Local AI Chat

Loading conversations...

Choose a model

The model is the part that performs the AI work. Downloading its weights lets this browser run it on your device instead of sending your text, images, or audio to an AI server. Your browser caches it for later use.

Loading model... Preparing on this device 0%

Start a conversation

Pick a prompt or write your own.

Local by default. The AI model runs in your browser via WebGPU. A search query leaves your device only if you enable an external web-search provider.

Your first message asks before downloading the model (…), once. After that it works offline.

Setting up your local model…

One-time download of … — it runs offline after this. Watch the progress bar above.

Web search on — next message only

• • • • Ready when you are

Chat & Model Settings

Configure parameters for local inference

System Instructions Guides model persona & tone

Thinking mode

Shows the selected model's reasoning in a collapsed panel before its answer.

Temperature 0.7

Precise (0.0) Balanced (0.7) Creative (1.5)

Max Generation Tokens 1024

256 1024 4096 (Long form)

Web Search (optional) No key saved

Optional & opt-in per message via the Search chip in the input dock. Your key is sent directly from your browser to the provider, not to SEDER LTD or an application backend, and is stored only in this browser's localStorage. Free keys: Tavily (1,000 credits/mo), Brave (free tier).

Chat with your files

Attach a PDF or text file as searchable knowledge for this chat

or drop one here

0 / 200,000 chars

PDF, Markdown, CSV, code and other text files are parsed 100% locally in your browser. Relevant excerpts are retrieved per message and injected into the chat context. If a long document needs the ~22 MB search model, WKO AI asks before downloading it. Scanned or password-protected PDFs cannot be read.