WebKaya

Private AI chat in Chrome. Ask anything, analyze your files with verified numbers, search when you ask. On-device by default.

# A private AI chat that lives in Chrome Chat naturally, work with your own files, and choose exactly where AI runs. Whether you're asking a question, exploring a spreadsheet, or extracting data from a receipt, WebKaya keeps you in control. ## Chat the way you expect Ask anything. Write, brainstorm, code, summarize, or edit text. Drop in a CSV, receipt, PDF, or image and ask questions about it. Find totals by vendor, explain what's in a document, spot unusual values, or pull out specific information. Every answer links back to the source. When numbers are involved, WebKaya verifies them on your device before showing them. Models can explain the results, but they cannot invent figures. ## Search the web only when you ask Web searches never happen automatically. Say "search the web for..." or paste a URL. WebKaya reads the relevant pages and answers with links to its sources. Results are clearly marked as coming from the web. If you enable Strict Privacy Mode, web access is disabled completely. ## Get real files back Need a spreadsheet or report? Ask for a CSV, Excel workbook, or PDF. The file is created on your device and appears as a download in the chat. ## Pick where AI runs ### On your device This is the default. Use Chrome's built-in Gemini Nano, WebLLM running on your GPU, or a model on your own Ollama server. No setup. No API keys. No data leaves your device. The live egress meter shows exactly how many bytes have been sent. For local models, that's zero. ### Confidential Cloud Bring your own key and run frontier-class open models including Llama 3.3 70B, Qwen3-VL, Kimi K2.6, GLM, and Gemma inside a hardware-attested confidential computing environment. Before any prompt is sent, WebKaya verifies the environment inside your browser. It checks the chip signature, AMD certificate chain, signed code, enclave measurement, and GPU attestation. If verification fails, nothing is uploaded. You can watch the verification happen in the Attestation panel and download the full transcript. ### Your cloud model Use Anthropic or OpenAI with your own API key. Before a prompt leaves your browser, WebKaya can automatically mask names, IDs, card numbers, email addresses, and phone numbers. This is enabled by default. ## Image understanding If your chosen model supports vision, it reads receipts, scans, and images directly. If it doesn't, WebKaya uses its built-in offline OCR instead. Every piece of extracted evidence shows which engine produced it. ## More tools when you need them Optional features include: * Verified Data Analyst dashboards * Offline Python with Pyodide * DuckDB SQL for large files * Local RAG over your documents * A no-code agent builder * A page reader for the current tab Agents can be shared as portable .wkagent files. They include prompts, tool settings, and the shape of the data they expect, but never your files. ## Privacy built into the design WebKaya has no accounts, no telemetry, no analytics, and no backend that stores your chats. Everything stays in your browser unless you explicitly choose a cloud model or web search. The live egress meter shows every byte leaving your device. Local-only agents cannot make network requests, and Strict Privacy Mode blocks all non-local communication. Privacy policy: https://www.webkaya.com/privacy.html