KivoAI 作成者: Wiktor Olf

KivoAI is a privacy-first autonomous browser assistant that navigates web pages, answers questions, and fills complex forms without exposing your personal information to the AI model.

Unlike conventional AI tools that transmit your raw screen and identity to third-party clouds, KivoAI enforces a strict local privacy boundary directly inside your browser. Sensitive inputs, passwords, and credentials are automatically masked on an internal canvas before any visual data is processed, keeping your personal details secure.

Key Features:

Unified Multimodal Agent: Ask a question or assign a goal in a single prompt box. KivoAI intelligently decides whether to answer with a structured explanation of the page or to autonomously plan and execute a sequence of clicks, typing, and selections.

Hands-Free Voice Dictation: Speak your instructions naturally. Built-in Voice Activity Detection (VAD) automatically detects when you finish speaking, transcribes your audio, and triggers the agent without requiring a single click.

Zero-Exposure Autofill: Save multiple isolated profiles (Personal, Work, Family). KivoAI prompts the model with symbolic placeholders (such as {{username}} and {{fullName}}). Real data is resolved and injected locally at the exact moment of execution, so the AI never sees your identity.

Visual PII Redaction: Passwords and sensitive form fields are detected and blacked out on an off-screen canvas prior to image downscaling and compression.

Local & Self-Hosted: Designed to pair seamlessly with your self-hosted Open WebUI, Ollama, or llama.cpp setup. Zero telemetry, zero external trackers, and zero dependencies.

Take control of your web automation with fast, local intelligence that respects your privacy.