This is not a bootable OS. Preview 0.2.0 installs as a host app. No clone, no cargo. Download the archive, run the script, launch Akasha OS Preview. Ce n’est pas un OS bootable. La Preview 0.2.0 s’installe comme appli hôte. Pas de clone, pas de cargo. Téléchargez l’archive, lancez le script, ouvrez Akasha OS Preview.

$ install --preview

# hw.req

• os GRANT Windows 10/11 x64 orou Linux x64
• gpu GRANT NVIDIA + recent driver+ pilote récent (nvidia-smi -L)
• disk ~8 GB free recommended (mid-tier GGUF pack). Binaries plus first-run models. disque ~8 Go libres recommandés (pack GGUF mid). Binaires plus modèles au premier lancement.
• cuda runtime ships in the package; the driver is enough runtime cuda fourni dans le paquet ; le pilote suffit
• macos DENY
• cpu-only DENY in Preview 0.2.0 cpu-only DENY en Preview 0.2.0

# Windows

# Windows

1. Download AgentOS-Preview-<ver>-windows-x64.zip from GitHub Releases.
2. Télécharger AgentOS-Preview-<ver>-windows-x64.zip depuis GitHub Releases.
3. Extract, then run:
4. Décompresser, puis :

.\install.ps1

Installs under %LOCALAPPDATA%\AgentOS-Preview and keeps var/ and etc/ on update. Launch Akasha OS Preview.

Installe sous %LOCALAPPDATA%\AgentOS-Preview et conserve var/ et etc/ à la mise à jour. Lancer Akasha OS Preview.

Without the installer:

Sans installateur :

$env:AOS_HOME = (Resolve-Path .)
.\bin\aos-session.exe

# Linux

# Linux

1. Download AgentOS-Preview-<ver>-linux-x64.tar.gz, extract.
2. Télécharger AgentOS-Preview-<ver>-linux-x64.tar.gz, extraire.
3. ./install.sh
4. Prefix: ~/.local/share/agentos-preview (non-destructive overlay). Run agentos-preview.
5. Préfixe : ~/.local/share/agentos-preview (overlay non destructif). Lancer agentos-preview.

# package

# paquet

bin/            daemons + CUDA runtime
share/models/   manifest.json (GGUF on first run)
share/modules/  notes.aospkg, ext-rt.aospkg
share/skills/   notes-writer, research, file-author, planner
share/mcp/      servers.yaml.example
data/models/    catalog.yaml
VERSION         build semver
var/            created at run

# first launch

# premier lancement

1. aos-session checks NVIDIA and disk, probes VRAM.
2. aos-session vérifie NVIDIA et le disque, sonde la VRAM.
3. Model setup: confirm the auto-best pack for your GPU tier, then GGUFs download into share/models/.
4. Réglage modèles : confirmez le pack auto pour votre palier GPU, puis les GGUF descendent dans share/models/.
5. bus → capkd → auditd → modeld → platformd → agentd
6. egui UI + multi-page tutorial. Closing the UI stops the daemons.
7. UI egui + tutoriel multi-pages. Fermer l’UI arrête les daemons.

Next: first run in the user manual.

Suite : premier lancement dans le manuel.

# troubleshooting

# dépannage

• NVIDIA GPU required → install the driver; nvidia-smi -L
• NVIDIA GPU required → installer le pilote ; nvidia-smi -L
• Model download failed → network for Hugging Face, or copy GGUFs into share/models/
• Échec téléchargement modèle → réseau vers Hugging Face, ou copier les GGUF dans share/models/
• Healthcheck failed → var/run/*.stderr.log (Troubleshooting button)
• Healthcheck failed → var/run/*.stderr.log (bouton Troubleshooting)
• Bus unreachable → always launch via aos-session
• Bus unreachable → toujours lancer via aos-session

# build from source # build depuis les sources

Releases are enough to run Preview. Build from source for a local package, patches, or development. Full write-up: docs/build and INSTALL.md.

Les Releases suffisent pour lancer la Preview. Compilez depuis les sources pour un paquet local, des patches ou du développement. Détail : docs/build et INSTALL.md.

• rust GRANT stable + wasm32-unknown-unknown
• cuda GRANT Toolkit (nvcc) · nvidia-smi OK
• cmake · ninja · clang/libclang (Linux)(Linux)

# Windows
.\packaging\build-preview.ps1 -SkipModels -RequireCuda

# Linux
SKIP_MODELS=1 REQUIRE_CUDA=1 ./packaging/build-preview.sh

# Dev run (repo root)
$env:AOS_HOME = (Resolve-Path .)   # Windows
export AOS_HOME="$(pwd)"           # Linux
cargo run -p aos-session --release

$ open docs/