Laworigin/DoWhat

DoWhat is an early-stage prototype (1 star, 18 days old) that implements the 'AI Screen Observer' pattern. While the privacy-first, local-first approach is a valid niche, the project currently lacks a technical moat or significant community traction. It builds on OpenClaw, suggesting it is more of an assembly of existing frameworks than a novel engine. Competitive Landscape: The project faces immense pressure from both well-funded startups like Limitless (formerly Rewind) and OpenAdapt, as well as OS-level integrations. Microsoft Recall (despite its PR hurdles) and Apple Intelligence are moving directly into this 'screen-aware' agent space. Defensibility: The score is low (2) because the core loop—periodic screenshots fed into a VLM for intent extraction—has become a standard design pattern in the 'Computer Use' agent category. Without a massive dataset of user interactions or a proprietary high-speed local vision model, it is easily reproducible. Frontier Risk: High. OpenAI (via 'Operator' and desktop apps) and Anthropic (via 'Computer Use') are aggressively optimizing models for exactly this type of interface interaction. As these labs provide more efficient 'vision-to-action' tokens, the value of thin wrapper applications like DoWhat diminishes unless they offer deep vertical integration into specific work tools (e.g., Jira, Slack, GitHub).