Sim groups computer actions and checks cursor placement before typing
Agent-workflow platform Sim has added support for 2–8 computer actions in one call. Its update also waits for confirmation that the cursor is at the end of a field before typing after a focus click. Changes to the desktop browser prevent text entry behind certain full-screen overlays, while the model receives a more compact view of application controls.
Artificial Intelligence··Morning
One call can contain several computer actions
Sim, a platform for building AI agent workflows, has added grouped computer-use actions in version 0.9.57. Its release document specifies 2–8 actions per call and reading application state in one round trip per element. The changes concern how software agents interact with computer interfaces while carrying out a task.[1]
Typing waits for a confirmed cursor position
When an agent clicks to focus a field, the system verifies that the cursor sits at the text’s end before it starts typing. An editor that cannot receive focus through accessibility controls can instead be selected by clicking. The release also addresses edge cases in grouped actions and click-to-focus behaviour.[1]
Full-screen overlays can block text entry
The desktop browser refuses to type behind an unmarked full-screen modal window or if all the sampled points miss their target. Sim also folds empty layout groups and compacts the accessibility tree presented to the model, the structure describing application controls. The strip showing activity no longer sits above the chat entry field. Agent-supplied tool arguments remapped from dedicated fields are also preserved. Google Drive operations keep the pagination token supplied by the agent.[1]