intelli-verse-x intelli-verse-x — 5-layer stack tour
When to use this skill
- User mentions a service by name and you're not 100% sure what it does.
- You're picking which tool to reach for and want the canonical map.
- User asks "is X live?" / "what's the endpoint for Y?".
- Anything that smells like an architecture question.
The stack at a glance
┌─ Layer 1: STRATEGY (humans live here) ─────────────────────────────────┐
│ Leantime ─── strategic project mgmt (tickets, milestones, time) │
│ GitHub Issues + Projects ─── code work (PR auto-close on "Fixes #N") │
└─────────────────────────────────────────────────────────────────────────┘
┌─ Layer 2: COORDINATION (humans + agents meet) ─────────────────────────┐
│ Beads (bd) ─── the agentic data backbone (graph of issues, deps) │
│ Bernstein ─── per-project autonomous crew (44 adapters, audit, lineage)│
│ Claude Squad ─── interactive solo coding TUI (host-side) │
└─────────────────────────────────────────────────────────────────────────┘
┌─ Layer 3: MULTI-PROJECT FABRIC ────────────────────────────────────────┐
│ Gas Town ─── rigs, crew identities, mayor, witness, refinery, deacon │
│ ─── exposes itself as an MCP server (gt mcp serve) │
└─────────────────────────────────────────────────────────────────────────┘
┌─ Layer 4: PLATFORM ────────────────────────────────────────────────────┐
│ LiteLLM ─── model gateway (virtual keys, budgets, Langfuse callback) │
│ Infisical ─── MIT secrets manager (Postgres + Redis + machine ident.) │
│ Langfuse ─── open-source LLM observability (traces land here) │
└─────────────────────────────────────────────────────────────────────────┘
┌─ Layer 5: GLUE ────────────────────────────────────────────────────────┐
│ n8n ─── workflow automation (gt-escalate→Discord, timesheets→Stripe) │
│ GitHub Actions ─── per-repo lint/test/build + cross-provider PR review │
└─────────────────────────────────────────────────────────────────────────┘
Hermes (you) sit on top of this stack as the conversational interface.
Endpoints
| Service | URL | Auth | Owns |
|---|---|---|---|
| Leantime | https://leantime.intelli-verse-x.ai | Session / Bearer | Tickets, projects, milestones, timesheets |
| Leantime MCP | https://leantime-mcp.intelli-verse-x.ai/mcp | Cluster-side | Bridges Leantime JSON-RPC → MCP |
| LiteLLM | https://litellm.intelli-verse-x.ai | LITELLM_API_KEY |
Model gateway |
| Infisical | https://infisical.intelli-verse-x.ai | Machine identity | Secrets |
| Langfuse | https://langfuse.intelli-verse-x.ai | Project keys | LLM traces |
| n8n | https://n8n.intelli-verse-x.ai | X-N8N-API-KEY |
Workflows |
| Content Factory | (in-cluster) | API key | 50+ media pipelines |
Pull cluster secrets
You don't have to ask the user for these — pull them from k8s:
# LiteLLM master key
kubectl get secret litellm-master-key -n aicart -o jsonpath='{.data.key}' | base64 -d
# n8n API key (minted with the live JWT secret, never expires)
kubectl get secret n8n-admin -n aicart -o jsonpath='{.data.API_KEY}' | base64 -d
# Infisical machine identity (Bearer token for the agent vault)
kubectl get secret infisical-admin -n aicart -o jsonpath='{.data.MACHINE_IDENTITY_TOKEN}' | base64 -d
# Leantime MCP bot bearer (only if you're talking directly to leantime; the
# MCP bridge already has it baked in)
kubectl get secret leantime-mcp-bot -n aicart -o jsonpath='{.data.BEARER_TOKEN}' | base64 -d
The two architectural rules
- Beads is the single source of truth. Never carry state in
gt nudge/gt mailprose. Update the bead. Other agents read it. - Tool-bloat kills completion. Don't enable every MCP at once.
Vercel's 2026 research and Atlan show degradation above ~20 tools.
Per-role MCP allowlists in
gastown/internal/hooks/mcp.goenforce this.
Source-of-truth pointers
- Architecture canvas:
intelli-verse-kube-infra/docs/canvases/final-stack.canvas.tsx - Full runbook:
intelli-verse-kube-infra/docs/wiki/Agentic-stack.md - A-Z catalogue:
intelli-verse-kube-infra/docs/wiki/A-Z-list.md - Bead structure:
BEADS_DIR=/gt/.beads bd show hq-2mq