APXInf: Product Facts & Availability

ApxInf is a Rust-based edge inference engine for embodied AI, shipping with an optimized PI-0.5 VLA model on Jetson Thor and Orin.

Product information

Information below comes from the product website or public documentation. Pricing and availability may change.

Category: Robot Development
Maturityi: Available
Deploymenti: Self-hosted
Buying / access: Buy or sign up directly
Source: Official websiteChecked Sep 20, 2026

Reality check

A demo shows what may be possible. These points help judge whether the product fits day-to-day use.

Works well when

Simulation, control, perception and repeatable experiments
Teams that can manage dependencies and hardware interfaces
Prototyping before committing to a physical deployment

What to check

Supported hardware, GPU requirements and version compatibility
Documentation, examples and community response
How simulation results transfer to the real robot

Not a magic solution

A guarantee that simulation equals a production cell
A replacement for safety validation on hardware
A low-maintenance toolchain without version ownership

About

ApxInf is an edge inference engine from the infinigence team, implemented in Rust with no other external dependencies and built around embodied AI. It ships with a highly optimized PI-0.5 VLA model for Jetson Thor and Orin devices, supporting BF16, FP8 and INT8 precisions, and exposes L1 model, L2 policy and L3 OpenPI-compatible websocket serving layers. Reported results include 92.8% success on LIBERO-10 with BF16 on Jetson AGX Thor and 72.45 ms P50 latency for two-view PI-0.5 inference. It is released under Apache 2.0.