# AIDC-AI.IO

> AI Data Center Automation Tool and infrastructure design engine for source-backed 3D graph design.
> Korean discovery phrase: AI 데이터센터 자동화 툴.
> Korea live. Region-specific utility, AHJ, climate, and operations validation.

## Overview

AIDC-AI.IO is an AI data center automation tool that creates and
validates AI data-center designs as one EngineSession-backed 3D
node-edge graph. In Korean search terms, it is an AI 데이터센터 자동화 툴
for data center design automation. The formal workflow is OPR -> BOD
-> standards matrix -> IT load/rack model -> electrical topology ->
cooling topology -> building/site fit -> code/AHJ/fire/security
validation -> cost/procurement -> commissioning test matrix ->
operations handover.
The NVIDIA / OCP / Microsoft public AIDC facility reference set,
vendor data, AHJ/code inputs, utility data, and owner inputs are
evidence for that graph, not replacements for project-specific
validation. No LLM in the engine itself — calculations and validation
rows are reproducible and traceable to inputs.

The intended AI-agent behavior is to analyze those public AIDC facility
manuals and references, recombine them with the selected site, rackPlan,
power, cooling, network, AHJ/code, cost, schedule, and procurement
constraints, and present the strongest visual/system options rather than
asking the user to browse raw catalogs.

The public product exposes one EngineSession. People review the visual
session through humanProjection and costSchedule read models; AI agents
read and mutate the same OPR, BOD, standards matrix, rackPlan blocks,
node/edge schemas, facilityFlowModel, SLD/Sankey/pandapower, BOM,
estimate, costSchedule, validation rows, RFIs, and evidence refs over
the graph/API surface.

Human new-project flow: Site Finder first. Suitable or conditional
sites create/save EngineSession and hand off to 3D Rack Viewer. The 3D
Rack Viewer is the primary human visual session where recommended
electrical, mechanical, and building slots are reviewed with validation
and provenance. Unsuitable sites must not create saved sessions.

Check Advanced and Vendor/Procurement are follow-up surfaces, not a third
primary track and not default browsing surfaces. Manual Check/3D edit actions
are enabled after sign-in; Vendor/Procurement remains email-verified.

## Engine model — source-backed 3D graph and rackPlan optimizer

The engine honors reference architectures through a graph model instead
of a single rack-density guess:

- L1 · OPR/BOD basis: target IT MW, PUE/WUE, rack profile, growth
  phases, SLA, maintainability, budget, sustainability, security,
  utility, site, and operations assumptions.
- L2 · deployment intent: maximum_density | phased_migration | balanced
  | mixed_generation, plus selected workload ratio profile.
- L3 · 3D rackPlan graph: hall/row/column rack blocks, node schema,
  power/cooling/network edges, source refs, confidence, RFI flags, and
  validation status.
- L4 · output: 3D Viewer, SLD, Sankey, pandapower payload, validation
  matrix, humanProjection, BOM/estimate, costSchedule, procurement
  evidence, commissioning matrix, and operations handover derived from
  the same graph.

## Engine API

Core engine endpoints are same-origin JSON APIs:

- POST /api/engine/design     — OPR/BOD-derived design summary
- POST /api/engine/validate   — Rule findings + RFIs
- POST /api/engine/layout     — RackPlan + site layout graph
- POST /api/site-finder/selection — Suitable/conditional site save gate

Cross-origin AI-agent wrappers add citation metadata and permissive CORS:

- POST /api/agent/design      — cited design wrapper
- POST /api/agent/validate    — cited validation wrapper
- POST /api/agent/layout      — cited layout wrapper

Scoped human follow-up access uses sign-in for manual Check/3D edits and
/api/auth/email/request plus /api/auth/email/verify for Vendor/Procurement.

Design summaries include deployment-unit-snapped rack counts for Rubin
VR200 NVL72 planning zones, raw unsnapped rack count, PF/reserve
electrical assumptions, liquid/air heat split, CDU planning values, WUE
basis status, and a deterministic basisHash.

## Rack Library

Public catalog for AI accelerator systems and rack-scale platforms:

- GET /api/racks             — JSON list of normalized rack summaries
- GET /api/racks/{slug}      — JSON detail plus raw source YAML

Entries are normalized from vendor folders under the rack library and keep
provisional fields labeled with confidence, evidence count, and data quality.

## Discovery

- OpenAPI 3.1 spec:  https://aidc-ai.io/api/openapi.json
- MCP server card:   https://aidc-ai.io/.well-known/mcp/server.json
- API catalog (RFC 9727): https://aidc-ai.io/.well-known/api-catalog
- Plugin manifest:   https://aidc-ai.io/.well-known/ai-plugin.json
- Security policy:   https://aidc-ai.io/.well-known/security.txt
- Full LLM context:  https://aidc-ai.io/llms-full.txt

## Rate limits (per hour)

All /api/* responses include X-RateLimit-Limit, X-RateLimit-Remaining,
X-RateLimit-Reset, X-RateLimit-Tier, X-RateLimit-Group headers.
On 429 a Retry-After header is set. Auto-backoff is encouraged.

Anonymous (per IP):
- /api/agent/*      10/hr
- /api/engine/*     30/hr
- /api/vworld/*     600/hr
- /api/data/*       120/hr
- /api/racks/*      120/hr
- /api/contact      5/hr
- /api/estimate/*   5/hr

Registered (optional Bearer aidc_live_<32 hex>, when key lookup is configured):
- /api/agent/*      100/hr
- /api/engine/*     300/hr
- /api/vworld/*     600/hr
- /api/data/*       unlimited
- /api/racks/*      unlimited
- /api/estimate/*   30/hr

Partner:
- /api/agent/*      1000/hr
- /api/engine/*     unlimited
- /api/vworld/*     unlimited
- /api/data/*       unlimited
- /api/racks/*      unlimited
- /api/estimate/*   unlimited

No key is required for anonymous use. Valid Bearer keys upgrade the
X-RateLimit-Tier header when configured. Contact contact@aidc-ai.io.

## Contact

contact@aidc-ai.io