Learn Mode

1. Start a conversation— open AI Chat and ask about a topic you want to study. Learn mode activates automatically when it detects study-related intent (e.g. "teach me about", "study materials for", "flashcards on").

2. Choose your depth — the AI asks what level of detail you want before building anything.

3. Watch it build — a dedicated side panel opens showing real-time progress as four artifacts are created in parallel.

4. Explore & extend — click any artifact card to open it. After completion, interactive coding suggestions let you create timelines, quiz games, and concept maps.

When a learn session starts, a side panel slides open showing step-by-step progress. Each step has an icon, label, elapsed timer, and animated indicator.

If some artifacts fail, the panel shows a "partial results" banner — the successfully created materials are still available.

After all artifacts are ready, the panel offers three interactive coding ideas tailored to your topic:

• Interactive Timeline — a scrollable timeline of key events with hover details
• Quiz Game — a gamified quiz with scoring, streaks, and animations
• Concept Explorer — a clickable concept map showing how ideas connect

Click any suggestion to have the AI build it as a live code artifact using Theo's code generation tools.

Learn-mode study videos render through Theo's video engine, which refuses any prompt that names a real or historical person (even public-domain figures like Friedrich Nietzsche or Marie Curie). Instead of failing the whole render, Theo now plans every shot using archetypes — era, role, attire, and distinctive features — so the video still teaches the subject without triggering the safety filter.

Three guard rails keep your render moving: the shot planner is steered toward archetypal language, a sanitizer pass rewrites any leftover names before the clips dispatch, and individual scenes that still trip the filter get one automatic rewrite + retry. If any single clip still can't render, Theo ships the remaining scenes as a shorter video with a soft heads-up instead of failing the whole job.

Learn sessions are tied to the conversation. Once the build kicks off, the pipeline runs on our durable execution layer — every phase (research, flashcards, quiz, study notes, presentation, podcast, video) is checkpointed independently.

Refresh the page, switch tabs, or close the browser? The build keeps going. When you come back, the learning panel reattaches to the same job and picks up right where the progress indicator left off — no need to start over. The in-thread Learn card also stays anchored to the conversation so you can reopen the panel at any time.

If a step ever stalls (a transient outage on a downstream provider, for example), an automated watchdog re-dispatches the job within a few minutes so completed steps stay cached and only the failed step has to retry.