The Components of Agent Memory
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Perplexity published a guide to agent memory, while Redis and PhoenixAI launched memory products the same week.
The race now centers on context infrastructure, not just the underlying models.
Background
AI agent memory is the mechanism an agent uses to retain, reuse, recall and refine information between sessions — what Oracle AI engineer Richmond Alake calls "the four Rs." As companies build agents that act over long tasks, several vendors are now selling memory and context as a standalone layer: Perplexity documented its Brain system and Agent API components, Redis expanded its Iris context engine, and PhoenixAI launched Contextbase. InfoQ separates the field into three parts — context (temporary information in the active window), memory (persistence beyond one interaction) and knowledge (underlying facts and rules).
How it unfolded
- Nov 2, 2026Perplexity published a technical guide to persistent agent memory, covering the Brain system in Perplexity Computer, the Agent API's memory-building components, and tests of whether stored context improves later work.
- Nov 4, 2026PhoenixAI's Contextbase, context infrastructure for AI agents, became available in private preview.
- Nov 5, 2026Redis announced the next chapter of Redis Iris, its real-time context engine for AI agents, with six capabilities including Redis Agent Memory and Redis Context Gateway.
- Nov 5, 2026PhoenixAI unveiled Contextbase at NEXT QUERY 2026 in Menlo Park, California.
Key terms
- AI agents
- A software entity that perceives its environment and acts autonomously to achieve goals, often improving through learning.
- private preview
- A testing stage of software development in which a limited group of users tries a product before its public release.
Source: Wikipedia
Who’s saying what
- Expert
- Oracle AI engineer Richmond Alake argues information does not equal intelligence: the model is access to information, while memory is the intelligence.
- Official
- PhoenixAI CEO James Li says an agent is only as good as its context, which today must be glued together from separate systems on every turn and comes out partial, inconsistent and often stale.
- Caution
- Perplexity notes its Agent API does not include a self-maintaining memory system, offering instead versioned procedures, conversation-context methods and tools for a developer-controlled store.