Google Cloud’s Always-On Memory Agent Replaces RAG and Embeddings With Continuous LLM Consolidation on Gemini 3.1 Flash-Lite

Most AI brokers neglect. They course of a request, reply it, then drop the context. Google Cloud’s generative-ai repository now ships a sample that tackles this directly. It is the Always-On Memory Agent, a reference implementation that treats reminiscence as a operating course of.

Always-On Memory Agent

Fundamentally, the mission is a light-weight background agent that by no means stops. It runs 24/7 as a steady course of, not a one-shot name. It is constructed with Google ADK (Agent Development Kit) and Gemini 3.1 Flash-Lite. Notably, it makes use of no vector database and no embeddings. Instead, an LLM reads, thinks, and writes structured reminiscence into SQLite. The mannequin selection targets low latency and low price for steady background work.

How It Works: Ingest, Consolidate, Query

Architecturally, an orchestrator routes each request to considered one of three specialist sub-agents. Each sub-agent owns its personal instruments for studying or writing the reminiscence retailer.

First, the IngestAgent handles incoming content material. It makes use of Gemini’s multimodal capabilities to extract a abstract, entities, subjects, and an significance rating. That structured file then lands within the recollections desk.

Next, the ConsolidateAgent runs on a timer, each half-hour by default. Like sleep cycles, it evaluations unconsolidated recollections and finds connections between them. Then it writes a synthesized abstract, one key perception, and these connections to the database. Consequently, the agent builds new understanding whereas idle, with no immediate.

Finally, the QueryAgent solutions questions. It reads all recollections and consolidation insights, then synthesizes a response. Importantly, it cites the reminiscence IDs it used as sources.