Download 1bit.MONSTER.

One binary. Every server and CLI. Pure C++, zero Python at runtime. Pick a format below — tarball, Debian package or AppImage — and be chatting with an NPU model in minutes. Latest release: 2026.08.04

Or install from source in one line.

clone · build · no pip

Prefer to build yourself, or want the NPU engine plus the coding-agent CLI? install.sh clones the repo and builds the single 1bit binary on your Strix Halo box.

curl -fsSL https://1bit.monster/install.sh -o install.sh
bash install.sh            # build + nothing else
bash install.sh --with-jarvis  # build + JARVIS launcher + config

Kernel check: Strix Halo (gfx1151) systems on Linux 6.19.x can hit an amdgpu display hang under sustained NPU/GPU load (issue #1). install.sh detects 6.19.x and warns — use 6.18.22-lts or the 7.x series.

Release packages.

built from the same staged tree · sizes & checksums load live

Binary tarball (13 MiB)

1bit-monster-2026.08.04-linux-amd64.tar.xz

Extract anywhere, run ./run.sh. The 1bit ELF holds every server + CLI; a loader script wires up librocm_cpp.so for you.

sha256 —

Download .tar.xz

Debian package (14 MiB)

1bit-monster_2026.08.04_amd64.deb

Installs 1bit (and its legacy symlinks) onto PATH plus the shared HIP kernel library and Vulkan shaders.

sha256 —

Download .deb

AppImage (21 MiB)

1bit-monster-2026.08.04-x86_64.AppImage

No install needed — chmod +x and run. Portable single-file desktop app for Linux.

sha256 —

Download .AppImage

Verify before running: SHA256SUMS · machine-readable: manifest.json

Every package is the same tree. The tarball, .deb and AppImage are built from one staged release tree (make package-site), so picking a format is purely about how you like to install — the contents are identical.

Other formats & channels.

configs in the repo — build when you need them

DockerDockerfile readypackaging/docker ↗ AURPKGBUILD readypackaging/aur ↗ Snapsnapcraft.yaml readypackaging/snap ↗ Flatpakbuild script readypackaging/flatpak ↗ RPMspec readypackaging/rpm ↗ OllamaModelfile — point at the NPU serverpackaging/ollama ↗

Requirements.

Strix Halo only, for now

Component	Requirement
Hardware	AMD Ryzen AI Max+ 395 (Strix Halo, gfx1151) — NPU or Radeon 8060S GPU
OS	Ubuntu 24.04 LTS or later (Debian-family); AppImage runs on most glibc Linux
Kernel	6.18.22-lts or 7.x — avoid 6.19.x (amdgpu display hang, issue #1)
ROCm runtime	ROCm/TheRock 7.15.0a runtime (`libhsa-runtime64`) for the GPU/HIP path; the NPU engine needs XRT (`amdxdna`)
Model file	A `.h1b`, `.q4nx` or `.gguf` model — see Models

Quick start.

tarball path

tar xJf 1bit-monster-2026.08.04-linux-amd64.tar.xz
cd 1bit-monster-2026.08.04
export HSA_OVERRIDE_GFX_VERSION=11.5.1
export HSA_ENABLE_SDMA=0
./run.sh chat           # NPU-native coding-agent CLI

Then point any OpenAI-compatible client at http://127.0.0.1:13305/v1 — see Getting started for the full walkthrough.