Project KRONOS: A Personal AI Operating System

The Challenge

A knowledge vault of 2,000+ notes is only as useful as the speed at which information can be retrieved. The FuSiON vault started as an Obsidian notebook and evolved into an AI-augmented productivity system — but the architecture underneath hadn’t kept pace. File-scan-based indexing meant:

• Rebuilding indexes manually after every significant vault update
• Agent queries scanning thousands of files on every request (3–8 second latency)
• External data (calendar, email, fitness) fetched live on every call with no caching
• Windows Task Scheduler managing dozens of fragile scheduled jobs

The goal of Project KRONOS was to close this structural gap by replacing the file-scan layer with a live, always-current database backend.

The Architecture

KRONOS introduces a persistent Python runtime sitting between the vault and the AI agent layer. It runs four concurrent threads:

Obsidian (human UI — unchanged)
        ↕ filesystem events
FuSiON Runtime (single Python process)
    ├── MCP Server       (38 tools, SQL-backed)
    ├── Watcher Thread   (vault → SQLite sync, <2s)
    ├── Scheduler Thread (APScheduler, replaces Task Scheduler)
    └── API Broker       (calendar / email / fitness, cached)
        ↕ fusion.db (SQLite, WAL mode)
        ↕ delta export/import
openFusion SQLite — replicated backup node
        ↕ FastAPI
Electron Frontend — runtime health console

Key Technical Work

1. Real-Time Vault Watcher

A filesystem event listener propagates any Obsidian file save to SQLite within 2 seconds. The database schema mirrors the vault structure (notes, tasks, projects, tags, links, journal entries) in WAL mode for concurrent read performance.

2. MCP Tool Rewrite (38 Tools)

All 38 Model Context Protocol tools were migrated from file-scan implementations to SQL queries. External signatures remain identical — the AI agent layer required no changes. Query latency dropped from file-scan baseline to sub-100ms for most operations.

3. API Broker

A caching broker pre-fetches Google Calendar, Gmail, and fitness data on a configurable schedule. MCP tool calls that previously hit external APIs live now return from cache in under 100ms.

4. Scheduler Consolidation

All scheduled FuSiON jobs (daily digest, index refresh, fitness sync, evening ritual) migrated from Windows Task Scheduler to APScheduler running in-process. Jobs are visible and controllable from the Electron console.

5. 5-Tier Python Environment Architecture

To manage dependency isolation across a mixed workload (real-time MCP gateway, direct ChromaDB operations, ML/vision, UI layer, development sandbox), the project uses five purpose-built virtual environments:

6. Electron Management Console

An Electron desktop app provides runtime health monitoring: watcher status, scheduler job list, API broker cache state, and SQLite stats. No note viewing — purely an operational dashboard.

Outcome

• Zero manual index rebuilds in standard working sessions
• Sub-100ms agent context delivery for task, project, and calendar queries
• Full vault-to-AI pipeline observable and controllable from a single console
• Scheduler, broker, and watcher lifecycle managed as a single process

KRONOS demonstrates that the same engineering discipline applied to enterprise software (event-driven architecture, WAL databases, connection pool management, multi-threaded service design) transfers directly to personal productivity infrastructure.