Google Always-On Memory Agent (github.com)

Google has unveiled an innovative always-on AI memory agent powered by the Agent Development Kit (ADK) and Gemini 3.1 Flash-Lite. Unlike traditional AI systems that forget information after processing, this agent operates continuously in the background, actively consolidating and connecting information much like the human brain. It eliminates the need for a vector database or embeddings, instead relying solely on a large language model (LLM) to read, think, and write structured memories. The agent can process a wide variety of file types—text, images, audio, video, and PDFs—through its multimodal IngestAgent, and is designed for easy integration with three ingestion methods: automatic file watching, dashboard uploads, and an HTTP API. This development is significant for the AI/ML community as it addresses limitations of current memory approaches, like passive retrieval or loss of detail over time. By actively consolidating information every 30 minutes, the ConsolidateAgent generates insights and connections across stored memories, enhancing the agent's ability to provide informed responses to queries. This innovation not only improves the functionality and reliability of AI agents but also demonstrates a shift toward dynamic memory systems that more closely mimic human cognitive processes, potentially setting a new standard for AI memory architecture.