NOUZStructured Memory for Obsidian and AI Agents
NOUZ helps agents find the right notes, links, and context in an Obsidian vault.
NOUZ helps agents find the right notes, links, and context in an Obsidian vault.
pip install nouz-mcpNOUZ turns an Obsidian vault into a local map of knowledge. Instead of a folder of files, the agent gets notes, links, levels, and nearby context. The structure stays under your control, while NOUZ helps keep the base organized and ready for precise agent work.
As a knowledge base grows, agents have a harder time choosing the right context: they read too many files, miss important links, mix drafts with conclusions, and rebuild material from scratch. NOUZ gives the agent precise actions over the base: read a note, inspect its place in the graph, follow links, suggest metadata, and find related material.
It does not replace thinking. It removes blind file wandering and leaves decisions where they belong: with the person using the system.
The agent can request a specific part of the base: one note, its parents, children, nearby links, or candidate connections. Work shifts from repeated blind reading to a reusable local index.
NOUZ surfaces possible metadata, links, duplicates, and drift signals. Nothing becomes part of the vault unless it is confirmed.
The semantic layer helps reveal intersections between notes, even when they use different words or live in different branches.
Through MCP, the agent gets practical tools: read a note, inspect its position, find neighbors, and suggest structure.
Start simple: Obsidian, YAML, and graph structure. Add semantics later, when the vault grows beyond manual links.
Obsidian vault structure: YAML metadata, parents, children, and note formulas. Works without embeddings or a complex config.
Adds classification, related ideas, bridges between themes, core_mix, and drift signals. Useful when the vault outgrows manual linking.
A strict five-level structure: Core → Pattern → Module → Quant → Artifact. Useful for research bases and team knowledge where order matters.
NOUZ helps expose structure, find disconnected notes, spot weak areas, and give the agent context without manually opening dozens of files.
Material from different domains can point to the same pattern in different language. NOUZ helps turn those intersections into a working research system.
The agent gets more than file access: it can see where a note sits, what it connects to, which material is nearby, and what to inspect next.
The easiest way to try NOUZ is with Obsidian. If your knowledge base already lives in Notion, Confluence, Google Docs, or GitHub, you do not have to migrate it: the same approach can be connected through MCP connectors or custom adapters.
Semiotronika can help audit and set up agent-ready knowledge bases: duplicates, outdated pages, weak links, and disconnected sections. The workflow is explicit: the agent reads the base and suggests changes; a person reviews and applies them.
A map of the current base: where structure has weakened, which topics are outdated, what duplicates exist, and where agents lack links.
Obsidian, NOUZ, metadata rules, a local index, and a workflow where humans confirm important changes before they land.
For team systems, NOUZ can act as a semantic layer beside the existing base without breaking Notion, Confluence, or another familiar tool.
The graph is built top-down. Each level has its role, and larger bases can add meta_root: a level-0 anchor that gathers domains into one system and stays out of semantic calculations.
| Level | Type | What Goes Here | Sign Source |
|---|---|---|---|
| L1 | Core | Knowledge domain — a broad area | Manual or etalon-based |
| L2 | Pattern | Thematic frame for child modules and quants | Manual or etalon-based |
| L3 | Module | Functional group, project | Manual or etalon-based |
| L4 | Quant | One idea — a guide, research, post | Child artifact type + content domain |
| L5 | Artifact | Raw material: logs, chats, configs | Content-structure heuristic |
pip install nouz-mcpStart with Obsidian: point NOUZ to your vault and connect it to an MCP client. LUCA works without embeddings; PRIZMA and SLOI need config.yaml and an embedding endpoint.
Example semantic mode:
mode: prizmaEnvironment variables:
export OBSIDIAN_ROOT=/path/to/vault
export EMBED_API_URL=http://127.0.0.1:1234/v1Run via MCP client:
{
"mcpServers": {
"nouz": {
"command": "python",
"args": ["-m", "nouz_mcp"],
"env": {
"OBSIDIAN_ROOT": "/path/to/vault"
}
}
}
}NOUZ is a practical MCP server; the theory is optional. For readers interested in the research frame behind the project: Recursive Self-Organization as a Universal Principle.