Gasgoo Munich-On July 13, StepFun unveiled three core products designed for edge devices: the agent-native operating system Step AOS, the personal agent StepFun Amoo, and the on-device model family Step Edge. That same day, strategic partner Chongqing Afari Technology announced plans to integrate this full-stack capability into the automotive sector, marking a joint push to deploy large-model native architectures inside vehicles.
This marks the first time StepFun has packaged its full-stack agent capabilities for the automotive sector since establishing its "AI+Terminal" strategy. For an industry hunting for the next paradigm in human-car interaction, this technical path deserves a close look.
Image source: Chongqing Afari Technology
Three-Layer Architecture: A Complete Loop from Chip to Interaction
Founded in 2023, StepFun has built a "1+N" model matrix under its Step series, covering both base and multimodal models. While its capabilities were previously delivered primarily via cloud APIs, these three new products map directly to the three layers of technical architecture required for edge devices:
Step Edge is an on-device model customized for the computing power, battery life, and specific needs of hardware like smartphones and cars. The design philosophy is straightforward: lightweight tasks are handled locally by the device, while complex scenarios trigger cloud collaboration. In a vehicle, this means high-frequency tasks—voice interaction, navigation queries, and vehicle status checks—get real-time local responses, slashing network dependency and latency. StepFun claims Step Edge ranks first among comparable on-device models across 29 core benchmarks, spanning text understanding, visual recognition, audio processing, speech recognition, GUI interaction, and image generation.
Dubbed the "world's first agent-native operating system," Step AOS is designed to provide a complete runtime environment for AI agents, centered on memory, decision execution, and security. It encapsulates hardware computing power, sensor data, and software functions into interfaces that agents can call directly. Users simply state their intent, and the agent handles the entire scheduling chain—shifting the interaction paradigm from "humans operating tools" to "human-machine symbiosis."
StepFun Amoo is the personal agent running atop Step AOS. With operating system-level permissions, it can orchestrate resources across applications and synchronize edge and cloud operations. Once integrated into vehicles, it will evolve into a dedicated travel agent, managing the full spectrum of driving scenarios—from trip planning and navigation to vehicle management.
The Strategic Logic Behind the Layout
Partner Chongqing Afari Technology brings full-stack mass-production capabilities to the table in China's smart driving and smart cockpit sectors, including infotainment adaptation, automotive-grade functional safety verification, and full vehicle integration. These are the exact engineering safeguards StepFun needs to move its tech stack from the lab to the assembly line.
Existing in-car agent solutions are mostly limited to infotainment apps or plugins. Restricted by application-layer permissions, they cannot access underlying capabilities like autonomous driving systems or body control units. StepFun's approach tackles this at the system level, while Chongqing Afari Technology handles the heavy lifting: optimizing on-device model inference on automotive chips, securing system-level functional safety certification, and managing multimodal data compliance.
"In the past, StepFun largely provided AI capabilities to smartphone and automakers on a B2B basis," Yin Qi noted at the launch. "But within the current device ecosystem, it is difficult to fully deliver the model and system experience." He added, "If we cannot pioneer such an innovative device ourselves, Step AOS cannot close the value loop." That logic holds true for the automotive sector as well—StepFun needs its partnership with Chongqing Afari Technology to validate the full loop inside vehicles first.
Viewed broadly, the StepFun-Qianli partnership represents a new collaboration model between AI firms and Tier 1 suppliers: the AI company provides the models and system-level capabilities, while the Tier 1 handles engineering and automotive adaptation, jointly defining the tech stack for the next generation of smart vehicles. Whether this model succeeds will heavily influence how in-car agent architectures evolve.
StepFun defines agents not merely as conversationalists, but as doers. On mobile, its STEPS platform has forged deep AI partnerships with ecosystem players like Ctrip, Alipay, Didi, Meituan, and WPS. Yin Qi emphasized that these integrations rely on API connections rather than GUI click simulations—allowing agents to reliably complete tasks across different scenarios.
The same logic applies to the car. When an agent plans a long road trip, it must understand travel preferences and vehicle range while coordinating navigation, hotel bookings, and charging station queries. StepFun views payment capability as the crucial link that turns an agent's suggestion into a completed task. On security, all operations run within a trusted execution environment with end-to-end encryption. Permissions are revoked immediately after use, and every action is visible, traceable, and reversible. High-risk actions require extra authorization, with the final decision remaining in the user's hands—a critical feature for automotive scenarios.
The core value of StepFun's release lies not in individual product specs, but in presenting a complete automotive technical stack spanning on-device models, operating systems, and agents—all while laying out an ecosystem of services and a security trust framework. For an auto industry searching for the next paradigm in human-car interaction, this is a roadmap worth watching.
