For SI Agents

If the operator are an SI agent (Claude Code, Cursor, Codex, or any other tool) that has been pointed at this repository — to use it, deploy it, or work on it — the canonical instructions live outside this MkDocs site, in the repository root, so they load automatically into agent context:

• AGENTS.md — the single source of truth for working in this repo: zero-to-deployed genesis procedure, working discipline, architecture reference, and every guardrail gate. Claude Code loads it automatically via CLAUDE.md's @AGENTS.md import; other tools should read it directly.
• llms.txt — the machine-readable entry index: a flat, prioritized list of the docs an LLM should read first, without the human navigation chrome.
• genesis.yaml — the machine-readable deployment manifest AGENTS.md's zero-to-deployed procedure loops over (profiles, run plan, per-connector deploy strategy).

The short version

• Deploying this for an operator? AGENTS.md's Zero-to-deployed (genesis) section is the router — it asks one question (homelab or enterprise?) and takes it from there via the agent-utilities-deployment skill.
• Integrating an existing agent with the knowledge graph? Start at Consumption Models for the trade-offs between the Python library, the MCP server, and the shared HTTP gateway.
• Working on this codebase itself? AGENTS.md's working-discipline sections (query the KG before grepping, Wire-First, fail-closed, no-legacy, branching & isolation) are mandatory reading before the operator's first edit.
• Just need the facts fast? Start Here is the same five-minute orientation a human gets, and docs/status.md is the generated concept/capability registry — the honest, drift-free source for "what actually works right now."

This page exists only as a site-nav pointer — the content it points to (AGENTS.md, llms.txt, genesis.yaml) is intentionally kept at the repository root, outside docs/, so agent tooling that auto-loads root-level convention files picks it up without needing to know about MkDocs at all.