Deploy AI

Pack requirements, preflight GPUs, land a checklist on the host. Weights stay content-addressed in the cache.

Pack a manifest

buraaq ai pack Qwen/Qwen3-8B

Writes target/ai/manifest.json and GPU_CHECK.txt:

min_vram_gb, recommended_backend, quantize, adapter_path (if you trained)
require_gpu when BURAAQ_AI_REQUIRE_GPU=1
No hundred-gigabyte weights inside the file

Ship preflight

buraaq ai ship Qwen/Qwen3-8B HOST

Prints local GPU vs min_vram_gb, then next steps for buraaq pack / buraaq ship. Strict mode:

$env:BURAAQ_AI_REQUIRE_GPU = "1"
buraaq ai ship MODEL HOST

Fails clearly if VRAM is insufficient.

Land an AI host

buraaq land --ai
buraaq land user@HOST --ai --cloud hetzner

Adds ai-check.sh and AI.txt to target/land/. On the host:

bash ai-check.sh
buraaq ai doctor
buraaq ai serve MODEL --bind 127.0.0.1

Never auto-installs GPU drivers. You install vendors yourself; the kit only checks.

Flow

train (optional) → ai pack → buraaq pack (.bur app)
                           → land --ai → ai-check.sh on GPU host
                           → buraaq ship HOST → ai serve

Back to Mind path or Ship overview.