00 — Verifikovani System Map (2026-06-22)

ALAI AI System — Verified System Map (input artifact)

Date: 2026-06-22 · Compiled by: John (tool-verified, not from memory) · For: Architect panel (Petter Graff lead)

This is the ground-truth snapshot the architect panel must reason over. Every line below was verified live this session (curl health, process list, DB reads, Slack API, log inspection).

1. Platform services (live status verified 2026-06-22)

2. Engine tools (non-skill/hook/agent)

• mc.js — Mission Control task engine (mc.db) — system core
• discover.js — unified discovery (tools/skills/agents/BookStack/RAG)
• Email suite — email-inbox, email-reactor, email-to-task, email-to-ticket, email-to-contact, email-audit, email-briefing
• slack.js / slack-bot.js — 19 channels (verified live)
• planka-sync.js / planka-admin.js — MC↔Planka bridge
• bookstack-sync.js / bookstack-webhook-relay.js / bookstack-staleness.js
• HiveMind (hivemind-*.js) — internal event bus + knowledge
• LightRAG tools (rag-*, lightrag-auth-helper)
• fiken-* (accounting), documenso-webhook, contacts.js
• Tool Shed (tool-shed.js) — running pid 1319, port 3050, 307 tools registered
• Library v3 (library.js) — 79 global skills, 12 companies, last sync today
• Dashboards (ceo / mc / health / tender / expansion); cost-tracker, agent-manager

3. Data flow (verified wiring)

MIGADU (1 acct, all domains; IMAP boxes: john/info/alai/dev/alem)
  → email-inbox.js (email-inbox.db) → classify (Ollama)
     ACTION → email-to-task.js → MISSION CONTROL (mc.db)   [dedup by message-id]
     INFO/SPAM → log only
     attachments → email-attachment-fetcher → PAPERLESS (archive.alai.no)
MISSION CONTROL ↔ PLANKA  (planka-sync.js bidirectional: createCard/moveCard/syncStatus)
MC/agent events → HiveMind event-bus → LightRAG (knowledge graph) + BookStack (docs)
SLACK = outbound notification layer (ceo-daily-digest / escalation-notify / cron-notify)
Fiken / Documenso / Vaultwarden = accounting / e-sign / secrets (on-demand)

4. Virtual companies (persona-agent model)

ALAI routes work to ~12 "virtual companies", each a persona-agent cluster (specialist-mapping.json):

• CodeCraft (architecture/backend) · Vizu (frontend/design) · FlowForge (devops/infra) · Proveo (QA) · Securion (security) · AgentForge (AI/ML/RAG) · Finverge (fintech) · Skybound (mobile/SaaS) · Lexicon (legal/docs) · Helixsupport (incident) · Proxima (marketing) · Resolver (cross-company systemic)
• John = AI Director / orchestrator (does NOT build); dispatches via MEHANIK gate → specialist → Proveo verify → MC done.

5. Event-sourcing surfaces (verified existing)

• evidence-ledger (SQLite, append; had a dup-insert bloat incident #102796 → dedup index)
• MC task history (mc_history — 542 rows this session window)
• HiveMind events (hivemind.db)
• Session checkpoints (auto-generated session summaries)
• NOTE: these are append-ish logs, NOT a formal event-sourced architecture (no single event store as source of truth, no projections/replay model). This is a key question for the panel.

6. Verified anomalies / smells (honest)

1. BookStack → Slack relay is DISABLED (bookstack-webhook-relay.js: "no Slack call made" — audit-log only).
2. Meta-agent cron daemon runs on schedule but produces 0 tasks / 0 action-items lately (idle loop).
3. Meta-skills pipeline (skill-creator/registry) exists but no new skills created since Jun 15.
4. HexaDB project just frozen (NO-GO) — a case of a bee/6×19 metaphor over-engineered into a DB with no real geo capability. Cautionary precedent: is more of the system metaphor-driven rather than need-driven?
5. mc.js done is gated by a per-session verdict file; research/advisory tasks need --force (friction).