Memanto tutorial: memory recall, RAG and temporal queries

This tutorial covers memory retrieval in Memanto, which the presenter calls Memento. It explains how an AI agent can request stored memories through commands, HTTP endpoints or the UI Playground.

The presenter describes recall as semantic search rather than keyword matching. Results include a title, content preview, memory type, confidence score, creation time and similarity score. Examples narrow results by type or tags, then combine a decision type, a backend tag and a limit of five.

The answer command uses retrieved memories as context for an LLM response. A limit controls how many context chunks it uses; one example selects the top ten. The speaker says this reduces the main agent's work and token use, but provides no measured comparison.

Temporal retrieval serves two purposes. The --as-of flag requests memory as it existed at a past date, such as May 15, 2026, or a relative time such as yesterday. The --changed-since flag requests changes since a chosen point. These examples explain how to inspect the information available when an agent made a decision.

The HTTP walkthrough starts a server and sends a recall request with an agent name and session token. It also covers temporal recall and an answer endpoint that returns an answer with its source. The tutorial closes with recall and RAG examples in the UI Playground.