Episodic memory
Episodic memory is the layer of an AI agent that records what was said and done across conversations, with timestamps. Each entry is a discrete observation: 'user said X at time T,' 'tool call Y fired at time T.' Queries against episodic memory are temporal.
Why it matters
A long-running agent that does not record episodic events at all forgets your conversation when the chat ends. A naive agent that loads every prior message into the prompt fails on context limits within days. The right structure is a separate episodic store, queryable temporally and semantically, that serves the prompt with relevant slices on each turn.
Episodic memory answers questions like "what did I tell you about my dog last week," "when did I last book Felipe's," "what was Sara's main objection in our last meeting." It is the source layer that the rest of memory consolidates from.
How Ajeris handles this
Ajeris episodic memory is a per-user timeline with semantic search for retrieval. Each conversation turn writes one or more entries with the channel, timestamp, and content. The retrieval layer at runtime queries by time window plus topic and feeds the relevant slice into the prompt. Daily dreaming consolidates important episodic notes into semantic memory.
