What problem does it solve?
Local AI workflows often lose context across interactions because session data is not reliably stored outside memory. This protocol persists session metadata locally to provide human-readable titles and consistent turn metadata for the Read Aloud extension.
Core Features & Use Cases
- Local storage architecture under the Antigravity path: ~/.gemini/antigravity/read_aloud/<sessionId>/ to house per-session state and snippet files.
- Metadata schema: extension_state.json containing session_title, current_turn_index, and last_updated to drive UI resolution and continuity.
- UI resilience and MCP compatibility: UI resolution reads extension_state.json to derive display names; MCP injections update the state atomically and reflect new titles in the UI.
Quick Start
Install the agent and configure it to store session metadata in extension_state.json under ~/.gemini/antigravity/read_aloud/<sessionId>/ and write new snippet files to read_aloud as MCP outputs.