Torres is an operating system, not a proprietary piece of plastic. We are building the sovereign software layer that bridges the massive computational power of your smartphone with the sensors and displays of fully integrated smart glasses.
The glasses do not run the models. Asking lightweight eyewear frames to execute continuous neural inference on top of high-throughput sensor capture generates extreme thermal throttling within minutes. Instead, the glasses serve strictly as an egocentric sensory array and optical HUD—capturing uncompressed visual and auditory telemetry and projecting visual primitives. The host smartphone acts as the heavy cognitive node in your pocket, absorbing 100% of the thermal load while executing the perception mesh and cognitive stack over low-latency local Wi-Fi.
True spatial computing requires an uncompromising seven-pillar hardware architecture: an integrated microphone array, an egocentric camera system, onboard speakers, a visual display, a Bluetooth Low Energy (BLE) radio, a high-throughput Wi-Fi radio, and explicit compatibility with third-party software development kits. Devices lacking any single component—such as audio-only frames or simple Bluetooth peripherals—are fundamentally incapable of running the Torres telemetry pipeline.
The Primary Testbed
TCL RayNeo X2 & X3 Pro
At present, Torres is actively developed and rigorously benchmarked on the RayNeo X2 hardware. Built on Qualcomm's Snapdragon XR2 silicon with open Android OS access, the RayNeo allows for direct ADB deployments. This enables Torres to bypass proprietary app store restrictions, streaming raw optical and acoustic telemetry directly over low-latency Wi-Fi to the host smartphone while rendering asynchronous spatial overlays onto its 6,000-nit binocular Micro-LED waveguides.
The Compatible Pipeline
While the RayNeo X2 serves as our initial development standard, our architectural goal is hardware neutrality across the entire open AR ecosystem. Any platform that provides developers with unmitigated access to raw camera frames, audio arrays, and display rendering without forcing cloud interception will be supported.
Rokid Glasses (2026 AI) & Inmo Air Series
Standout edge platforms engineered with dedicated Wi-Fi radios and open Android SDKs. Their architectures avoid BLE bottlenecks, enabling continuous optical frame capture and low-latency audio transmission directly to our local smartphone cognitive engine.
OpenSource-Ai-Glasses (OSAIG)
A completely open, bare-metal hardware and software initiative. Operating directly on embedded Linux with native RTSP streaming support, OSAIG provides an unconstrained pipeline for local video ingestion and sovereign computing.
Enterprise & Tethered Companion Systems
Platforms leveraging Android XR, Snapdragon Spaces, or direct high-bandwidth USB-C connections—including the XREAL Aura, Xiaomi Wireless AR, and Lenovo ThinkReality A3. While utilizing differing tether mechanisms, their sensory suites and wide fields of view provide robust spatial anchoring.
Snap SPECS (5th Gen)
A standalone platform powered by dual Snapdragon processors. While Snap OS and Lens Studio enforce a closed-garden ecosystem, its quad-camera tracking array and 51-degree LCoS waveguides represent significant spatial hardware.
| Platform | Display & Optics | Sensory Camera Suite | Audio Architecture | Radios & Streaming | Ecosystem Profile |
|---|---|---|---|---|---|
| TCL RayNeo X2 / X3 Pro Active Testbed | Binocular Micro-LED 30° FOV / 6,000 nits |
12MP RGB + Monochrome Spatial Tracker | Triple-mic array with voice isolation; dual directional speakers | Wi-Fi 6, Bluetooth 5.3 (Low-latency Direct) | Open Android via ADB; direct hardware frame routing |
| Rokid Glasses (2026 AI) | Monocular Micro-LED 30° FOV / 1,500 nits |
12MP Ultra-compact sensor | 4-microphone array with AI ENC; directional speakers | Wi-Fi 6, Bluetooth 5.3 | Snapdragon AR1 SDK; cloud-bypass streaming compatible |
| Inmo Air 2 / Air 3 | Binocular Micro-OLED 26° FOV / High Contrast |
8MP HD Camera | Integrated near-field mics; open-ear / bone conduction | Wi-Fi Sync, Bluetooth 5.4 | Open Inmo Android SDK; direct APK deployment |
| OpenSource-Ai-Glasses | Monocular 640x480 30° FOV (Modular) |
1080p RGB Sensor | Integrated electret mic; onboard micro-speaker | Wi-Fi 802.11b/g/n, Bluetooth 5.3 | Bare-metal Linux (RV1106B); native RTSP streaming |
| Xiaomi Wireless AR | Binocular Micro-OLED 58 PPD / 1,200 nits |
Integrated spatial tracking cameras | Onboard spatial mics and micro-speakers | Proprietary high-speed Wi-Fi (3ms latency), BLE | Snapdragon Spaces XR; electrochromic dynamic dimming |
| XREAL Aura (Project Aura) | Binocular Micro-OLED 70° FOV / 120Hz |
High-Res RGB + Dual 3D Environment Sensors | 4-mic array; spatial sound speakers | Wi-Fi 6E, Bluetooth 5.3 | Android XR / Jetpack SceneCore; tethered companion routing |
| Lenovo ThinkReality A3 | Dual Micro-OLED 45° FOV / 1080p per eye |
8MP RGB + Dual Fisheye Tracking Cameras | 3 near-field microphones; integrated stereo speakers | Host PC/Phone Tether (USB-C DisplayPort) | Snapdragon Spaces / OpenXR; multi-screen virtual workspace |
| Snap SPECS (5th Gen) | Dual LCoS Waveguides 51° FOV / 16M Colors |
4 Cameras (Dual RGB + Dual Infrared 6DoF) | Multi-mic array; spatial audio speakers | Wi-Fi 6, Bluetooth 5.3 | Snap OS 2.0; Lens Studio gated sensor access |
The Blacklist: Hostile Ecosystems
Torres is fundamentally designed as an anti-panopticon. We cannot guarantee your safety, or the privacy of the people around you, if the hardware manufacturer is actively hostile to data sovereignty. Any hardware OEM caught intercepting, secretly uploading, or telemetry-harvesting camera frames will be permanently blacklisted from our ecosystem.
The Meta Ray-Ban Ban
All Meta hardware is permanently blacklisted. The Meta ecosystem is designed explicitly to turn your localized reality into their centralized training data. This was fully realized during the I-XRAY scandal, when Harvard students demonstrated that Meta Ray-Bans could be weaponized with PimEyes face search to auto-dox innocent strangers in public.
The NameTag Scandal
Despite Meta spending weeks denying the existence of a facial recognition feature, ongoing lawsuits have alleged that Meta illegally scraped Facebook and Instagram photos to secretly build "NameTag" into their glasses without user consent. A platform that steals biometric data to engineer mass surveillance endpoints is fundamentally incompatible with the Torres mission.
The Telemetry Guarantee
We refuse to build on platforms that utilize restrictive bridge SDKs or thermal-throttled DeviceSession rules designed to force users through corporate cloud endpoints. If a device requires an API key that logs queries to a third party, it will not run Torres.