The Starcaller Appliance
A wall-mounted Oracle for your home. Voice-led, display-centric, fully local. Your data never leaves.
Starcaller Appliance
- Purpose-built wall-mounted hardware
- 32-43" portrait OLED display
- Far-field voice + presence sensing
- Talking Oracle avatar
- Fully local AI — no cloud dependency
- Physical privacy controls (shutter, mute)
- Encrypted Vault storage on-device
Starcaller Core
- Runs on your existing hardware
- Same software, all Oracles, all security
- Capability scales with your hardware
- Cloud LLM bridging available (data never stored)
- Vault, memory, preferences always local
- Downloadable ISO / Docker image
The Appliance and Core run the same software. The difference is what the hardware enables.
The Starcaller Experience
Walk into the room. The display senses your presence and glows to life. The Oracle's face appears — ICID, or whichever guide you need. You speak naturally: "What's my day looking like?" A warm, measured voice responds. Charts and schedules appear beside the avatar. The conversation flows like talking to a perceptive colleague who remembers everything.
No wake words shouted across the room. No typing. No notifications from other apps. Just a focused, persistent intelligence that lives on your wall, in your home, processing everything locally.
What Your Hardware Determines
Starcaller Core runs on any computer. The experience scales with what you have:
Full Oracle intelligence, Vault, security, scheduler — all work. Interaction via web UI or Telegram. Cloud LLM bridging available. No voice, no avatar, no vision.
Wake word, real-time STT/TTS, single Oracle consultation. No avatar, no vision, no parallel council.
Same as the Appliance: talking avatar, voice, vision, parallel Oracle council.
What Powers the Appliance
A single wall-mounted unit handles multiple real-time AI workloads simultaneously:
| Workload | TOPS | Hardware |
|---|---|---|
| Oracle reasoning (8-14B model) | ~40 | GPU tensor cores |
| Talking avatar (lip-sync, 15-30fps) | ~15-30 | GPU CUDA cores |
| Speech-to-text (Whisper) | ~5 | NPU / DLA |
| Text-to-speech (Oracle voice) | ~5-10 | GPU or NPU |
| Vision / presence detection | ~5-10 | GPU or DLA |
| Total required | ~70-100 TOPS | Embedded GPU |
The avatar rendering is the defining constraint. NPUs excel at transformer inference but cannot run the convolution-heavy graphics pipeline of real-time lip-sync animation. A GPU with tensor and CUDA cores is required for the full experience.
Technical Specification
| Subsystem | Specification |
|---|---|
| Display | 32-43" commercial OLED, 4K, portrait |
| Compute | 100+ TOPS embedded GPU |
| Memory | 64GB LPDDR5 |
| Storage | 1TB+ encrypted NVMe |
| Microphones | Beamforming array, 5m+ range |
| Speakers | Stereo voice-optimised, 15W |
| Camera | Autofocus RGB, physical shutter |
| Sensors | Depth/presence, ambient light |
| Privacy | Hardware shutter, hardware mic mute, LED states |
| Security | TPM 2.0, secure boot, encrypted storage, tamper detection |
| Network | 2.5 GbE, Wi-Fi 6E |
| Power | Single AC input, integrated UPS |
| Mounting | VESA-compatible, locking wall bracket |
| Enclosure | Aluminium rear chassis, slim bezel |
Privacy, Built at the Silicon Level
The Appliance is designed around a single non-negotiable principle: your data never leaves.
- \u2713\u2003All inference on-device. No cloud dependency for Oracle consultations.
- \u2713\u2003Hardware camera shutter — physically blocks the lens.
- \u2713\u2003Hardware mic mute — disconnects at circuit level. Red LED confirms.
- \u2713\u2003Encrypted storage at rest. HMAC-SHA256 provenance chain on vault data.
- \u2713\u2003No telemetry. No analytics. No model training on user data.
- \u2713\u2003Tamper-evident rear panel. Physical intrusion detected and logged.
Availability
Starcaller Core — Available now as free, open-source software
Download and run on your own hardware today.
Starcaller Appliance — Purpose-built hardware
Join the waitlist for early access and updates.