An Artifact is a structured deliverable created by the agent to accomplish its task and communicate its progress and thinking to the human user. Artifacts include rich markdown plans (Implementation Plans), code diffs, architecture diagrams, images, and browser recordings.

As agents become more autonomous and execute complex tasks over longer periods, Artifacts enable asynchronous collaboration. You do not need to carefully monitor every individual tool call or step synchronously; instead, you review high-level deliverables at key milestones.

Reviewing Artifacts Across Surfaceslink

Artifacts are primarily generated during the agent's Planning Mode and are accessible across both Antigravity 2.0 and the Antigravity CLI.

Antigravity 2.0link

The desktop app features a visual sidebar and review pane specifically optimized for displaying, organizing, and managing rich Artifacts.

Capabilities: You can inspect interactive plans, review visual code diffs, and play back browser recordings of the agent's UI actions directly within the app interface.

Antigravity CLIlink

In the lightweight terminal interface, Artifacts are managed using a fast, keyboard-driven review panel.

Workflow: When the agent generates or modifies files that require your approval, a notification appears in your terminal status bar.

Interactive Steering and Feedbacklink

Feedback is a core mechanism of the Artifact workflow. Depending on your configuration, the agent will pause at intermediate milestones and request review on its plans or code edits before executing them.

Steering the Agent: If an artifact (like an Implementation Plan) does not align with your goal, you can provide inline text feedback to steer the agent's thinking in the proper direction before it modifies any local files.
Granular Control: This approval loop ensures that you remain in the driver's seat, allowing the agent to operate with high autonomy while maintaining strict human-in-the-loop validation.