Kameleoon's AI centres on Prompt-Based Experimentation (PBX), now a five-agent lifecycle. Ideate scans a site and returns prioritised, evidence-based hypotheses, free on a Basic tier for all Web Experimentation customers, with a paid Premium tier for unlimited analyses. Build generates production-grade UI variations on the live site from natural language, with Master Prompts, Figma import and a browser extension. Configure sets targeting, goals and traffic allocation, including bandit methods, from prompts. Analyze surfaces business takeaways through the Takeaways menu. Ship pushes winners into production through the MCP server.
KAI, the in-app AI assistant, answers questions from documentation and experiment data, and acts directly on the account through the same tools.
The MCP server (https://mcp.kameleoon.com/mcp, OAuth) exposes 37 tools covering the full experiment and flag lifecycle (create, duplicate, start, pause, resume, stop, delete), goals, segments, targeting rules, results and raw variation code, including transpose_winning_experiment for one-call winner rollout. Non-developers can use it as a Claude custom connector, with documented Airtable and Notion export workflows.
AI Predictive Targeting scores each visitor's conversion probability for real-time activation, and AI-generated Takeaways consolidate insights across stopped experiments.