Memanto setup: persistent agent memory and Codex integration

This tutorial sets up Memanto, described in the metadata as an open source memory layer for AI agents and coding assistants. Installation starts with pip install memanto, followed by configuration with a Moorcheh API key. The local CLI and REST interface use that backend; the tutorial does not establish an offline deployment.

The walkthrough creates an agent and stores facts, preferences, and decisions with memory types, tags, and confidence values. In the demonstrated setup, creating an agent activates it and starts a six-hour session. The speaker distinguishes recall, which returns memory chunks, from answer, which generates a response grounded in retrieved memories. Examples filter decisions and limit retrieval to two results.

Document ingestion covers PDF, Word, CSV, and Markdown. An uploaded brand guide supplies answers with a named source. The tutorial then connects Memanto to Codex. The speaker says this lets the coding assistant retain preferences across interactions and slows context-window growth, but provides no measured comparison.

The dashboard shows memory sources, analytics, and conflict handling, with controls for the answer model and retrieval settings. Finally, memanto serve exposes a local REST API so applications and agent frameworks can call the memory layer over HTTP.