I open-sourced DomA: an AI automation agent for the browser

说明与用途

DomA 是一个 Chrome/Edge 扩展，提供侧边栏 AI 代理，保留用户会话和扩展。它使用截图覆盖、Jev 内部循环减少 LLM 调用，并通过可重用技能提升真实网站的自动化准确性。

问题

Browser automation agents often fail on real sites due to fragile selectors, slow LLM round‑trips, and lack of reusable procedures.

结果

DomA reduces LLM round‑trips, improves accuracy on real sites, and enables reusable automation skills while preserving user session data.

工作流

• 1. User opens DomA side panel in Chrome/Edge
• 2. Agent captures a screenshot with numbered overlays (Set‑of‑Mark grounding)
• 3. Agent selects index‑based actions (click, type, navigate)
• 4. Optional Jev inner loop executes multiple grounded steps locally
• 5. Agent verifies progress and continues until goal achieved or hands back to main LLM
• 6. Successful procedures are saved as Agent Skills for future reuse

证据

DomA is a Chrome/Edge extension, not a standalone AI browser and not a hosted “cloud browser.”

来源

The real pain isn’t “can it click?” — it’s accuracy

来源

Jev is DomA’s optional inner loop on browser_screenshot (act/verify)

来源