Core concepts

Neuronz.ai stores a few kinds of records — facts, knowledge documents, past work, topics, rules and a few more — links them together, and points your agent at the ones that match what you are doing. Here's what each kind is for.

Words you'll meet

• Harness — the coding agent you use: Claude Code, the Claude desktop app, or Oh-My-Pi.
• Profile — the scope your memory lives in, usually one per project.
• Record — one stored item: a fact, a knowledge document, an action, a topic, a rule.
• Fact — one short statement worth remembering.
• Knowledge — a longer reference document, kept whole.
• Action — a record of what happened: work the agent did, or a discussion with you.
• Topic — a recurring issue or piece of work, tracked over time.
• Rule — an instruction your agents must follow, approved by you.
• Voice — how your writing should read in a given situation.
• Persona — what is known about you, or a colleague, as a person.

Profiles

A profile is the scope your memory lives in, usually one per project, so a client repo's context never shows up in your side project. The agent picks the profile from the directory it runs in. Many directories can share one profile; a directory that matches none gets its own. Almost everything below belongs to one profile. See profiles.

Facts

The core of your memory: one short fact per record, never a wall of pasted text. A fact stands on its own or is derived from a knowledge doc, and it has a status — verified, unverified, or a plan.

• History is kept — correcting a fact keeps the old version in history, and the new one becomes the current fact.
• No duplicates — a new fact is checked against what's already stored, so the same thing isn't saved twice.

A fact can also be about a person — you, or a colleague you work with. That is what powers personas: what is known about you arrives at session start, and a colleague's profile is pointed at when a prompt is about them.

Knowledge

Longer reference material — runbooks, docs, infrastructure notes. A knowledge document keeps its full text, split into sections so recall can point at the passage that matters rather than the whole file. Facts taken from a document link back to it, so a short fact is found first and the full document is one step away when you need more. See knowledge docs.

Actions

An action records what happened: a fix, a deploy, a pull request, a run of one of your commands, or a discussion with you. The plugin records a session's work automatically, and the agent can log an action itself with a summary of what was done or discussed, what was weighed and why, and what was left open. Actions are how a new session finds out what earlier sessions already did and picks up a thread where it was left.

Topics

A topic tracks a recurring issue or a piece of work over time, so the same problem adds to one topic instead of creating duplicates. A task topic has a priority and a status and appears on the dashboard's kanban board. Recall does not return topics; the agent may be shown related topics as links it can open. See topics.

Rules

Rules are the instructions your agents must follow ("always do X"). An agent can propose a rule, but only you can approve it; nothing becomes a rule on its own.

Each rule carries conditions saying when it applies. Some Neuronz.ai checks itself (the tool about to run, the file being edited); others only the agent in the session can judge ("when the command touches production"). A rule whose conditions don't hold is left out.

What reaches the agent is the rule's name, with the call that fetches its text — never the text itself. That happens on every prompt, and in Claude Code also right before a tool call a rule applies to. See rules.

Recall

How the right context reaches the agent. On every prompt, Neuronz.ai searches your facts, actions and knowledge against what you just asked. What arrives is a pointer: how many records matched, and the call to read them. The agent then reads the records it needs. Your voice arrives the same way. See auto-recall.

• Keyword and meaning — both kinds of search run together, so recall finds the right record even when the wording doesn't match.
• Reranking — results are re-ordered by how well they fit the prompt.
• Linked records — when the agent reads its matches, recall also follows one step along the graph to pick up closely related records.

The graph

Records are linked to each other: a fact derived from a knowledge doc, an action that fixed a topic, and so on. Recall uses these links to reach the longer document behind a short fact only when it is needed. See the knowledge graph.

More

• Voice — stores how your writing should read in different situations and applies it to drafts.
• Skills & assets — skills, commands and agents stored in Neuronz.ai and placed into your projects at session start.

Next

• All features — a page for every capability, in depth.
• MCP tools — the tools that read and write all of this.