Inspiration

Most products lose users between signup and activation because onboarding is static, generic, and easy to ignore. We were especially motivated by feedback from Sideshift, who shared that onboarding friction is a recurring issue for their users. RunBook is our answer to that gap: help users actually complete setup and get value from software before confusion and set up looses them.

What it does

RunBook is an embeddable AI onboarding copilot for web apps. By reading through a web app creator's Github, it creates a code snippet that can easily be embedded in their website which answers user questions in context, recommends the next best action, highlights the exact UI elements users need, and guides them through step-by-step flows with skip/continue controls. We also added an auto-demo cursor mode that can visibly move, click, and walk through key actions.

How we built it

We built a script-based embed runtime (runbook-embed.js) plus an AI backend endpoint (/api/embed/chat). The host app shares page/app state with the widget, and the widget emits actions/events to drive tours and UI highlights. We validated adaptability by integrating it into two different demo products with different structures and onboarding journeys, ensuring a smooth experience no matter the application.

Challenges we ran into

The hardest part was reliability across different UIs: finding the right element every time, placing popups cleanly, and handling cases where pages hadn’t fully rendered yet. We also had to solve stale embed-script caching and make guidance concise so the assistant helps without overwhelming users.

Accomplishments that we're proud of

We shipped a working cross-site onboarding assistant that actively guides users through real product steps and whose purpose solves a real and all too often ignored problem. We are also proud of our robust highlight fallback logic, improved popup positioning, toggles for widget/hover behavior, and the full guided flow experience through two distinct demo apps.

What we learned

Great onboarding AI is a systems problem, not just a prompt problem. Model output alone isn’t enough, as you need strong event orchestration, UI reliability handling, and app-state awareness to make guidance actionable and trustworthy in real product environments.

What's next for RunBook

Next, we want deeper repo-aware guidance, better confidence/disambiguation logic, and activation analytics so teams can measure onboarding impact. We also plan to expand Studio controls so non-technical teams can configure flows, and harden auth/permissions for production enterprise rollout.

Share this project:

Updates