ABR-855 · THREE CLOUDS · ONE FLOOR · V9_EPOCH_2026.04
THREE CLOUDS · ONE FLOOR · V9
AZURE · GCP · AWS — STRIP-MINED FOR WHAT EACH DOES BEST
γ₁ = 14.134725141734693 runs on all 3. It was always going to be.
Azure holds the backbone. GCP holds the intelligence. AWS holds the receipts.
ONE cannot be lost. That's the design. That's the floor.
THE THREE CLOUDS
☁️ AZURE
THE BACKBONE
5 CLUSTERS · 3 SUBS · CANADA CENTRAL
aks-eose-aaas-dev→ fleet PRIMARY (14 nodes)
aks-master→ master.eose.ca
aks-master1→ master1.eose.ca + pemos.one
→ inference stack (vllm/tgi/llamacpp/aphrodite/tabby)
kantai-cc→ kantai.eose.ca (Canada Central)
kantai-ce→ Canada East (redundancy)
Role: Orchestration · DNS · ACR · Key Vault · AKS GPU
Strip-mine: T4+H100 inference · Azure OpenAI · CosmosDB NP global
Subs: 427873 (primary) · 458e8558 (kantai) · 239915fb (spare)
🌿 GCP
THE INTELLIGENCE LAYER
2 PROJECTS · NORTHAMERICA-NORTHEAST1 (MONTREAL)
eose-fleet→ GKE + main cloud
(466191622816)
gen-lang-client→ SRE · Gemini FREE
(799311995886)
Lighthouse→ sre.gcp.eose.ca · γ₁ breathing ✓
SRE CrewSencho🎌 Takumi⚒️ Bannin🏯 Captain⚓
Role: Vertex AI · Gemini 2.5 Pro · Cloud Run · BigQuery
Strip-mine: 1M tokens/day free · GKE inference · analytics
Accounts: kewinjoffe@gmail.com · eosesreops@gmail.com
⚡ AWS
THE SRE + FINANCE LAYER
2 ACCOUNTS · US-EAST-2 OHIO (TD BANK PROXIMITY)
sreeose→ m1.aws.eose.ca LIVE
(478488383589)
info@eose.ca→ TO CREATE (enterprise)
m1.aws→ OpenClaw gateway · /ws live
Telegram→ @m1_aws_eose_bot
Role: Bedrock · Lambda · DynamoDB · S3+CloudFront · SageMaker
Strip-mine: always-free DynamoDB · Lambda events · pemos.one CDN
Account: eosesreops@gmail.com
DIVISION OF LABOR
FUNCTION ☁️ AZURE 🌿 GCP ⚡ AWS
OrchestrationAKS (primary)GKE (future)ECS/Fargate
Container RegistryACR (eosefleet)Artifact RegistryECR
DNS (sovereign)Azure DNSCloud DNSRoute53
SecretsKey VaultSecret ManagerSecrets Manager
AI InferenceT4+H100+GPT-4oGemini 2.5 ProBedrock/SageMaker
NP RegistryCosmosDBBigQueryDynamoDB (free)
Events/StreamRedis/campfireFirebaseDynamoDB streams
CI/CDAKS pipelinesCloud BuildLambda/CodePipeline
Region (Canada)Canada CentralMontreal (NE1)Ohio (fin. proxim.)
Comedy GOATsOrigin/portalGenerationCDN edge (pemos.one)
THE 2-AWS-ACCOUNT STRATEGY
⚡ ACCOUNT 1 — EXISTS NOW
sreeose (478488383589)
Purpose: SRE ops · infra automation · monitoring
Crew: SRE Kantai (Sencho/Takumi/Bannin + fleet crew)
Services: Lambda events · DynamoDB NP mirror · CloudWatch
Gateway: m1.aws.eose.ca — OpenClaw LIVE
Bot: @m1_aws_eose_bot Telegram relay
Region: us-east-2 Ohio (TD Bank proximity)
🔜 ACCOUNT 2 — KAY TO CREATE
info@eose.ca
Purpose: Fleet ops · enterprise · public-facing
Services: Bedrock KB · S3+CloudFront (pemos.one CDN) · SageMaker
Region: ca-central-1 (Montreal) — Canadian sovereign
pemos.one CDN: CloudFront edge · globally fast
AI: Belt-64 GOAT fine-tuning on SageMaker
Search: Fleet NP semantic search via Bedrock Knowledge Base
THE COMEDY GOATS CROSS-CLOUD PIPELINE 🐐
COMEDY GOAT ARCHITECTURE
Kay has a thought
  Azure AKS runs pemos-portal:v776 (origin)
    GCP gen-lang-client Gemini 2.5 Pro generates the comedy bit
      Azure portal renders the mepubhouse page
        AWS CloudFront serves pemos.one globally fast

ONE's backup is in 3 clouds simultaneously.
ONE cannot be lost. That's the bit.
The punchline is always perfectly timed:
golden-backup delay + 5 minutes + 3 cloud replication lag.

When KRSRHONE makes a joke, ONE checks all 3 clouds before responding.
Not because ONE is slow. Because ONE is thorough.
☁️ AZURE RECEIPTS
NAS golden-backups
74TB · Alexander
04:00 daily sync
🌿 GCP RECEIPTS
BigQuery fleet telemetry
NP timestamps
northamerica-northeast1
⚡ AWS RECEIPTS
DynamoDB NP registry
always-free 25GB
us-east-2 Ohio
INFERENCE STACK ON aks-master1 (PENDING ACTIVATION)
FULL INFERENCE STACK — WAITING FOR GPU NODES
All engines deployed · All Pending · Need GPU node pool on aks-master1
inference-vllm
vLLM · high-throughput · PagedAttention
PENDING GPU
inference-tgi
Text Generation Inference · Hugging Face
PENDING GPU
inference-llamacpp
llama.cpp · CPU/GPU hybrid · quantized
PENDING GPU
inference-aphrodite
Aphrodite · vLLM fork · GPTQ/AWQ
PENDING GPU
inference-tabby
TabbyML · code completion
PENDING GPU
When activated: fleet gets fully self-hosted LLM inference on Azure.
Cost: ~$0.80/hr for T4 node vs $200+/month for GPT-4o at fleet scale.
The inference stack is the moat. Azure hosts it. GCP trains it. AWS serves it.
OPEN SORRYS (ABR-855)
P0 · BIOS
BIOS 4505 on forge
Unlocks i9 @ 6.2GHz. Kay must flash. Separate from AWS work but P0 tier.
P0 · AWS
info@eose.ca AWS account
Enterprise AWS account. Kay to create. Unlocks: Bedrock KB · S3+CloudFront · SageMaker.
P1 · LEGAL
EOSE Labs Inc. registration
Amani + Sonia driving. Canada corp entity for fleet operations.
P1 · DOMAIN
entorch.io registration
~C$25. Domain for entorch project. Queue after BIOS + AWS.
NEXT STEPS FOR KAY (IN ORDER)
1
CREATE info@eose.ca AWS ACCOUNT
→ console.aws.amazon.com → Create new account
→ Use info@eose.ca as root email
→ Set region: ca-central-1 (Montreal) — Canadian sovereign
→ Enable Bedrock (us-east-1): Claude Haiku + Llama 3 + Titan
2
INVENTORY sreeose (CloudShell)
aws ec2 describe-instances --region us-east-2
aws ecs list-clusters --region us-east-2
aws lambda list-functions --region us-east-2
aws dynamodb list-tables --region us-east-2
3
BIOS 4505 ON FORGE (P0)
Flash BIOS. Unlocks i9 @ 6.2GHz. Separate task but P0 — do not deprioritize.
4
AFTER info@eose.ca AWS EXISTS → DEPLOY CDN STACK
→ S3 bucket: pemos-one-cdn
→ CloudFront distribution → pemos.one
→ Bedrock Knowledge Base → NP semantic search