# Shadow

> Shadow is a French cloud computing company founded in 2015 and chaired by OVHcloud founder Octave
> Klaba. It streams complete high-end Windows desktops (Shadow PC, Shadow PC Pro), sells
> Nextcloud-based storage as Shadow Drive, and sells sovereign GPU compute as Shadow GPU. Shadow GPU
> is the only programmable product, and it is programmable only through the standard OpenStack API:
> a curated OpenStack 2024.1 deployment across seven regions in France, Germany, the United States
> and Canada. Shadow publishes no OpenAPI, no AsyncAPI, no first-party SDK or CLI, no MCP server and
> no A2A agent card.

GENERATED by API Evangelist on 2026-08-17 from the Shadow public developer surface. Shadow does not
publish an llms.txt of its own — https://gpu-instances.shadow.tech/llms.txt returns HTTP 200 but the
body is an HTML language-redirect shell, not an llms.txt.

## What is programmable

- [Shadow GPU (OpenStack API)](https://gpu-instances.shadow.tech/en/): GPU instances (NVIDIA RTX 2000
  Ada, RTX A4500) driven through OpenStack service APIs — Nova (compute), Neutron (networking),
  Octavia (load balancing), Designate (DNS), Cinder (block storage), Glance (images), Placement,
  Barbican (secrets), CloudKitty (usage rating). Dashboards are Skyline and, on FRSBG01, Horizon.
- Authentication is OpenStack Keystone, not OAuth. The identity endpoint answers anonymously with an
  Identity API v3.14 version document at `https://auth.<region>.os.shadow.tech/v3`. Everything past
  that needs a token from a domain-scoped password (`user#domain`) or an Application Credential.
- The rest of the service catalog is returned only inside an authenticated token response, so
  per-service endpoints are not publicly enumerable.

## Identity endpoints (probed 2026-08-17, HTTP 200, Keystone v3.14)

- FRSBG01 Strasbourg, France: https://auth.frsbg01.os.shadow.tech/v3 — TLS certificate EXPIRED
  2026-07-01; a strict-TLS client cannot connect
- FRSBG02 Strasbourg, France (Edge): https://auth.frsbg02.os.shadow.tech/v3
- FRDUN02 Dunkirk, France: https://auth.frdun02.os.shadow.tech/v3
- DEFRA01 Frankfurt, Germany: https://auth.defra01.os.shadow.tech/v3
- USWDC01 Washington DC, USA: https://auth.uswdc01.os.shadow.tech/v3
- USPOR01 Portland, USA: https://auth.uspor01.os.shadow.tech/v3
- CAMTL01 Montreal, Canada: https://auth.camtl01.os.shadow.tech/v3

Each region is an independent deployment with its own dashboard. There is no global endpoint and no
cross-region API. Credentials are region-scoped.

## Docs

- [Shadow GPU documentation](https://gpu-instances.shadow.tech/docs/)
- [Getting started](https://gpu-instances.shadow.tech/docs/getting-started/)
- [Platform architecture (which OpenStack services are in scope)](https://gpu-instances.shadow.tech/docs/getting-started/architecture/)
- [OpenStack CLI configuration](https://gpu-instances.shadow.tech/docs/cli-guide/configuration/)
- [Security measures and Application Credentials](https://gpu-instances.shadow.tech/docs/getting-started/security-measures/)
- [Project and user management](https://gpu-instances.shadow.tech/docs/getting-started/project-management/)
- [User management with the CLI (Adjutant)](https://gpu-instances.shadow.tech/docs/cli-guide/user-management/)
- [Regions and endpoints (portal URLs)](https://gpu-instances.shadow.tech/docs/advanced/regions/)
- [Storage classes (Wood / Silver / Gold)](https://gpu-instances.shadow.tech/docs/advanced/storage-classes/)
- [Instance statuses and the lock mechanism](https://gpu-instances.shadow.tech/docs/advanced/instances-statuses/)
- [Feature limitations](https://gpu-instances.shadow.tech/docs/advanced/limitations/)
- [Kubernetes Cluster API on Shadow](https://gpu-instances.shadow.tech/docs/gpu-paas/cluster-api/)
- [FAQ](https://gpu-instances.shadow.tech/docs/faq/)
- Upstream API reference Shadow points at: https://docs.openstack.org/api-ref/ and
  https://docs.openstack.org/2024.1/user/

## Commercial

- [Pricing and cost simulator](https://gpu-instances.shadow.tech/en/#pricing) — Pay as you go
  (per-minute), Monthly plan (fixed, quota overage billed), Enterprise (negotiated quotas and SLAs).
  RTX 2000 Ada from €0.29/h (≈ €220/month); RTX A4500 from €0.35/h (≈ €250/month). Spot / On Demand /
  Dedicated 24/7 availability models.
- No self-service signup. Access is requested through a "Talk to an expert" form; the page advertises
  24h activation. An autonomous agent cannot onboard itself.
- [Terms of use](https://shadow.tech/us/terms-of-use/)
- [Privacy policy](https://shadow.tech/us/privacy-policy/)

## Operations

- [Status page (Status.io)](https://status.shadow.tech/) — RSS at
  https://status.shadow.tech/pages/5bbcb1b0b0936904c004bbeb/rss. Tracks Shadow PC, Shadow Drive, the
  shared web services and the data centres; it has NO Shadow GPU / OpenStack component.
- [Help centre (Zendesk)](https://support.shadow.tech/hc/en-us)
- No public SLA, no changelog or release-notes page, no deprecation policy, no error-code reference.

## API Evangelist artifacts in this repository

- apis.yml — the APIs.json profile
- authentication/shadow-authentication.yml — Keystone password + Application Credential model, roles,
  identity endpoints
- conventions/shadow-conventions.yml — services in scope, instance state machine, storage classes,
  what is NOT documented
- conformance/shadow-conformance.yml — standards assessed with evidence
- lifecycle/shadow-lifecycle.yml — status page, versioning, SLA posture, retired surfaces
- plans/shadow-plans-pricing.yml — the three billing models and per-configuration prices
- rate-limits/shadow-rate-limits.yml — quota model; no published request-rate limits
- packages/shadow-packages.yml — zero first-party SDKs; the upstream OpenStack packages Shadow's docs
  tell you to install
- cli/shadow-cli.yml — the documented command surface (upstream openstack CLI, not first-party)
- mcp/shadow-mcp.yml — searched absence of any MCP server
- well-known/shadow-well-known.yml — the /.well-known probe, including two false-positive traps
- security/shadow-domain-security.yml — TLS/HSTS/DNS probe

## What Shadow does not publish

- No OpenAPI, Swagger, GraphQL SDL or AsyncAPI document on any host
- No first-party SDK in any language, and no GitHub organisation
- No MCP server (hosted or stdio) and no A2A agent card
- No llms.txt, no /.well-known/api-catalog, no security.txt of its own
- No error-code reference, no request-rate limits, no idempotency contract
- The former developer portal at developers.shadow.tech now 403s and redirects to the help centre;
  business.shadow.tech returns 401