La boutique est ouverte

Nous construisons les machines Workstation AI pour les sites web, les agents IA et l'entraînement de LLM — configurez-les et achetez-les en ligne.

Obtenir les prix pro
Produits IA
Labs IAAgents OpenAIAgents ClaudeGrok BotCRM natif Workstation AIMarketingTous les produits
Solutions IA
IA appliquéeAgents IAIA générativeAutomatisation par IAMontée en compétence IAOffres IA pour PMELaboratoire IA d'entrepriseStations de travail IAMatériel et logiciels NVIDIACalcul accéléré NVIDIAIA privéeCloud privé GPUClusters GPUIA en périphérieIA par secteur
Services IA
Modernisation des plateformesIngénierie numériqueFondations data et IAOpérations autonomesConseil IAAutomatisation DevOpsCybersécuritéDéveloppement logicielCréation d'agentsMise en place MLOpsLocation de GPUCloud privé GPU
À propos
EntrepriseNotre histoirePartenairesTémoignages clients
Centres de données IA
Centres de données IACloud privé GPUClusters GPULocation de GPU
Analyses IASérie JevLabo de benchmarksChoix de modèles
Notes techniques
Documentation
OS LinuxUbuntuRed Hat Enterprise LinuxSUSE LinuxAlpine LinuxDebian, Rocky & AlmaLinux
Plateformes cloud d'IAAWS & Amazon BedrockGoogle Cloud & Vertex AIMicrosoft Azure & AI FoundryClouds et places de marché GPU
Centres de données IA privésPour les cabinets d'avocatsPour les cabinets comptablesPlateformes sur siteCloud entièrement sécurisé
OutilsRing PromoterJobshout
Actualités
Blog
Nous contacterSe connecter

Stations de travail IA, logiciels multi-agents IA, infrastructure GPU et solutions d'agents intelligents pour les entreprises modernes.

Nous contacter

Solutions IA

IA appliquéeAgents IAIA générativeAutomatisation par IAMontée en compétence IAOffres IA pour PMELaboratoire IA d'entrepriseIA par secteur

Produits

Tous les produitsCRM & ERP natifs Workstation AIMarketingAgents OpenAIRing PromoterJobshout

Entreprise

À proposEntrepriseCentres de données IAPourquoi WorkstationPartenairesTémoignages clientsTarifsContactTenthMatrix

Ressources

Analyses IANotes techniquesDocumentationOS LinuxPlateformes cloud d'IACentres de données IA privésBlogRechercherPlan du site
Bureau de Belgique
Workstation SRL, Rue Vanderkindere 34, 1180 Uccle, BrusselsBE 0751.518.683Lun - Ven : 9h00 - 18h00 CET
+32 492 45 67 46
Siège social au Royaume-Uni
77-79 Marlowes, Hemel Hempstead HP1 1LFItinéraire : prenez la sortie 20 de la M25, Outer LondonN° d'entreprise: 11641870Lun - Ven : 9h00 - 18h00 GMT
+44 7515 356 146
Bureau de l'Inde
#159 Sector 9, Pocket 1, DDA Flats, 110077 Dwarka, New Delhi
+91 98881 98841
Bureau de Taïwan
3F, No. 2-4, Aisi Rd., Ren'ai Dist., Keelung City 200, Taiwan基隆市仁愛區愛四路2之4號3樓Lun - Ven : 9 h 00 - 18 h 00, heure de Taïwan (GMT+8)
+886 926 892 701+32 492 45 67 46
Bureau en Estonie
Aleksander Puškini tn 30-25, 20306 Narva, EstoniaLun - Ven : 9 h 00 - 18 h 00, heure d’Estonie (EET)
+32 492 45 67 46

© 2026 Workstation AI. Tous droits réservés.

ConfidentialitéSuppression des donnéesCookiesConditions d'utilisationPlan du site web

NVIDIA, le logo NVIDIA, RTX, RTX PRO, Blackwell, CUDA, NVLink, MGX, Omniverse, Spectrum-X, Quantum, BlueField, DGX et HGX sont des marques commerciales et/ou des marques déposées de NVIDIA Corporation aux États-Unis et dans d'autres pays. AMD, le logo AMD Arrow, EPYC, Instinct, Ryzen et Threadripper sont des marques commerciales d'Advanced Micro Devices, Inc. Dell et PowerEdge sont des marques commerciales de Dell Inc. ou de ses filiales. Intel et Xeon sont des marques commerciales d'Intel Corporation ou de ses filiales. AMAX, AceleMax et LiquidMax sont des marques commerciales d'AMAX Engineering Corporation. Tous les autres noms de produits, logos et marques appartiennent à leurs propriétaires respectifs. Nous les utilisons uniquement pour identifier les produits que nous décrivons, fournissons ou prenons en charge ; leur utilisation ne signifie pas que leurs propriétaires nous cautionnent.

Accueil / Notes techniques / Technology
DevOpsCI/CDAI AgentsKubernetesSRE

Workstation AI Shop on Ring Promoter: Orbit, Version-Verified Health and the Path from INT to PROD

Technical guide: shop Helm/CI mapping, Orbit semantics, version-verified health for the AI agent, secrets per ring, promote API, and a production checklist

12 septembre 2026Technology6 min read

This article is the release-engineering deep dive for Workstation AI Shop — the storefront at shop.workstation.co.uk that sells GPU machines and uses a fully autonomous AI Sales Agent to configure baskets. Companion blog (stakeholder-friendly): Promoting the AI Shop through the rings. Control-plane reference: ringpromoter.com.

Watch on YouTube · Product: ringpromoter.com

Ring Promoter Orbit: Workstation AI Shop deploying at INT with version 7cd70f6, rings int test acc prod, TTFB 71ms London, application list sorted by latency
Production Orbit board during a Workstation AI Shop promotion. App label matches the Ring Promoter registry name; blue node = deploy in progress; Earth = PROD / live users.

1. System under promotion

Workstation AI Shop is a Next.js server app (Helm chart shop-chart) on k3s. The browser talks only to the shop; the shop BFF calls OpsAPI in-cluster (http://workstation-opsapi/). Payments use Stripe Checkout with webhooks on OpsAPI. Catalogue freshness is a CronJob (market/sync.mjs). The AI Sales Agent uses Anthropic (ANTHROPIC_MODEL, typically claude-sonnet-5-5, CHAT_MAX_TURNS capped) — if the key is missing, chat shows offline and the rest of the shop still works.

Public hosts

  • INT: https://int-shop.workstation.co.uk
  • PROD: https://shop.workstation.co.uk

Edge path (prod): Cloudflare DNS (DNS-only) → wslproxy → Traefik on the edge node → shop Ingress. Prod has been pinned to a dedicated edge worker where required for latency and certificate locality.

INT TEST ACC PROD flow for Workstation AI Shop

2. Why Ring Promoter owns the hop, not the build

CI still builds and pushes wsw-shop:<sha7>. Ring Promoter answers a different question: is this exact version healthy in ring N, and is it allowed into ring N+1 right now? That matches the product framing on ringpromoter.com:

  1. One ring at a time — promote always targets the next ring.
  2. Source must prove health before anything deploys.
  3. Target must earn health after deploy (retries, then auto-rollback).
  4. Failure undoes itself to the previous recorded version.
  5. Every seed / promote / rollback is written to history.

Shop-specific risk amplifies those rules: the AI agent, market sync and BFF trust keys must move together. A “green Helm job” that leaves an old chat binary answering is a silent sales failure.

3. Reading the Orbit screenshot (release semantics)

From the console capture used in this article:

  • Application name: Workstation AI Shop (human label on the roster and Orbit node).
  • Version in flight: 7cd70f6 (short git SHA — matches image tag convention).
  • Current ring: INT — status Deploying; copy such as “has reached int, N rings to go.”
  • Pipeline legend: int → test → acc → prod across the top of Orbit.
  • Signals: blue node = deployment in progress; green trail = how far the newest version earned; Earth = PROD / live users.
  • SLO breadcrumbs: TTFB ~71ms, location London, GB — useful in the same pane as promotion state so latency regressions are not a separate tool hop.

Lanes (APP × INT/TEST/ACC/PROD matrix) is the complementary view when you need to see whether PROD lags the SHA that already spans ACC.

4. Mapping rings to shop deploy mechanics

Ring Shop mechanism Gate intent
INTGitHub Actions shop-deploy.yaml on main paths under shop/** / chart → deploys intEngineer smoke: /api/health, agent turns, catalogue read
TESTRing Promoter promote INT→TEST (same image SHA)Automated regression; market sync dry observations
ACCPromote TEST→ACCStakeholder UX on AI basket; quote PDF; optional Stripe test mode
PRODPromote ACC→PROD (+ production password). Chart prod may also be workflow_dispatch with TARGET_ENV=prod — Ring Promoter should remain the ledger of what customers runHuman gate; live secrets from vault path workstation-shop/prod/config

Important: treat Helm/workflow deploy as the deployer implementation detail. Ring Promoter remains the system of record for which SHA is current / previous per ring and whether the hop was rolled back.

5. Version-verified health for an AI storefront

Configure the shop health endpoint (or a dedicated readiness field) so Ring Promoter can assert the deployed SHA — via health_version_field or health_version_header as documented on the product site. That catches:

  • Old ReplicaSets still answering while the new Deployment is stuck.
  • Edge/cache serving a previous HTML shell in front of a mixed API.
  • Agent worker and web process skew (chat binary ≠ storefront SHA).

Also verify the source ring reports the same version before promote — so you never promote a SHA that INT already lost.

6. Secrets and AI agent continuity across rings

Per devops/SHOP-DEPLOY.md:

  • Vault: workstation-shop/<env>/config → ESO → shop-secret-<ns>.
  • CI runtime Secret shop-runtime-<ns> may supply ANTHROPIC_API_KEY, OPSAPI_ADMIN_API_KEY, FIRECRAWL_API_KEY.
  • SHOP_BFF_KEY must match OpsAPI’s key for that env or the BFF is treated as anonymous.
  • Prod uses an isolated vault token / SecretStore so rotating INT cannot touch PROD.

Promotion does not invent secrets — but a ring that is “healthy” while the agent is offline is a business-visible defect. Include agent liveness in the health definition you gate on.

7. CI integration shape

After the image is published and INT is seeded (or after the int workflow lands the SHA), promotion is fail-closed HTTP:

curl --fail -sS -X POST "$RP/api/apps/$APP/seed" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d "{\"ring\":\"int\",\"version\":\"$VERSION\"}"

curl --fail -sS -X POST "$RP/api/apps/$APP/promote" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"from_ring":"int"}'

Non-2xx fails the job. 422 means the operation ran, failed health, and rolled back — evidence is in Ring Promoter history / job steps, not only in GitHub Actions logs. Optional ?async=1 returns a job id for live step polling (lock → source health → deploy → target health → record).

8. Auto-promote policy we recommend for the shop

  • ON for early rings when smoke + version health are strong (e.g. INT→TEST).
  • HOLD on ACC so product can sign off AI basket behaviour.
  • Never config-enable auto-promote into PROD — require production password / human promote.
  • Rollbacks remain exempt from the prod password so incidents are not blocked.

Optional gates (maintenance windows, version-specific Go/No-Go, Jira CR, Grafana panels with max_age) run before deploy so a failed gate leaves every ring untouched.

9. Failure modes this design prevents

Failure Without Ring Promoter With protocol
Skipped stagingHotfix workflow_dispatch to prodCannot skip rings by design
Green job, old chatHTTP 200 from previous ReplicaSetVersion-verified health → rollback
Half-updated prodCI timeout aborts mid-rolloutDetached op context + advisory lock
Secret skew INT vs PRODShared token blast radiusIsolated prod SecretStore + ring-scoped config
Unclear “what’s live?”Slack archaeologyOrbit/Lanes + per-ring history

10. Operational checklist (shop release)

  1. Confirm image wsw-shop:<sha7> published.
  2. INT host healthy; agent completes a basket configuration turn; /api/health reports SHA.
  3. Promote through TEST → ACC; re-run market sync observation if catalogue code changed.
  4. Stakeholder sign-off on ACC (AI copy, quote terms, checkout/quote path).
  5. Promote to PROD with production password; verify shop.workstation.co.uk SHA + TTFB.
  6. Leave Orbit open during soak — blue should clear to healthy green on PROD.

11. Related reading

  • Ring Promoter product · Compare vs Argo CD / Kargo / CI
  • Every release earns production · Protocol deep dive
  • QA gates before production
  • Which tool answers which release question
  • In-repo runbook: devops/SHOP-DEPLOY.md

Published by Workstation — GPU storefronts and agentic sales with a promotion control plane you can show in a release review.

Partager cet article

Plus dans Technology

Portables pour développeurs IA : exécuter des LLM en local sous Linux avec des GPU de 16 GB et 24 GB (guide 2026)

Portables pour développeurs IA : exécuter des LLM en local sous Linux avec des GPU de 16 GB et 24 GB (guide 2026)

Ce qui tient sur un GPU RTX 5080 ou RTX 5090 Laptop, le calcul de mémoire, les modèles ouverts qui valent la peine, la configuration d'Ubuntu 26.04 avec la pile NVIDIA, Ollama, llama.cpp et vLLM, les assistants de code, l'affinage LoRA, la sécurité, et quand passer le travail sur un serveur

Lire la suite
Tester des systèmes ML et LLM impossibles à prouver corrects : une boîte à outils Python d'éléments probants

Tester des systèmes ML et LLM impossibles à prouver corrects : une boîte à outils Python d'éléments probants

Des éléments probants à la place des preuves, en environ 400 lignes de Python avec une sortie réelle : métriques par segment avec intervalles bootstrap et contrat de réussite ou d'échec, tests de propriétés et métamorphiques avec Hypothesis, calibration et prédiction conforme, contrats Pydantic et évaluations de non-régression pour les sorties LLM, contrôles de dérive, barrière en CI et vérification exacte d'un petit réseau de neurones

Lire la suite
TorchCurves en pratique : courbes B-spline et de Legendre apprenables dans PyTorch

TorchCurves en pratique : courbes B-spline et de Legendre apprenables dans PyTorch

Installer torchcurves, ajuster et dimensionner une courbe, choisir entre B-spline et Legendre ainsi que l’application d’entrée, construire une courbe de calibration monotone et un modèle additif sur des données tabulaires, comparer avec SciPy, scikit-learn, un KAN et un MLP, exporter les coefficients en JSON, tester et exploiter en production, avec des sorties réelles

Lire la suite