// ACTIVE LOOPS: 14 · // DATA ROUTED (24H): 1.2TB · // LAST DEPLOY: 14 MINS AGO · // MCP GATEWAYS ONLINE: 47 · // LEGACY ERP BRIDGES: 23 · // BACNET NODES: 1,284 · // UPTIME (30D): 99.987% · // QUEUE DEPTH: 3
// INSIDE THE LOOP

How the Engine works.

A scrollable node map showing how an Axiom intake ticket becomes a deployed Model Context Protocol server in 72 hours. Every station, every artifact, every shell command in the chain.

// NODE MAP · 6 STATIONS
node://n-01
H+00:00

Intake

artifact: intake.json

Client submits a request in the portal. Routing AI parses the ask, normalizes docs, schemas, and screenshots.

POST /api/intake → { type: 'integration', priority: 'high' }
node://n-02
H+01:00

Scope Split

artifact: milestones.yaml

Milestones detected and ranked. Senior engineer assigned. Estimate locked. Client approves before meter starts.

axiom scope --milestones=3 --owner=senior-eng
node://n-03
H+04:00

Hour-Zero Scaffold

artifact: mcp-server.ts

Staging deploy. Auth shell, error capture, preview URL. MCP gateway boilerplate generated from schema.

axiom scaffold --target=mcp --transport=stdio,http
node://n-04
H+12:00

Legacy Mapping

artifact: schema.map.ts

Proprietary ERP endpoints reverse-engineered. EDI/X12 → JSON translation layer. Vector store seeded.

axiom map --legacy=sage --target=mcp --emit=zod
node://n-05
H+48:00

Tooling & Tests

artifact: tools/*.ts · *.test.ts

MCP tools defined with strict schemas. Integration tests against real endpoints. LLM eval harness wired up.

axiom test --suite=integration --against=staging
node://n-06
H+72:00

Production Deploy

artifact: prod.deploy.log

Push to prod edge. Telemetry online. Runbook handed off. Loop closes. Next intake opens.

axiom deploy --prod · p95=47ms · loop closed
// PRINCIPLES

Three rules that keep the loop closing.

One request, one loop.

We don't context-switch mid-sprint. Concurrent work runs on its own concurrent track.

Ship before polish.

Day-one preview URL. Real data over mockups. Polish in the same loop, not in a separate phase.

AI handles the toil.

Codegen, schema diffing, scaffolding, test boilerplate — automated. Humans on architecture and judgment calls.