How Google Just Gave AI a Real Memory (and Why Your SME Can’t Ignore It)

You know the frustration. You chat with an AI assistant, it nails the first task, but the moment you switch topics, it has completely forgotten your name. It feels like talking to a goldfish. Your business context—customer history, internal processes, past decisions—vanishes the second the conversation ends. Google just dropped a solution that changes this completely.

What Happened

Google Cloud open-sourced what they call the “Always-On Memory Agent”. Instead of processing a request and throwing away the context, this agent runs like a background process in your business—24 hours a day, 7 days a week. It doesn’t just retrieve data; it actively digests it.

The architecture is refreshingly simple. It has three specialist sub-agents. The IngestAgent takes everything you give it—PDFs, images, audio, video, even meeting recordings. The ConsolidateAgent runs a “sleep cycle” every 30 minutes, linking related memories together and writing new insights without being prompted. The QueryAgent then answers questions using this living, breathing memory.

The agent builds new understanding while idle, with no prompt.

Critically, it does all this without a vector database or embeddings. It uses