Screenshot to Toolset
Turn casual screenshots into page tools an agent can call
- Problem
- People who want an agent to work common websites — job hunting, applications, form filling — find that screenshot-plus-description is slow, token-hungry, and must be retaught every time. Sparse page-toolset projects exist, but the cost of writing tools and maintaining selectors falls on small teams and individuals who cannot sustain it.
- Solution
- The user simply screenshots a page or demonstrates a few clicks, and the tool identifies page elements to generate a reusable page-action script registered for the agent to call. Toolsets are shareable and versioned, and when a page changes the user gets a breakage alert plus a repair suggestion.
- Users
- Heavy agent-on-browser users, automation hobbyists, and small teams that need to batch-process internal admin systems
- MVP
- A record mode where a few clicks and a screenshot produce an action script; export to a common toolset format; three prebuilt site tools as examples; breakage alerts when scripts fail; Chromium only
- Revenue
- Free tier with 3 toolsets; subscription for unlimited toolsets and cloud sync; sell site-specific tool packs separately
- Why now
- Agent browser control is moving from demos into daily use, but page-tool production is still artisanal; record-and-replay plus selector generation is not hard, and an accumulating community toolset can become the moat