AI Planning from Docs

Stop writing demo scripts by hand

Most demo video workflows start the same way: someone opens a doc, manually writes a shot list, creates a script, sends it for review, and iterates. By the time the script is approved, the feature has already changed.

AI Screen Recorder’s planning stage eliminates that loop entirely.

How it works

You submit a job with your app URL and one or more documentation URLs — your help center, your README, your changelog. The planner sends those URLs to our Firecrawl-powered scraper, which reads the full content and passes it to an LLM with long-context understanding.

The model reads your docs the same way a solutions engineer would: understanding which features matter most, what each one accomplishes, and what a viewer needs to see to understand it.

The output is a structured plan: a prioritized list of features, each with a title, a one-sentence intent, and explicit success criteria — what must be visible in the recording for that chunk to be considered complete.

What you get

• Automatic feature discovery — no need to enumerate features manually
• Intent-driven demos — each chunk has a clear purpose, not just a list of clicks
• Success criteria — the recording agent knows when it’s done
• Priority ordering — highest-value features recorded first
• Configurable scope — add a brief to focus (“Cover invoicing, payments, and reporting — skip admin settings”)
• Max chunk cap — set a limit to control cost and video length

Built for real documentation

The planner handles a wide range of doc formats: markdown READMEs, help center articles, inline code docs, changelogs, and structured HTML. If your users can read it, our planner can use it.

Model routing

The planning stage uses a long-context LLM optimized for comprehension and cost efficiency, with Claude Sonnet as the fallback for complex or ambiguous documentation sets. Per-client model preferences can override the default routing.