Enterprise AI agents deployed in real-world settings face a persistent problem: they forget what they learned and regress to earlier mistakes. Retrieval-augmented generation (RAG), the dominant architecture for grounding AI agents in factual information, only solves half the problem by surfacing relevant documents. It offers no mechanism for agents to retain validated decisions or build upon previous discoveries.
Rippletide, a startup operating within the Neo4j ecosystem, addresses this gap with a decision context graph framework. The system adds structured memory, time-aware reasoning, and explicit decision logic to enterprise agents. The critical innovation centers on "non-regressivity," a term co-founder and chief scientific officer Yann Bilien coined to describe agents that freeze validated action sequences and compound improvements over time.
This distinction matters enormously in production environments. An AI agent handling customer service might correctly resolve a billing issue on Monday, then revert to an incorrect approach on Tuesday because it has no persistent memory of validated solutions. RAG architectures retrieve documents but lack mechanisms to track which agent decisions succeeded and which failed. They treat each query as independent, ignoring the agent's own experiential learning.
Decision context graphs solve this by storing three layers of information: the decisions the agent made, the outcomes it achieved, and the temporal sequence connecting them. When an agent encounters a new problem, it can reference its own decision history and build on solutions that worked before. This prevents regression while enabling compound learning.
The architecture gains particular relevance as enterprises deploy agents for high-stakes processes like financial compliance, supply chain optimization, and healthcare workflows. In these domains, reverting to previously disproven approaches creates operational risk and erodes user trust. Non-regressivity becomes a compliance necessity, not a feature.
Rippletide's approach sits at the intersection of graph databases, agent design, and business process automation. By treating an agent's