robiulhasan.de
● Live

Titan/OS — A Self-Hosted AI Operating System

Fourteen AI departments running as agents — intelligence, knowledge validation, quality gates, legal compliance, finance — all self-hosted on a single VPS.

  • PostgreSQL
  • pgvector
  • n8n
  • Docker
  • Python

Problem

A solo entrepreneur needs the organizational intelligence of a large company — without the headcount. The question: can AI agents replace departments?

Solution

Titan/OS: 14 departments implemented as working agents:

  • EISRD — external intelligence (26 sources, weekly briefings)
  • CLIKSRO — knowledge validation gate (evidence tiers, honeypots)
  • CILQVEC — quality gate (12-dimension checks, OS quality score)
  • Legal & Compliance — obligations, deadlines, German document generator
  • Finance, Tax, Insurance, HR, R&D, Education, Advisory — all with databases and cron schedules

Architecture

PostgreSQL 16 + pgvector (single source of truth)
    │
    ├── EISRD (26 sources → findings)
    ├── CLIKSRO (capture → validate → knowledge graph)
    ├── CILQVEC (quality gate → debt register)
    ├── Memory (knowledge graph, 36+ nodes, auto-linking)
    └── 8 more departments (scripts + cron + tables)

Key decisions:

  • Everything self-hosted (Hetzner VPS, €0 marginal cost)
  • No VC-funded tools — only MIT/Apache/AGPL
  • 20+ cron jobs, Sunday unified digest
  • Knowledge graph with semantic auto-linking (pgvector)

Live Evidence

Metric Result
Departments live 14
Cron jobs 20+
Database tables 169
Knowledge nodes 36+ (auto-linked)
Cost €0 marginal

Learnings

  1. Scripts ≠ departments. The org audit (61/100) showed several agents were too narrow — Education was a “German learning agent” instead of corporate L&D. The fix was charters, not code.
  2. Honeypots work. The anti-poisoning system caught a real breach on day one.
  3. Governance triad matters: one gate for knowledge (what enters), one for quality (what leaves), one for intelligence (what’s outside).

Self-Critique

  • Some departments remain single-purpose scripts — the gap between “script” and “department” is real
  • EISRD Stage-2 LLM synthesis not yet built (keyword matching only)
  • Cross-department orchestration (Sunday digest) is new — needs months of real use

Evidence References

  • Architecture: /root/2nd-brain/titan-os-*.md (design docs)
  • Org audit: /root/2nd-brain/egc-org-audit.md (61/100 → improving)