Documentation Index
Fetch the complete documentation index at: https://mintlify.com/RealComputer/GlassKit/llms.txt
Use this file to discover all available pages before exploring further.
The rokid-overshoot example is the simplest way to get live AI scene descriptions running on Rokid Glasses. A temple tap starts streaming; the glasses send live camera video over WebRTC to the FastAPI backend, which creates and manages an Overshoot VLM stream. As Overshoot returns inference results, the backend relays each line back over a WebSocket to the glasses, where it appears as a rolling auto-scrolling log on the HUD. Tap again to stop. This example is also the foundation that rokid-overshoot-openai-realtime builds on for more advanced multi-modal workflows.
User Experience
- Tap the temple area to start streaming.
- Tap again to stop.
- While running, new result lines appear at the bottom of the HUD and the view auto-scrolls.
- Each new run starts with a clean screen.
Architecture
Rokid Glasses (Android)
├── Temple tap → start/stop
├── Camera → WebRTC offer → Backend (FastAPI)
│ ├── Creates Overshoot stream (HTTP POST /streams)
│ ├── Returns WebRTC answer SDP to glasses
│ └── Opens Overshoot WebSocket → relays result text
└── WebSocket ← backend ← Overshoot inference events
└── HUD: append result line, auto-scroll
| Component | Location | Language |
|---|
| Glasses app | rokid/ | Kotlin |
| FastAPI session manager | backend/ | Python 3.12 |
rokid/) handles temple-tap start/stop, camera capture, WebRTC offer/answer flow, and HUD rendering. OvershootSessionClient manages the WebRTC and backend WebSocket. MainActivity appends each result line to a rolling log and auto-scrolls while the session is active.
The backend (backend/) is built around OvershootSessionManager. On session create it sends an HTTP POST /streams to Overshoot with the WebRTC offer, processing config, and inference prompt, receives the answer SDP, then opens a WebSocket to wss://api.overshoot.ai/v0.2/ws/streams/{stream_id} to receive inference events. Each result field in those events is forwarded to the Android app over the backend WebSocket. The manager also runs a keepalive task to renew the Overshoot stream lease, and reconnects the WebSocket with exponential backoff if it drops.
Requirements
Configuration
Configure the glasses app
Set the backend base URL in rokid/local.properties:BACKEND_BASE_URL=http://<YOUR_BACKEND>
Configure the backend
cd backend
cp .env.example .env
# Set OVERSHOOT_API_KEY in .env
Optional Backend Overrides
These environment variables let you tune inference behaviour without changing code. Default values are defined in backend/session_manager.py.
| Variable | Default | Description |
|---|
OVERSHOOT_API_URL | https://api.overshoot.ai/v0.2 | Overshoot API base URL. |
OVERSHOOT_PROMPT | "You are observing a first-person POV. Describe the scene in second person. Write at most three short sentences." | Inference prompt sent to the VLM. |
OVERSHOOT_MODEL | Qwen/Qwen3-VL-30B-A3B-Instruct | Overshoot model identifier. |
OVERSHOOT_PROCESSING_TARGET_FPS | 6 | Target frame rate for clip processing. |
OVERSHOOT_PROCESSING_CLIP_LENGTH_SECONDS | 0.5 | Duration of each processed clip. |
OVERSHOOT_PROCESSING_DELAY_SECONDS | 0.5 | Delay before each clip is sent for inference. |
Run the Backend
cd backend
uv run --env-file .env fastapi dev main.py --host 0.0.0.0
Run the Glasses App
Connect Rokid Glasses and enable Wi-Fi
adb devices # confirm device is visible
adb shell cmd wifi status # check connection
adb shell cmd wifi set-wifi-enabled enabled
adb shell 'cmd wifi connect-network "NAME" wpa2 "PASSWORD"'
adb shell cmd wifi status # confirm connection
Optional: wireless ADB
adb shell ip -f inet addr show wlan0 # check glasses IP
ping -c 5 -W 3 <IP> # first ping may time out
adb tcpip 5555 # enable remote ADB mode
adb connect <IP> # connect over Wi-Fi
adb devices # verify remote connection
Build and run
Open the rokid/ directory in Android Studio, select Rokid Glasses as the target device, and run the app.To rebuild manually after code changes:cd rokid && ./gradlew :app:assembleDebug
Key Files
| File | Description |
|---|
rokid/…/MainActivity.kt | Temple-tap start/stop controls and rolling result log UI. |
rokid/…/OvershootSessionClient.kt | WebRTC offer/answer flow and backend WebSocket handling. |
rokid/…/activity_main.xml | Monochrome HUD layout with auto-scrolling log view. |
backend/main.py | FastAPI app lifecycle and HTTP/WebSocket route handlers. |
backend/session_manager.py | Overshoot session orchestration: REST create/delete, keepalive, WebSocket relay, reconnect/backoff, and env-driven processing config. |
backend/.env.example | Environment template with OVERSHOOT_API_KEY and optional overrides. |
Next Steps
For a more advanced pattern built on the same Overshoot foundation, see the Proactive Drink-making Coach. It adds OpenAI Realtime for spoken guidance, a recipe workflow, and server-authoritative HUD state, but uses the same WebRTC brokering and Overshoot session management patterns you see here.