A virtual service with no container. It holds the switches, API keys and model lists for OpenAI, Anthropic and OpenRouter. All requests to these providers go through LiteLLM.

An enabled provider with a key adds model routes to LiteLLM. Applications call those models through LiteLLM like any other model. Routing is described in the LiteLLM README.

• Category: llm
• Kind: virtual
• Tracks: all, data-eng, gen-ai-creative, gen-ai-eng, gen-ai-rag, ml-eng, trading

No port and no Kong route. Call the models through LiteLLM.

Provider	Switch (default disabled)	API key	Model list
OpenAI	CLOUD_OPENAI_SOURCE	OPENAI_API_KEY	OPENAI_USER_MODELS
Anthropic	CLOUD_ANTHROPIC_SOURCE	ANTHROPIC_API_KEY	ANTHROPIC_USER_MODELS
OpenRouter	CLOUD_OPENROUTER_SOURCE	OPENROUTER_API_KEY	OPENROUTER_USER_MODELS

A provider adds LiteLLM routes only when its switch is enabled and its key is set. Each *_USER_MODELS value is a comma-separated list of the model names the wizard activated. The bootstrapper derives LITELLM_OPENAI_ENABLED, LITELLM_ANTHROPIC_ENABLED, LITELLM_OPENROUTER_ENABLED and LITELLM_ENABLED_PROVIDERS from the three switches.

• Required dependency: litellm
• Optional dependencies: none
• Runtime calls: none (LiteLLM calls the provider APIs)

SOURCE Variable	Default	Values
CLOUD_OPENAI_SOURCE	disabled	enabled, disabled
CLOUD_ANTHROPIC_SOURCE	disabled	enabled, disabled
CLOUD_OPENROUTER_SOURCE	disabled	enabled, disabled

LiteLLM is the only consumer. Disabling a provider removes its routes when the LiteLLM configuration is next generated.

The wizard asks for each provider on every track. CLI flags:

• --cloud-openai-source enabled|disabled (and --cloud-anthropic-source, --cloud-openrouter-source).
• --openai-api-key <key> saves OPENAI_API_KEY to .env and enables the provider. --anthropic-api-key and --openrouter-api-key work the same way.
• --openai-models <names> saves OPENAI_USER_MODELS. --anthropic-models and --openrouter-models work the same way.

• Service manifest: services/cloud-providers/service.yml
• LiteLLM integration: LiteLLM README

Support tier: experimental — Capability contract declared; no cited cold-start, workflow, or upgrade qualification run yet (evidence at v0.1.0).

Capability	Status	Verification	Notes
Virtual cloud-provider selection	supported	tested	Atlas independently enables OpenAI, Anthropic, and OpenRouter and exposes their selected models through the always-on LiteLLM gateway without running a provider container.
Key-gated model registration	supported	tested	A provider contributes LiteLLM rows only when its source is enabled and its server-side API key is non-empty; disabling it removes those routes on regeneration.
Uncatalogued cloud model metadata	partial	tested	User-selected names outside the curated catalogs are routable, but Atlas synthesizes generic capability metadata and cannot infer provider-specific limits or modalities.
Live cloud completion validation	not-supported	untested	Atlas statically tests selection, key isolation, and rendered routing but performs no live credential, entitlement, model-availability, or provider certification. Successful requests still depend on external accounts and APIs.
Direct cloud-provider service endpoint	not-supported	documented	This virtual manifest owns configuration only; applications must use LiteLLM because Atlas creates no cloud-provider container, port, or Kong route.

No upstream calls.

Service	Category
litellm	llm

Open the full-size diagram for a full-screen view.

No high-confidence opportunities identified.

No high-confidence opportunities identified.

No high-confidence opportunities identified.