What it does
Cortex updates its memory as observations arrive. Every write lands first in a structure called Afferent. A structure called Thalamus opens sessions, retrieves what a turn needs and assembles the context for the answer. A structure called Executive writes one plan per turn that retrieved something. The na8ve agent is a barebone, open-source terminal agent with long-term memory that connects to Cortex. Install it with npm install -g github:na8ve/na8ve-agent, then start it in the folder you want it to work in. Type /login and choose na8ve. Your browser opens, you approve, and the agent finishes on its own. The agent keeps what matters across sessions and answers why questions from what it holds. It cites where each answer came from. It needs Node.js 22.19 or newer and git. The agent is in early access, with no official releases yet.
Capabilities
- Memory is kept at three levels: sensory, episodic and semantic. The abstractor folds many episodes into semantic conventions.
- A retracted memory cannot be retrieved by a search or by its own vector. Memories derived from it are flagged stale. na8ve reports this was verified 16/16.
- A live console shows Cortex as a brain, a graph, a live view and a signal trace. It also includes a chat view.
- Every write is recorded as a delta with the role that made it. Every derived memory keeps its sources as provenance. Every answer lists the memories it was built from.
- Each organisation gets its own learning model and memory space. A search in one organisation's memory cannot draw a candidate from another's.
When to use it
- You want answers to why questions drawn from stored facts, with citations to where each answer came from.
- You need to withdraw a stored memory so that it can no longer be retrieved.
- Your organisation is evaluating an AI system and needs auditability and governance.
