Skip to content

master5d/NAUTILUS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

536 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NAUTILUS — Sovereign Personal Knowledge Mesh & AI Pilot

Nautilus is a local-first autonomous environment for personal knowledge management, multi-dimensional semantic indexing, and secure AI agent orchestration. It is built upon the SOVRN Architecture blueprints—a set of sovereignty-first, zero-telemetry, and highly resilient technical principles.

Nautilus Concept Flowchart

🏗 Monorepo Architecture

Nautilus is divided into modular directory boundaries:

  • /apps/knowledge-graph (Nooscope UI): A WebGL-powered 3D force-directed node visualization dashboard (Next.js 16 + React + Three.js). Features custom GraphRAG searches and supports a robust offline fallback (walkVault) sweeping local Obsidian directories when database servers are offline.
  • /core/enerv (ENERV Indexer): A fast, schema-first environmental compass (Python CLI). Validates project directories against strict JSON contracts (.facets/meta.json) and powers the local facet ingest pipeline.
  • /core/desops (DesOps Hub): Centralized design governance — pipeline v4 «hub-and-spokes» (doctrine + agent-layer skills + semgrep token gates on 15 lab projects + nightly audit feeding the labwatch panel) extended by v5 machine-readable contracts: per-project design/identity.json (style descriptor + variance/motion/density dials + palette; test-identity.ps1) and diagram governance (lint-diagram.ps1 structural lint for svg/drawio/archify-html; «derived, not drawn» doctrine with desops:derived provenance markers). Manages GLOBAL_DESIGN.md/tokens.json (master tokens) and the tools-registry.
  • /core/vending (Code Vending Machine): Reuse-first registry of tested, hash-pinned code artifacts (CATALOG.md + registry/). Agents vend.py search before generating, vend.py add after; includes the SOVRN §7/§8 supply-chain format gate (strict pre-store + secops/hooks/pre-commit).
  • /core/gte (Gummiberry Time Engine): De-poisoned time estimation for agent-written plans — T_model/T_glue estimates in hours:minutes instead of trained-in human weeks, a plan linter (pre-commit, blocking) and an estimate-vs-actual calibration ledger. Protocol: skills/gummiberry_time.md.
  • /config (Shared Service Configs): Houses LiteLLM provider configurations and holds the dynamic service registry updated by the Port Broker.
  • /scripts (Lifecycle & Port Broker Automation): Prevents port collisions on local Windows hosts via socket availability sweeps, automatically synchronizing dynamic ports across envs (.env, .env.local, .hermes/config.yaml).
  • /hermes (AI Pilot Gateway): The central orchestration context. Listens to Telegram bot directives, manages cumulative dry-runs, and dispatches tasks to specialist coding agents (Claude Code, Codex, Hermes — see config/orchestrator.json, switch via Labwatch :4002 or swarm -Switch).
  • /skills (Local Skill Docs): Agent skill definitions, including the Resonance Audit skill for cognitive bandwidth management, nautilus_mesh, shredder, visual_anchor, and the mined self-improvement family (mine-automations, wrap, fleet-doctor, intake — canon here, deployed by copy to ~/.claude/skills; see docs/loop-doctrine.md).
  • /garage (Fleet CMDB): Hardware fleet registry (fleet.json), workload assignments, migration plans and two dashboards for the SOVERN local nodes (Mac Studio M4 Max, Mac minis, Hetzner hub): the legacy self-contained dashboard.html generator and the interactive Prefab dashboard (app.py + serve_edit.py — filters, search, guarded write-back editing of unit/workload fields straight into the CMDB JSON; prefab export --bundled for an offline artifact). First onboarded Prefab use-case (see core/desops/doctrine/tools-registry.md).
  • /labwatch (Observability / Eagle Eye): system single pane on :4002 (canonical instance on the M4; tailnet https://sovrn-mini.taile5b8dd.ts.net:8443) — five tabs: 🦅 Fleet (hosts × services grid from signed push-reporters on m4/surface/hetzner), 🗺 Topology (garage CMDB joined with live probes, drift/rogue chips), 💰 Usage (quotas, budgets, 7-day tokens), 🛡 Ops (SecOps posture, DesOps audit, orchestrator switcher + agent capability cards parsed from docs/<agent>-capabilities.md), 📖 Шпаргалки (plain-language owner cheat-sheets from docs/cheatsheets/*.md). Collectors, telemetry spool, watcher alerts → Telegram.
  • /secops (Sovereign Shield): Security posture — gitleaks config, Claude posture baseline + audit hook, hardening guides, Wazuh/Falco/CrowdSec infra configs. Absorbed as a subsystem 2026-06-11.
  • /docker: docker-compose stack and env templates for the database/infra services.

External context (outside this repo): C:\telo\Efforts\Simmering — the back-burner zone for ideas fermentation; C:\telo\Atlas\Maps\Interests MOC.md — the central index for high-resonance "Drive" projects.


📂 Sovereign File System Recommendations

Nautilus enforces a Zero-Clutter Root Policy to minimize "Folder Tax" and maximize focus. For a canonical Solo Vibe Coder setup:

  1. ACE Core: The project root (C:\telo) must only contain the three ACE pillars (Atlas, Calendar, Efforts), essential logs, and minimal system configs (.env, .mcp.json).
  2. Agent Redirection: All agent-specific working files (keche, configs, persistent state) must be relocated to a centralized C:\agents\<AgentName> directory and mapped back via hidden Symbolic Junctions.
  3. STIER Dimension: Efforts should be tagged with ⚡ Drive or 🛠️ Duty prefixes to enable the Resonance Audit and maintain energy balance.

🧠 Sovereignty-First Principles

  1. Autonomy Under Failure: The stack is designed to operate sovereignly if every cloud goes dark. The inference fleet is two Macs — M4 Studio 128GB (the beast, gpt-oss-120b = local floor) + M4 mini — reached over the tailnet; local Obsidian directories serve as the guaranteed data floor. (Models no longer run on the laptop — retired 2026-07-16; it had no CUDA runtime and only ran on CPU.)
  2. Zero Telemetry Constraint: Opaque development environments with closed telemetry tracking are strictly prohibited. Open, local-first tools (Zed, Warp, Claude Code) are leveraged exclusively.
  3. Gateway Agnosticism: All Large Language Models are routed through the always-on M4 LiteLLM gateway (https://sovrn-mini.taile5b8dd.ts.net/v1) with dynamic failover arrays (Free cloud pools first → the sovereign beast local-floor as the terminal floor).

Self-improvement контракт проектов (ground truth / drift / activation): docs/loop-doctrine.md. Выбор оркестрационного лейна (hermes / delegates / Workflow / squad): docs/orchestration-doctrine.md.


🚀 Quick Start

1. Environment Preparation

Initialize the environment configuration by copying the template and setting up your keys:

cp .env.example .env

2. Inference Gateway

Inference is served by the always-on M4 LiteLLM gateway on the Mac fleet — nothing to launch on the laptop. Point clients at LITELLM_URL (default https://sovrn-mini.taile5b8dd.ts.net/v1); the SessionStart ensure-gateway hook verifies it and SSH-kickstarts the M4 daemon if it is down. (The old laptop master-restart.ps1 / Port Broker bootstrap was retired 2026-07-16.)

3. Integrated CLI Commands

The Python-powered facet CLI provides a unified control interface:

  • Ingest Documents: Chunk, hash, embed, and transactionally merge files into your knowledge graph with automatic metadata inheritance:
    facet ingest path/to/note.md
  • Audit Directory Contexts: Recursively scan all project directories for schema drift or missing configurations:
    facet audit
  • Open 3D Visualization: Instantly map the current directory context in the WebGL Nooscope UI:
    facet visualize .

🎙 Echo Voice Capture

Echo — the local-first dictation app — can drop voice notes straight into the mesh. Echo is the sensory audio bridge: you speak, and your words become a note Nautilus ingests.

In Echo's Settings → Advanced, set the Voice capture folder to a Nautilus/Obsidian inbox directory and add a trigger phrase (e.g. запиши в наутилус). Saying "<phrase> …" writes a timestamped markdown note (with source: echo frontmatter) into that folder instead of typing it into the active app. The normal facet ingest / n8n pipeline then consumes it — no extra wiring, fully local and sovereign.


📖 Deeper Reading

The full system architecture, bitemporal memory designs, and multi-phase rollout lifecycles are documented in the Nautilus White Paper:

About

ENERV — a fast metadata facet-system for organizing tech and knowledge roots from the command line.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors