Amber Edge Assistant
Access small, fast, free open-weight models from the Edge Gallery. Run completely locally on Windows, Linux, macOS, iOS, and Android with zero telemetry.
Download Amber AI
Bundled native binaries and standalone PWA installers for all 5 major platforms.
Windows 10 / 11
64-bit standalone package. Embedded JVM runtime bundled with zero Python or Java install needed.
Android
Optimized for Snapdragon, MediaTek, and Tensor NPU acceleration on phones and tablets.
macOS
Universal binary with Apple Silicon (M1–M4) Metal acceleration and Intel x86_64 support.
Linux
Universal runner for Ubuntu, Debian, Fedora, Arch Linux, and openSUSE distributions.
iOS (iPhone & iPad)
Zero App Store friction. Standalone fullscreen Progressive Web App with offline service worker.
Open Source Hub
View full source code, Kotlin Multiplatform build configs, and quantized GGUF weights on GitHub.
Small, Free Edge Models
Curated lightweight models under 4GB RAM. Download quantized weights or select for in-app execution.
Qwen 2.5 0.5B Instruct
Sub-500MB multilingual titan. Excellent coding, reasoning, and mathematical logic with an ultra-low memory footprint.
SmolLM2 135M Instruct
The world's premier sub-100MB model. Instantaneous execution on smartwatches, wearables, and low-end phones.
SmolLM2 360M Instruct
Superb reasoning-to-size ratio. Squeezes maximum instruction adherence into 240MB with minimal battery drain.
Llama 3.2 1B Instruct
Meta's official edge foundation model with massive 128k context and high instruction fidelity on mobile devices.
Qwen 2.5 1.5B Instruct
Rivals prior-generation 7B models in complex coding and reasoning benchmarks. Balanced for laptops and phones.
DeepSeek R1 Distill Qwen 1.5B
Revolutionary on-device reasoning engine with step-by-step thinking traces (<think>...</think>) before answering.
Gemma 2 2B IT
Built upon Gemini technology stack. Class-leading world knowledge and exceptional conversational flow.
Llama 3.2 3B Instruct
Meta's flagship compact model. Rich, nuanced responses, software development assistance, and 128k context.
Phi-3.5 Mini 3.8B Instruct
Microsoft's edge champion. Outperforms many larger models on reasoning, math, and code generation.
curl -X POST http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "qwen-2.5-0.5b",
"messages": [
{"role": "user", "content": "Explain quantum superposition in simple terms"}
]
}'