TechNode that one lifeboat manufacturer cut a product-design cycle from a month to a week, although those are company-reported results rather than independently audited benchmarks. ¹

The company plans to open-source FengKan AIOS by the end of October and is exploring overseas distribution across several regions. It has also set an ambition of reaching 80 to 100 enterprises across 10 industries by 2027. Those targets are forward-looking plans, not guarantees of adoption. ¹

Our opinion

The interesting part of PeakVision AI’s pitch is not the familiar promise that agents will replace every app. It is the decision to treat deployment, hardware and data control as part of the product rather than obstacles left for customers to solve later. That could matter to organisations that cannot send sensitive work to a public cloud, but local AI brings its own costs: hardware has to be bought, maintained and kept useful as models change.

PeakVision AI therefore has a sharper proposition than a generic chatbot startup, but the difficult proof will be repeatable results outside showcase deployments. Opening FengKan AIOS could help others inspect and extend the platform. For now, the company has outlined a credible direction for private enterprise AI, not yet demonstrated that it can turn that direction into a durable global business.