Google PM open-sources Always On Memory Agent, ditching vector databases (venturebeat.com)

Google's senior AI product manager, Shubham Saboo, has made a significant move in the AI landscape by open-sourcing the "Always On Memory Agent" on GitHub, designed to tackle the challenge of persistent memory in agent systems without relying on traditional vector databases. Built using Google's Agent Development Kit (ADK) and the cost-efficient Gemini 3.1 Flash-Lite model, this initiative highlights a shift towards more autonomous systems that continuously ingest, consolidate, and retrieve information. This agent offers a practical implementation that aims to simplify the architecture of agent systems, particularly appealing to developers looking to reduce operational complexity and costs. The implications are notable for the AI/ML community, as the agent operates with a simplified structure that eliminates the need for embedding pipelines and vector storage, instead leveraging a continuous memory approach through SQLite for structured data. However, this release raises governance questions about how memory is managed, including data retention and auditability, which are critical for enterprise adoption. The response from the industry underscores that while the technical capability to maintain persistent memory is advancing, ensuring that these systems operate within safe, inspectable boundaries is equally vital for widespread implementation in the enterprise context.