Windows and macOS. The Windows packages are built from 3203c39 on main, the macOS package from f833336 on main. Installation and troubleshooting: see README.md inside each package.

What's Changed

• Edge0 145B and Edge0 120B. Two latent-MoE models for a single NVIDIA GPU, run by edge0-strata.
• New inference engine, edge0-infer. A single on-device MoE engine in C++ on GGML, with one native API, replaces the Python framework and the four separate platform apps. It includes a CLI (run, chat, bench, serve, info), an OpenAI-compatible local server and the Edge0 Chat desktop app. This release supports Edge0 35B; Edge0 8B follows in a later release.
• macOS (Apple Silicon). Edge0 145B and Edge0 120B also run on Metal, and Edge0 Chat comes as a macOS package with the engines inside. (#214)
• Anthropic Messages API. The local server also answers POST /v1/messages and POST /v1/messages/count_tokens for native models, and LAN authentication accepts x-api-key. (#180)

Windows x64

edge0-chat-0.2.0-windows-x64.zip

• Edge0 35B runs on the CPU (AVX2) or on a GPU through Vulkan.
• Edge0 145B and Edge0 120B run on an NVIDIA GPU (CUDA) of compute capability 7.5 or newer (RTX 20 / 30 / 40 / 50) with at least 16 GiB of GPU memory, 20 GiB of system memory, and the NVIDIA CUDA Toolkit 13 installed.

SHA256: c7146b76d51c7986723b81a3643893e1ccbb09ce4f365a5090dc99e476bb31b9

Windows Arm64

edge0-chat-0.2.0-windows-arm64.zip

• Edge0 35B runs on the CPU or on a GPU through Vulkan.
• Edge0 145B and Edge0 120B run on an NVIDIA GPU (CUDA) of compute capability 12.1, such as the NVIDIA RTX Spark N1X, with at least 16 GiB of GPU memory, 20 GiB of system memory, and the NVIDIA CUDA Toolkit 13 for Windows Arm64 installed. The CUDA code is built for sm_121 only.

SHA256: f64accb2c963058bc4ff1c9b064ccb35dc62ccdf8881b9a797d1ddab136f500a

macOS (Apple Silicon)

edge0-chat-0.2.0-macos-arm64.zip

• macOS 26 or later on Apple Silicon. edge0-infer and edge0-strata are inside Edge0 Chat.app. The app is signed ad hoc and not notarized: on first start, allow it under System Settings > Privacy & Security (see README.md in the package).
• Edge0 35B runs on Metal or the CPU.
• Edge0 145B and Edge0 120B run on Metal from their q3k folders only; the NVFP4 folders do not run on macOS.

SHA256: 4ece2c38302afd7b9610cd4cfadad9a82586c882d83b5eb3b31782efac6ea4e5

Models

Model weights are not included in the packages.

• Edge0 145B: Edge0-145B-NVFP4, Edge0-145B-q3k
• Edge0 120B: Edge0-120B-NVFP4, Edge0-120B-q3k

Full Changelog: v0.1.1...v0.2.0

Android only. macOS and iOS remain the 0.1.0 builds.

Android (arm64, Android 13+)

edge0-0.1.1-android-arm64.apk

Built from d4a2fa02fcf0 on main, which includes #155. versionName 0.1.1 (versionCode 2), arm64-v8a only. Signed with the same certificate as 0.1.0, so it can replace that install. Not published on Google Play. Allow installation from this source, then open the APK.

Models has Import. Choose a .gguf and the app copies it into its private storage. USB debugging is not required. For 8B, import lora_edge0_8b-gguf.gguf the same way, then load the model.

SHA256: a4cf0253f0d8169948cad8688a7cac6e5d1b500994f95558d03a9a8f96a546a2

Signing certificate SHA-256: bd2ea50337e879bdc190041765877f949ae9ad959aaf27c7255e5e8a8a714e26

macOS and Android are built from 7a203f2 on main. The iOS IPA also includes bf267d4, which lets the app read a model copied in through the Files app.

macOS (Apple Silicon)

edge0-0.1.0-arm64.zip

This machine has no Developer ID Application certificate and no notarization credentials, so the app is ad-hoc signed. Gatekeeper will block it after download. After unzipping, clear the quarantine attribute, then open it:

xattr -dr com.apple.quarantine edge0-0.1.0-arm64.app
open edge0-0.1.0-arm64.app

SHA256: 6e7824230d5201c02bc0a192969841e0ad19482c7847030220aa0f4db7ca34fb

Android (arm64, Android 13+)

edge0-0.1.0-android-arm64.apk

versionName 0.1.0 (versionCode 1), arm64-v8a only. Signed for sideloading, not published on Google Play. Allow installation from this source, then open the APK.

Model weights are not in the package. After install, copy a converted GGUF in with the in-app picker (files/models/), or stage one with android/tools/model/push_models.sh. See android/README.md section 1.4.

SHA256: 6ed9e44b912216c99bc05986a549dbc574b22adc2c6a35ee7bdd5e6129d43196

Signing certificate SHA-256: bd2ea50337e879bdc190041765877f949ae9ad959aaf27c7255e5e8a8a714e26

iOS (iPhone, iOS 17+)

edge0-0.1.0-ios.ipa

Unsigned sideload build, bundle id com.example.edge0phone, app version 1.0 (1). Install with Sideloadly, AltStore, or SideStore and your own Apple ID; the tool re-signs the IPA. A free Apple ID signature lasts about 7 days.

Model weights are not in the package. After install, connect the phone and copy a model into the app's documents with Finder:

• 8B: Edge0-8B-A1B-preview/ containing model.safetensors
• 35B: repacked/ containing model.safetensors.index.json, tokenizer.bin, and experts-L00.bin through experts-L39.bin

The folder can sit directly in the app's documents, or under Models/.

SHA256: 7083ff771019ec4dc1f283c5de83794b6d227bdb950a9a961864ead587b0a5e5