inAI Chat

Runs in your browser · model files load from chain inscriptions on this domain

Loading the assistant…

First visit can take several minutes while weights are fetched and cached locally.

Starting…

Hardware acceleration — WebGPU (nvidia‑smi ≠ browser GPU)

NVIDIA can work in CUDA but stays invisible to Chat until the browser is allowed to use the GPU stack WebGPU expects.

  1. Chrome: Menu → Settings → search hardware → turn Use graphics acceleration when available ON → Relaunch.
  2. Edge: edge://settings/system → same toggle → Restart.
  3. Check diagnostics: Chrome chrome://gpu · Edge edge://gpu — ensure WebGPU is not disabled; note “Vulkan” / “Passthrough” on Linux NVIDIA.
  4. Linux + NVIDIA proprietary: fully quit Remote Desktop nested sessions if possible; Wayland/X11 quirks exist — compare a normal desktop session vs RDP/VNC-only.
  5. Firefox often lacks production WebGPU; use Chrome or Chromium for WebLLM.
  6. External verifier: webgpureport.org — if it fails here, inAI cannot fix that from inscription code.
Technical settings