Agentkit lets you build custom AI chatbots trained on your own content and embed them on any website — no code required. Whether you run a SaaS, an online store, a docs site, or a service business, Agentkit turns your knowledge into a chatbot that answers customer questions 24/7. Train it on what you already have. Point Agentkit at your website, upload PDFs, or paste in text. Your chatbot learns from it automatically and answers based on your actual content — not made-up information. Re-crawl on a schedule so it stays in sync as your site changes. Embed it anywhere in one line. Drop a single script tag onto any website — WordPress, Shopify, Webflow, or custom HTML — and your chatbot appears as a clean, customizable widget. Match it to your brand with custom colors, avatars, welcome messages, and suggested questions. The widget is lightweight and mobile-friendly. Built for teams and agencies. Agentkit supports both personal and team workspaces, so agencies can manage chatbots for multiple clients from one dashboard. Invite teammates, assign roles, and work together on training, prompts, and conversation review. Each workspace's data stays fully separated. Test, tune, and review. Use the built-in playground to test your chatbot before going live. Adjust the system prompt, swap models, and see exactly which sources it used for every answer. Review real conversations, flag bad replies, and keep improving over time. Track usage, popular questions, and leads from the dashboard. Pick your AI model. Agentkit runs on OpenRouter, giving you access to the latest models from OpenAI, Anthropic, Google, Meta, and more. Choose the right balance of cost, speed, and quality for your use case — and switch anytime without retraining. Capture leads automatically. Collect emails and qualify leads right inside the chat. Push them straight to your CRM, helpdesk, or email tool with webhooks and built-in integrations. Secure and private. Built on Supabase and Postgres with row-level security and encrypted storage. Your data stays yours and is never used to train outside models.