Local AI readiness
Checking your browser…
This does not download an AI model.
Local-first AI lab
See what AI your browser can run.
Test your device, download a small open model, and run it privately—without creating an account, entering an API key, or sending your text to a server.
- No API key
- No prompt upload
- Works after download
runanyai / local-route live
01
InspectCheck browser capabilities
done02
MatchChoose a compatible model
local03
RunKeep user data on device
privateinputyour browserresult
Device check
Your browser is the compute.
We inspect only capabilities exposed by your browser. Results remain on this page and are not saved or transmitted.
WebGPUFor faster, larger models
CheckingWebAssemblyReliable CPU fallback
CheckingCPU threadsParallel local processing
CheckingModel storageBrowser-managed cache
CheckingLive local model
Run real AI on your CPU.
This first experiment detects the tone of English text. The model runs entirely inside your browser using WebAssembly.
Model 001
Local tone reader
First run: approximately 70 MBOne quantized model plus small tokenizer and runtime files.
Your text will not be included in the model download request.
Model downloads, plainly
The model comes to you.
Traditional AI sends your prompt to a company’s server. Local AI reverses that: the model files download once, then your device performs the work.
You approve the download
Nothing large downloads when the page opens. The model starts only after you press run, with its approximate size shown first.
Your browser caches it
Model weights are static numerical files. Your browser stores them in its site cache so later visits can reuse them.
Your input stays local
Once loaded, the model processes text in browser memory. This demo has no endpoint that receives your input or output.
Public model hostModel filesDownloaded on first use
≈70 MB→
Your browser cacheLocal modelReusable on later visits
Your text never travels↺
Your devicePrivate resultCPU inference
What comes next
One runtime, progressively more capable.
We start with a small CPU model because it works on the widest range of devices. Later experiments will remain explicit about size, hardware, and where processing happens.
Available nowDevice check + tone modelCPU · about 70 MB
NextPrivate summarizationWebGPU · larger download
LaterLocal transcriptionCPU or WebGPU