The supplied GitHub repository describes Univer Workspace as open source and says that people and AI agents can create, collaborate, and review together. It lists support across spreadsheets, documents, presentations, relational tables, and canvases. The repository also shows 19 branches, 16 tags, and a latest commit dated September 25, 2026; recent entries include fixes for agent plugins, data upgrades, previews, and comparison-view synchronization.
The mechanism is a workflow boundary: browser editing, an agent interface, and a CLI point at the same office-oriented workspace, while the project’s supplied description places agents in isolated worktrees and has changes checked before merging. That separates generating a proposed edit from accepting it into a shared artifact. It makes review, comparison, and permissions part of the product surface rather than afterthoughts around a chat response.
Why now is that the repository shows active work on the operational seams of this model—agent plugins, data upgrades, previews, and synchronized comparison views. The implication is that document automation may increasingly be judged by whether agents can produce reviewable, attributable changes across connected business artifacts, not merely by the fluency of generated prose. A second implication is that teams adopting these systems will need governance for merge authority and access rights alongside model selection.
Evidence boundary: the supplied public source is a project repository and supports claims about the project’s stated design and visible development activity; it does not prove production reliability, security posture, adoption, or the quality of agent outcomes. Watch whether the project documents a stable permission model, auditable review history, and repeatable deployment guidance for multi-user environments. This begins a distinct analytical record because no prior New Runtime record with usable editorial content was supplied, despite related-item identifiers in coverage.