Your model. Your machine.

Run a real language model locally with WebGPU. Add private files for retrieval, or chat with the base model directly. After the model is downloaded, inference runs on your device.

0 knowledge files active

Enter to send · Shift+Enter for newlineBrowser-local AI can be slower than cloud models

MODEL MANAGERWEBGPU

Model

First load downloads model files. Later loads use browser cache. WebGPU is fastest; CPU/WASM is the compatibility fallback.

DEVICE

WEBGPU—

GPU—

MEMORY—

MODEL CACHEBrowser

Checking device capabilities…

INFERENCE

Temperature 0.70 Higher values make responses more varied; lower values make them more deterministic. Max response 512 Maximum new tokens the model may generate for one response. RAG passages 4 Number of semantically relevant chunks inserted into the model context. Chunk words 160 Approximate passage size used when indexing uploaded knowledge.

SESSION

Your model. Your machine.

Unlock vault

Paste text

System prompt

Local AI

Load a local model