Talk to your Google TV. It looks at the screen and does it for you.
An AI voice remote for Android / Google TV, built with Kotlin and Jetpack Compose (Material 3 Expressive).
Say "play Dhruv Rathee's latest video", "resume my show on Nuvio" or "volume up 3", and TV Pilot does it on your TV.
- Vision agent. A multimodal model (Gemini 3.5 Flash-Lite by default, or any OpenAI-compatible API that takes images) gets a screenshot of the TV. It picks one action (tap, D-pad key, swipe, type, open app, search, wait), runs it, then looks again until the job is done. A live card shows the TV screenshot, a checklist of steps, and a Stop button.
- Instant remote commands. Pause, volume, back, home and similar never wait on a model.
- Voice, your way:
- Gemini 3.5 Transcribe. Handles Hinglish and code-switching, and is biased toward the names of your installed apps.
- Phone. Android's built-in speech recogniser.
- Parakeet. Your own server (any OpenAI-style
/v1/audio/transcriptionsendpoint).
- Custom instructions. Rules the agent always follows, per app. For example: "In Nuvio, after pressing Play, wait for providers, skip 'Support the project', pick PenguPlay."
- Bluetooth remote. The phone acts as a Bluetooth HID media remote, so your own button presses reach the TV with no ADB lag. The agent always uses ADB, so the two never clash.
- Home-screen widgets. A frosted 2×2 mic and a 4×2 remote with status and play, volume and home buttons. Tap the mic to start talking.
- Backup brain (optional). A small decision model (e.g. Liquid AI d1) on your own server takes over if the vision API fails or runs out of quota.
- Polished UI. A focus home with a big morphing mic, a smooth mic-to-bar transition, an AMOLED black mode, and a remote and apps sheet.
- Works with any app. YouTube / TizenTube, Netflix, Prime Video, JioHotstar, Crunchyroll, Nuvio, Stremio, Plex and more. Installed TV apps are detected on their own.
voice ─▶ transcription ─▶ quick command? ──yes──▶ key press (Bluetooth or ADB)
│ no
▼
vision agent loop: screenshot ─▶ model ─▶ one action ─▶ repeat until done
│ API error
▼
optional backup decision model
The phone talks to the TV directly over network ADB with dadb. No PC, no root, and no app on the TV.
- On the TV: turn on Developer options (Settings › System › About › tap Build 7 times). Then turn on USB debugging and Network debugging (on some TVs, ADB over network).
- Find the TV's IP: Settings › Network & Internet › your Wi-Fi.
- In TV Pilot › Settings:
- Enter the TV IP (port
5555), then tap Save and connect. Accept "Allow USB debugging?" on the TV and tick Always allow. - Brain: choose Vision model and paste a Gemini API key from AI Studio. The default URL is
https://generativelanguage.googleapis.com/v1beta/openai. - Voice: pick Google › Gemini API (the key field can stay blank to reuse the brain key) or Phone.
- (Optional) Remote: turn on Bluetooth remote and tap Pair with TV.
- Enter the TV IP (port
- Tap the mic and talk.
All server fields ship empty. Bring your own keys and servers.
Requirements: JDK 21, Android SDK 37.
./gradlew assembleReleaseOptional build-time defaults (kept out of the source):
./gradlew assembleRelease -Pd1Url=https://your-server -Pd1Key=... -PasrUrl=https://your-asr-server
# or the env vars D1_URL, D1_KEY, ASR_URLAdd -PabiSplits=1 for separate arm64-v8a and armeabi-v7a APKs.
To keep updates installable over older builds, sign with your own fixed keystore: -PtvpilotKeystore=/path/to/your.keystore.
Kotlin · Jetpack Compose · Material 3 Expressive (1.5.0-beta01) · DataStore · OkHttp · dadb (ADB) · ML Kit text recognition · Android Bluetooth HID Device API · Gemini API (OpenAI-compatible chat + Interactions API for transcription)
Keys and settings stay on your phone (DataStore). Screenshots and audio go only to the APIs you configure.
MIT © 2026 Syed Arshaan (@Ichiro786)