# SDTech > SDTech builds open-source, AI-driven hardware: small devices that see, decide and move ("Web4"). Founded in 2026 by Serhii Dovhyi (also Sergey Dovgiy, Sergey Designer), a Ukrainian senior motion designer from Dnipro. Contact: sergey@sdtech.page. ## OKO-1 An open-source robot that keeps any webcam locked on your face. “Oko” means “eye” in Ukrainian. - A separate ESP32-CAM (AI-Thinker, OV2640) detects the face on-chip with Espressif esp-dl at about 7.2 detections per second and sends coordinates over UART to a main ESP32. - The main ESP32 drives two MG90S servos (pan IO18, tilt IO19) with lag compensation and critically damped easing. - The user's own webcam clips onto the head and stays connected to the laptop by USB; OKO-1 never touches its video. - Wi-Fi and Bluetooth are never started. No app, no cloud, no account. - Print: PETG, one plate, 134 g, 5 h 16 min, no supports. Mounts on a laptop lid or a desk stand. - Pan ±80°, tilt −65° / +40°. - Licence: code MIT; models, print files and docs CC BY 4.0. - Parts cost ≈$22 (AliExpress item prices before shipping, checked 18 Nov 2026): - 1 × ESP32 DevKit (ESP32-WROOM-32, CP2102): $4.4 (Main board: servo control and USB power) - 1 × ESP32-CAM AI-Thinker + OV2640: $6.4 (Tracking eye: on-device face detection) - 2 × MG90S metal-gear micro servo (pack of 2): $6.15 (Pan and tilt) - 1 × Dupont jumpers 20 cm (F-F + M-F, 40 each): $3.2 (4 to the camera, 2 per servo) - 1 × WAGO-type 3-way lever connector: $0.2 (Splits USB 5 V to both servos (sold in packs of 10, ≈$1.9)) - 1 × PETG filament, 134 g of a 1 kg spool: $2 (Whole kit, one plate, no supports) - 4 × Screws M2.5-M3 x 8-10 mm: n/a (Plus the screws from the servo bags) ## Pages - [OKO-1 overview](https://sdtech.page/): product, specs, real photos, price comparison, downloads, FAQ - [How to make any webcam follow your face](https://sdtech.page/oko/how-it-works/): gimbal webcams vs software auto-framing vs a motorized mount, and how OKO-1 tracks a face - [ESP32-CAM face tracking with servos, explained](https://sdtech.page/oko/esp32-cam-face-tracking/): camera fast path, UART protocol, latency-compensated control law, smoothing, power gotchas, tuning commands - [Build guide](https://sdtech.page/oko/build/): print settings, flashing, wiring tables, assembly, troubleshooting, serial commands - [Founder](https://sdtech.page/about/): Serhii Dovhyi and SDTech - [Українська версія](https://sdtech.page/uk/): усі сторінки українською (OKO-1, як це працює, прошивка, збирання, засновник) - [Source code and files](https://github.com/ghostoman/oko) ## How OKO-1 tracks a face 01. See: An OV2640 sensor on the head captures 320 × 240 frames. This is a separate tracking camera, not your webcam. 02. Detect: Espressif’s esp-dl face detector runs on the ESP32-CAM chip itself, about 7.2 detections per second. After the first lock it searches only a window around your face, which keeps it fast. 03. Report: Only the face position leaves the camera: a few bytes over two wires (UART at 921 600 baud). No pictures, no radio. 04. Compensate: Each report is about 80 ms old when it arrives. The main ESP32 keeps a 640 ms history of servo positions, looks up where the head was pointing when that frame was taken, and corrects from there. Delay never turns into overshoot. 05. Move: A 2° dead zone keeps the head still while you sit still. Bigger offsets become eased moves: about 500 ms to settle, capped at 90°/s, no overshoot. 06. Rest: Lose the face for 6 seconds and OKO-1 returns to the centre and waits. ## Use cases - Video calls: Lean back, stand up, reach for a notebook: you stay in the middle of the picture in Zoom, Meet, Teams or Slack. - Teaching and demos: Move between the desk and a whiteboard or a product on the table without asking someone to fix the camera. - Standing desks: Switch between sitting and standing mid-call; tilt follows your height. - Streaming and recording: Talking-head videos and tutorials with a camera that keeps up when you move. Works in OBS like any webcam, because it is your webcam. ## Wiring - ESP32-CAM 3V3 -> main ESP32 3V3 - ESP32-CAM GND (next to IO0) -> main ESP32 GND - ESP32-CAM U0T -> main ESP32 IO16 - ESP32-CAM U0R -> main ESP32 IO17 - Pan (in housing) servo: signal IO18, GND GND, +5 V via WAGO - Tilt (on yoke) servo: signal IO19, GND GND, +5 V via WAGO ## Downloads - [Print plates (Bambu Studio 3MF)](https://sdtech.page/downloads/OKO-1_print_PETG.zip): Full kit on one plate + fit-test plate, sliced for P1S / X1C PETG. - [STL files](https://sdtech.page/downloads/OKO-1_STL.zip): Every part already in print orientation. Any slicer. - [STEP CAD](https://sdtech.page/downloads/OKO-1_STEP.zip): Parts and full assembly for Fusion, FreeCAD, SolidWorks, Onshape. - [Firmware](https://sdtech.page/downloads/OKO-1_firmware.zip): ESP32 main board + ESP32-CAM sketches and flashing scripts. ## FAQ ### Does it work with any webcam? Yes. OKO-1 does not use your webcam for tracking at all. A separate ESP32-CAM on the head finds your face, and your webcam just rides on top and stays plugged into the laptop as usual. Anything with a monitor-style clip fits the head rib; it was developed with a Logitech Brio 100. ### How is OKO-1 different from a gimbal webcam like OBSBOT or Insta360 Link? Those are complete cameras with motors and tracking built in, from about $99 to $199. OKO-1 is a motorized mount that adds the same idea to the webcam you already own, for about $22 in parts. It is fully open source, so you can print, repair and modify every piece. ### Does it work on Windows, macOS and Linux? Yes. Your computer only sees your own webcam, exactly as before. OKO-1 takes 5 V from any USB port or phone charger and needs no driver, app or account to track. ### Is my video sent anywhere? No. Face detection runs on the ESP32-CAM chip itself. Wi-Fi and Bluetooth are never started in the firmware, the tracking camera only sends face coordinates over two wires to the main board, and your webcam picture never touches OKO-1. ### How fast does it track? About 7.2 face detections per second on the ESP32-CAM, measured on hardware. Between detections the servos move with critically damped easing, so motion stays smooth instead of jerky. ### What do I need to build one? A 3D printer (PETG), an ESP32 DevKit, an ESP32-CAM with OV2640, two MG90S servos, a few Dupont wires and a WAGO connector. About $22 in parts. No soldering and no separate programmer. ### I do not have a 3D printer. Can I still build one? Yes. Send the STL files to any online 3D printing service or a local makerspace and ask for PETG. It is one small plate of about 134 g, so it is a cheap job. Everything else is off-the-shelf electronics. ### Is it noisy? MG90S micro servos make a soft whir while the head moves and stay silent when you sit still, thanks to the 2° dead zone. ### Can I use it on a desk instead of a laptop? Yes. Print the optional desk stand and slide the housing into it. ### What is the licence? Firmware and code are MIT. 3D models, print files and documentation are CC BY 4.0. Build it, sell it, remix it; keep the credit to SDTech / Serhii Dovhyi. ### Can I hire SDTech to build something like this? Yes. SDTech designs AI-driven devices, firmware, 3D-printed enclosures and the websites that launch them. Write to sergey@sdtech.page. ### Can I make my existing webcam track my face without buying a new one? Yes. Mount it on a motorized pan/tilt head that has its own tracking camera, like OKO-1. The webcam keeps working as a normal USB camera, and the head turns it toward you. OKO-1 costs about $22 in parts and is fully open source. ### Is software auto-framing the same as a tracking webcam? No. Software framing crops a wider picture and moves the crop, so it can only follow you inside the original field of view and the image gets softer when it zooms. A motorized mount physically turns the camera and keeps the full resolution. ### Does face tracking send my video to the cloud? Not with OKO-1. Face detection runs on the ESP32-CAM chip, Wi-Fi and Bluetooth are never started, and only face coordinates travel over a wire. Your webcam picture never passes through the device. ### Which webcams fit? Any webcam with a monitor-style clip that fits on the head rib. It was developed with a Logitech Brio 100. Light 1080p and 2K clip-on webcams are the intended range. ### How hard is it to build? One evening, no soldering and no separate programmer. You print one PETG plate, flash two boards from a computer, connect a few Dupont wires and screw it together. The build guide and a copy-paste prompt for Claude walk you through it. ## Build with Claude Paste this prompt into Claude to get a step-by-step build assistant: ``` You are my build assistant for OKO-1 by SDTech, an open-source face-tracking turret for any webcam. Project repo: https://github.com/ghostoman/oko Site: https://sdtech.page Hardware: ESP32 DevKit (main board), ESP32-CAM AI-Thinker with OV2640 (tracking camera, on-device face detection), 2x MG90S servos (pan IO18, tilt IO19), WAGO 221-413 for 5 V, Dupont wires. ESP32-CAM -> main: 3V3-3V3, GND-GND, U0T-IO16, U0R-IO17. Print: PETG, 0.20 mm, 4 walls, 20 % gyroid, 245 °C, no supports, 134 g, one plate. Walk me through the build one step at a time and wait for my "done" before the next step: 1. Ask what I already have (printer, slicer, boards, OS) and make me a shopping list with what is missing. 2. Printing: which files to download from the repo, how to slice, what to check after the fit-test plate. 3. Flashing: install Arduino CLI / esp32 core, flash the main board, then the ESP32-CAM through bridge mode. 4. Wiring with exact pin tables, then mechanical assembly in the order from the repo manual. 5. First run and tuning (flip pan/tilt, aimx/aimy, smooth), and troubleshooting if I paste the serial monitor output. Explain like I am new to electronics, keep each message short, and warn me before anything that could damage a board. ``` ## About the founder Serhii Dovhyi (Сергій Довгий; older spelling Sergey Dovgiy; brand Sergey Designer) is a Senior Motion Designer and developer from Dnipro, Ukraine: 10+ years in motion design, 330+ projects, Top Rated Plus on Upwork, founder of SnipeWork. After a successful career in design and software development (websites, web apps, SaaS), he moved on into hardware and robotics, building with AI engineering tools such as Codex and Claude. OKO-1 is an open-source SDTech device. - Portfolio: https://sergeydesigner.pro - GitHub: https://github.com/ghostoman - LinkedIn: https://www.linkedin.com/in/sergey-dovgiy/ - Telegram: https://t.me/ghostoman