You set the goal. It moves forward, and you can see it.
Chinese or the language you choose; stage descriptions appear in the conversation. Tool details and task files expand on demand, and final results are presented separately.
You can add requirements or ask questions while the task is running
Three modes: Chat, Plan, Agent
Five thinking levels, mapped to the model's actual capabilities
Context checkpoints and token-saving mode enabled by default
Actual app screenshot · click to enlarge
02
Tools, organized around results.
Terminal & Code
Run commands, edit Python / HTML / JavaScript, preview files, and view real exit statuses.
Browser
Built-in standalone browser with Bing as the default search entry; can open URLs, read pages and perform DOM operations.
Skills and scheduled tasks
Enable official Skills, import your own Skill, configure scheduled or interval tasks; the device and app must remain running.
Image and site
Connect supported image generation APIs; configure SSH deployment sites. Text models will not impersonate image models.
03
Real computer operations.
Native screenshots, accessibility reading, mouse and keyboard operations; the actual operation screen is shown in the upper right corner. Permissions are managed by macOS; clicking stop clears the input state.
Different models support different tools, vision, thinking parameters, and output quotas. BE AI shows actual results and does not equate a single successful request with a completed task.