Google Cloud Releases Always-On Memory Agent, Replacing RAG and Embeddings with Continuous LLM Integration

Always-On Memory Agent is a reference implementation by Google Cloud in the generative-ai repository. It treats memory as a continuously running process, built on Google ADK and Gemini 3.1 Flash-Lite. Unlike traditional RAG and embedding methods, it does not use vector databases or embeddings. Instead, an orchestrator routes to three sub-agents: Ingest, Consolidate, and Query, which continuously read, connect, and write structured memory to SQLite. For developers building AI applications requiring long-term context, this eliminates the need for vector databases and reduces infrastructure complexity. Continuous LLM integration based on Gemini 3.1 Flash-Lite may lead to more efficient memory management but also incurs more frequent LLM calls and associated costs. Developers need to evaluate the trade-off between real-time performance and cost.