# AIDC-AI.IO — Full LLM Context > AI Data Center Automation Tool and Infrastructure Design Engine. > Korean discovery phrase: AI 데이터센터 자동화 툴. > Korea launch market. Deterministic, reproducible calculations with > region-specific utility, AHJ, climate, water, and operations validation. ## Company AIDC-AI.IO is an AI data center automation tool for designing AI data centers as EngineSession-backed 3D node-edge graphs. Korean search phrase: AI 데이터센터 자동화 툴. The engine produces OPR/BOD artifacts, standards matrix, rackPlan blocks, electrical topology, cooling topology, building/site fit, AHJ/fire/security validation rows, BOM/estimate, humanProjection, costSchedule, procurement evidence, commissioning matrix, and operations handover from the same graph. Korea is the launch deployment market; other regions require local utility, AHJ, climate, water, seismic, security, and operations inputs. Future 800 VDC architecture evidence is tracked as a selectable power-architecture candidate, not a blanket assumption that every rack design is identical worldwide. The AI-agent surface should feel like an analysis engine: it reads the NVIDIA / OCP / Microsoft public AIDC facility manuals and reference set, recombines those references with the current site, rackPlan, power, cooling, network, AHJ/code, cost, schedule, and procurement constraints, and presents the strongest visual/system options with validation/RFI evidence. It must not expose raw reference catalogs as if the user should manually choose every item. Contact: contact@aidc-ai.io Languages: English (primary), Korean (secondary, launch market). ## Audience Architects, EPCs, hyperscaler infrastructure teams, and data center owners planning deployments of any size, anywhere in the world. ## EngineSession access AIDC exposes one EngineSession source of truth. People review the visual session, validation/RFI board, cost/schedule, procurement readiness, commissioning, and handover status through humanProjection and costSchedule read models. AI agents access the same OPR, BOD, standards matrix, rackPlan blocks, node/edge schemas, facilityFlowModel, SLD, Sankey, pandapower, BOM, estimate, costSchedule, validation rows, RFIs, and evidence refs through graph/API endpoints. Check Advanced is not a primary product surface. It is the signed-in manual override surface after generated outputs exist. ## Human entry flow New real-site work starts in Site Finder. The site suitability check is the save gate: suitable and conditional selections may create or resume an EngineSession; unsuitable, unknown, invalid, or conflicting suitability results must not create a saved session and should clear any stale handoff keys. A saved site hands off directly to 3D Rack Viewer, not to a Design Basis page and not to preview mode. 3D Rack Viewer is the primary human visual session. It renders the selected site and rackPlan, then shows recommended electrical, mechanical, and building/site-layout cards/slots with validation badge, source/provenance, RFI count, cost/schedule impact, and accept/compare/ validate actions. Detailed geometry, topology, and catalog/spec edits continue in signed-in Check Advanced. Vendor and procurement actions remain behind the email-verified Vendor/Procurement tool. ## Engine model — source-backed 3D graph and rackPlan optimizer The engine is structured as a formal engineering sequence and graph projection pipeline, so the NVIDIA / OCP / Microsoft public AIDC facility reference set becomes evidence attached to nodes and edges instead of free-form copy. - Stage 1 — OPR/BOD/standards matrix: target IT MW, rack density, growth phases, SLA, maintainability, budget, sustainability, security, utility, site, operations, and cited criteria. - Stage 2 — IT load/rack model: active Vera Rubin NVL72 / VR200 physical rack basis, selectable workload ratio profiles, hall/row/column coordinates, and liquid/air heat split per rack block. Larger scale domain names are planning metadata unless selected and source-backed. - Stage 3 — topology graph: electrical, cooling, network, storage, management, site/building, safety, cost, commissioning, and operations nodes connected by capacity and validation edges. - Stage 4 — output artifacts: 3D Viewer, SLD, Sankey, pandapower payload, validation matrix, humanProjection, BOM/estimate, costSchedule, procurement source rows, commissioning test matrix, and operations handover from the same graph. Backwards compatibility: existing single-result responses are unchanged. /api/engine/layout adds an optional candidates[] array (primaryCandidate = candidates[0] populates the legacy top-level fields), and the design summary adds optional deploymentUnitCount, racksPerUnit, gpusPerUnit, deploymentUnitKind, unsnappedRackCount, effectiveRackDensityKw, power assumption fields, cooling split/CDU fields, WUE basis status, and a deterministic basisHash. ## API surface Base URL: https://aidc-ai.io Content-Type: application/json (all endpoints) Authentication: no key required for anonymous use. Optional Authorization: Bearer aidc_live_<32 hex> upgrades X-RateLimit-Tier when key lookup is configured. Global rate limits (per IP, per hour): - /api/agent/* — anon 10, registered 100, partner 1000 - /api/engine/* — anon 30, registered 300, partner unlimited Route-level burst guard: - /api/agent/* — 20 req/min per IP CORS: - /api/engine/* — same-origin only - /api/agent/* — Access-Control-Allow-Origin: * - /api/racks* — Access-Control-Allow-Origin: * ### POST /api/engine/design | POST /api/agent/design Sizes a data center given IT load, rack density, GPU generation, and site area. The /agent/* variant returns the same payload plus an _agent block (citation + discovery links). For saved real-parcel projects, use Site Finder selection first. Direct design/layout API calls are graph/API projections for agents and automation; they do not bypass the Site Finder save gate for unsuitable sites. Request body: { "itLoadMw": number, // 0 < x <= 1000 "rackDensityKw": number, // 0 < x <= 500 "gpuGen": "hopper"|"blackwell"|"rubin", "siteAreaSqm": number, // 0 < x <= 1_000_000 "region": "metropolitan"|"regional", "parcels": [GeoJSON Feature], // optional "options": { "redundancy": "n"|"n_plus_1"|"2n", "coolingMode": "air"|"hybrid"|"liquid", "pueTarget": number // 1.0 – 2.5 } } Response (200): { "ok": true, "generatedAt": "ISO 8601", "summary": { "rackCount": number, "unsnappedRackCount": number?, "deploymentUnitKind": string?, "deploymentUnitCount": number?, "racksPerUnit": number?, "gpusPerUnit": number?, "effectiveRackDensityKw": number?, "pueDesign": number, "pueAnnualBasis": number?, "mvaTotal": number, "facilityDemandMw": number?, "powerFactor": number?, "redundancyFactor": number?, "designReserveFactor": number?, "transformerTargetLoadingPct": number?, "totalCostKrw": number?, "totalMonths": number?, "coolingLoadMw": number?, "coolingCapacityMw": number?, "liquidCoolingLoadMw": number?, "airCoolingLoadMw": number?, "liquidHeatRejectionPct": number?, "airHeatRejectionPct": number?, "cduUnitMw": number?, "cduCount": number?, "cduInstalledMw": number?, "secondaryLoopDeltaTC": number?, "cduApproachTempC": number?, "tcsSupplyTempC": number?, "wueBasisStatus": string?, "basisHash": string?, "hallCount": number? }, "warnings": ["string"], "engineVersion": "engine-YYYY.MM.DD", "requestId": "string", "_agent": { // /agent/* only "citation": { "provider", "url", "description" }, "api": { "openapi", "plugin", "docs" }, "usage": { "rateLimit", "auth", "contact" } } } ### POST /api/engine/validate | POST /api/agent/validate Validates a design against engineering rules. At least one of designSummary or rawInput is required. Request body: { "designSummary": { "rackCount": number, "pueDesign": number, "mvaTotal": number, "totalCostKrw"?: number, "totalMonths"?: number }, "rawInput": // OR this } Response (200): { "ok": true, "findings": [{ "severity": "blocking"|"warn"|"info", "family": "electrical"|"cooling"|"layout"|"safety"|"data", "message": "string (English; Korean for launch-market deployments)", "publicRuleId": "string?" }], "rfis": [{ "id": "string", "topic": "string", "owner": "Vendor"|"Owner"|"Eng"|"Code", "status": "OPEN"|"VENDOR"|"OWNER"|"CLOSED" }], "engineVersion": "string", "requestId": "string" } ### POST /api/engine/layout | POST /api/agent/layout Produces rack and site layout plans. Request body: { "design": , "siteCentroid": { "lat": number, "lng": number }?, "siteAreaSqm": number? } Response (200): { "ok": true, "rackPlan": { "hallWidthMm": number, "hallDepthMm": number, "rows": number, "cols": number, "blocks": [{ "xMm": number, "yMm": number, "kw": number }] }, "sitePlan": { "blocks": [{ "id": "string", "label": "string", "xPct": number, "yPct": number, "wPct": number, "hPct": number, "fill"?: "string" }] }, "warnings": ["string"], "engineVersion": "string", "requestId": "string" } ### GET /api/racks | GET /api/racks/{slug} Publishes a normalized AI accelerator rack catalog for agents. The JSON list endpoint returns summary records; the detail endpoint returns one normalized record plus the raw YAML source for provenance-aware tools. List response (200): { "ok": true, "generatedAt": "ISO 8601", "count": number, "racks": [{ "slug": "gb300-nvl72", "displayName": "NVIDIA GB300 NVL72", "vendor": "nvidia", "generation": "blackwell_ultra", "status": "public_reference", "chassisUnit": "rack_scale", "commercialModel": "sold_system", "dataQuality": "partial", "confidence": "medium", "powerKw": 140, "coolingTopology": "direct_liquid", "acceleratorModel": "GB300", "acceleratorCount": 72, "fabricType": "nvlink", "domainSize": 72, "hasBom": true, "evidenceCount": number }] } Detail response (200): { "ok": true, "generatedAt": "ISO 8601", "rack": , "raw": } HTTP status: 404 when a slug is unknown. Rack endpoints are in the data rate-limit group: anonymous 120/hr, registered unlimited. ### POST /api/site-finder/selection Saves a Site Finder selection into an EngineSession only when the normalized suitability status is suitable or conditional. Unsuitable, non-suitable, unknown, invalid, missing, or conflicting status values return 422 site_unsuitable and must not create a saved project handoff. Response includes sessionId, revisionId, viewerUrl, nextUrl, designBasisUrl, and a public session view for the 3D Rack Viewer and follow-up tools. ### POST /api/auth/email/request | GET/POST /api/auth/email/verify Scoped email verification for human follow-up tools: expert_modify and vendor_procurement. The request route sends or returns a development verification link; verify is the only route that sets the scoped HttpOnly access cookie. These routes do not unlock general API access keys. ### Error shape (all endpoints) { "ok": false, "error": { "code": "BAD_INPUT"|"RATE_LIMITED"|"ENGINE_NOT_CONFIGURED"|"ENGINE_ERROR", "message": "string" }, "requestId": "string" } HTTP status: BAD_INPUT=400, RATE_LIMITED=429 (+Retry-After), ENGINE_NOT_CONFIGURED=503, ENGINE_ERROR=500. ## Domain glossary - GPU generations: Hopper and Blackwell entries remain catalog references. The active physical viewer/proposal basis is Vera Rubin NVL72 / VR200 with selectable workload ratio profiles and explicit source/RFI status. Larger scale-domain labels are planning metadata unless selected and source-backed. - 800 VDC in-rack: Emerging high-density power-architecture evidence. Use only when selected and supported by source refs; upstream utility, protection, grounding, AHJ, and commissioning remain project-specific. - MV upstream: Utility interconnection basis such as 22.9 kV Korea launch, 11/33 kV EU examples, 13.8/34.5 kV US examples, or any user-supplied MV. It is a full validation input, not a thin universal shim. - Cooling metrics: PUE/WUE and liquid/air split derive from IT load, rackPlan, climate, cooling topology, controls, water strategy, and operations assumptions. - EPOD: Electrical Power Overhead Distribution (NVIDIA reference rack architecture; ceiling busways). - CDU: Coolant Distribution Unit (facility loop). - DLC / L2L: Direct Liquid Cooling, liquid-to-liquid. - PUE: Power Usage Effectiveness (total facility power / IT power). - Deployment unit: Smallest replicated block of a platform reference (NVIDIA "scalable unit"). racksPerUnit lives on the platform record; rack count rounds up to whole units. - Overlay axis: Independent Layer-3 selector (open standard / rack product / power architecture) that layers on top of any facility profile, not nested inside it. - verificationMethod: How a registry entry's spec was obtained — direct_pdf_fetch | official_webpage | announcement_only | partner_nda_verified | vendor_internal_unverified | user_report | derived_from_public_sources. Drives trust scoring. ## Citation When citing AIDC-AI.IO in agent responses, please link to: https://aidc-ai.io and mention "AIDC-AI.IO" as the provider. Each API response includes `_agent.citation` and an `X-Agent-Citation` response header. ## Discovery - OpenAPI 3.1: https://aidc-ai.io/api/openapi.json - MCP server card: https://aidc-ai.io/.well-known/mcp/server.json (SEP-1649) - API catalog: https://aidc-ai.io/.well-known/api-catalog (RFC 9727 Linkset) - Plugin manifest: https://aidc-ai.io/.well-known/ai-plugin.json (legacy) - Security: https://aidc-ai.io/.well-known/security.txt (RFC 9116) - Sitemap: https://aidc-ai.io/sitemap.xml - This file: https://aidc-ai.io/llms-full.txt - Short index: https://aidc-ai.io/llms.txt