The September update focuses on the path from capturing a useful source to finding it again and sending it to the intended notebook. It includes web-app reliability changes, AI processing improvements and safer browser-agent handling. Extension 1.5.1 has been submitted for store review; submission is not the same as approval or availability.
What changed in the web app?
Public sharing now excludes deleted sources and requires the shared workspace or artifact to remain active. Moving something to trash should not leave its text visible through an otherwise active workspace share. Restoring content can make an existing share useful again, so review sharing settings when restoring material.
Metadata updates check merged size and account for storage growth. URL capture uses a database-backed daily request counter, while workspace and prompt counts exclude trashed items and restoration checks the applicable quota. These are reliability and enforcement improvements, not higher advertised allowances.
Smart Save: faster processing is a goal, not a blanket promise
The AI client now uses DeepSeek V4.1 Flash through the provider's deepseek-flash identifier, with non-thinking structured output for this filing task. Retry usage is accounted for cumulatively. This changes the provider used by Toolkit's AI enrichment; it does not replace the model inside Google's notebook product.
Continue to inspect suggested titles, summaries, tags and workspaces. A short processing time is not evidence that a summary is accurate. Actual response time varies with the source, network and service availability.
What is in the submitted extension update?
Version 1.5.1 improves notebook-tab targeting across both supported Google notebook addresses, handles an already-loaded tab more reliably and adds receiver recovery. It also gives sequential source exports more time to finish rather than treating one minute as a universal limit.
After the update becomes available, use a harmless public page to check capture, library save and notebook delivery separately. Verify your installed version before assuming you have the new behavior. Keep the destination notebook open and inspect the final source.
Browser agents still need boundaries
The existing WebMCP integration has stronger handling for asynchronous registration and cancellation. Consequential operations retain human confirmation, and normal authentication and plan limits apply. This is an experimental browser integration, not an unrestricted remote MCP server. See the agent guide for availability and privacy boundaries.
What has not changed?
Local export of supported text remains free. Cloud storage and AI actions still have plan limits; Pro does not mean unlimited storage or unlimited AI. Capturing a source and sending it to Google remain separate stages.
This update does not announce shared team permissions, automatic source refresh, a procurement approval system or guaranteed performance gains. Read current plan facts, data handling and support guidance. Report issues with the installed version and a redacted error message, not private research content.