Providers

Praetor supports 25 providers. Set at least one API key and Praetor detects which provider you configured.

The registry is the single source of truth: registerProvider() calls in packages/core/src/runtime/providers/providerRegistry.ts determine the auto-detection order, the environment variable map, the default base URLs, the default models, the display names, and the API type. Adding a provider is one registerProvider() call.

Auto-detection order

Order matters. Local providers are listed first, so auto-detection prefers them when configured; cloud providers follow, with OpenAI last as the most reliable fallback.

The 25 providers

#	--provider value	Display name	Env variable	Default model
1	ollama	Ollama (Local)	OLLAMA_API_KEY	llama3.2
2	vllm	vLLM (Local)	VLLM_API_KEY	meta-llama/Llama-3.2-3B-Instruct
3	deepinfra	DeepInfra	DEEPINFRA_API_KEY	meta-llama/Llama-3.3-70B-Instruct-Turbo
4	anyscale	Anyscale	ANYSCALE_API_KEY	meta-llama/Llama-3.3-70B-Instruct
5	replicate	Replicate	REPLICATE_API_TOKEN	meta/meta-llama-3.3-70b-instruct
6	cohere	Cohere	CO_API_KEY	command-a-plus-05-2026
7	mistral	Mistral AI	MISTRAL_API_KEY	mistral-large-latest
8	groq	Groq	GROQ_API_KEY	llama-3.3-70b-versatile
9	together	Together AI	TOGETHER_API_KEY	meta-llama/Llama-3.3-70B-Instruct-Turbo
10	perplexity	Perplexity	PERPLEXITY_API_KEY	sonar-pro
11	fireworks	Fireworks AI	FIREWORKS_API_KEY	accounts/fireworks/models/llama-v3p3-70b-instruct
12	xiaomi	Xiaomi MiMo	XIAOMI_API_KEY	mimo-v2-flash
13	mimo	MiMo	MIMO_API_KEY	mimo-v2.5
14	deepseek	DeepSeek	DEEPSEEK_API_KEY	deepseek-v4-flash
15	glm	GLM (Zhipu AI)	ZHIPU_API_KEY	glm-4.7
16	xai	xAI (Grok)	XAI_API_KEY	grok-2-latest
17	bedrock	AWS Bedrock	AWS_ACCESS_KEY_ID	anthropic.claude-3-5-sonnet-20241022-v2:0
18	agnes	Agnes AI	AGNES_API_KEY	agnes-2.5-flash
19	stepfun	StepFun	STEPFUN_API_KEY	step-3.7-flash
20	minimax	MiniMax	MINIMAX_API_KEY	MiniMax-M3
21	openrouter	OpenRouter	OPENROUTER_API_KEY	openai/gpt-4o-mini
22	anthropic	Anthropic	ANTHROPIC_API_KEY	claude-sonnet-4-6
23	google	Google Gemini	GOOGLE_API_KEY	gemini-2.0-flash
24	azure	Azure OpenAI	AZURE_OPENAI_API_KEY	gpt-4o
25	openai	OpenAI	OPENAI_API_KEY	gpt-4o

Environment variables that are not obvious

Several providers use a variable you would not guess from the name:

• Cohere uses CO_API_KEY (with COHERE_API_KEY accepted as a fallback).
• GLM / Zhipu uses ZHIPU_API_KEY, not GLM_API_KEY.
• Replicate uses REPLICATE_API_TOKEN (with REPLICATE_API_KEY accepted as a fallback).
• AWS Bedrock is selected by AWS_ACCESS_KEY_ID presence and also needs AWS_SECRET_ACCESS_KEY and AWS_REGION.

Overrides

Each provider accepts the same three override shapes:

Pattern	Example
Base URL	<PROVIDER>_BASE_URL — e.g. OPENAI_BASE_URL, AZURE_OPENAI_BASE_URL
Model	<PROVIDER>_MODEL — e.g. OPENAI_MODEL, AZURE_OPENAI_MODEL
Host (local runtimes)	OLLAMA_HOST, VLLM_BASE_URL

Azure OpenAI additionally uses AZURE_OPENAI_API_VERSION.

Failover

Providers form a configurable auto-failover chain. praetor init scans for configured keys, tests connectivity, measures latency, and writes the resulting chain to .commander.json. Per-provider circuit breakers track error rates and open after consecutive failures, moving traffic to the next provider in the chain.

Model tiers

PRAETOR_MIN_SENSITIVE_MODEL_TIER sets a floor for the model tier used on sensitive steps. Tiers are eco, standard, power, and consensus; the default is standard. A step that needs a higher tier will not be routed to a lower one.

Credential handling

Provider keys are resolved vault-first, then environment. The encrypted secrets vault (PRAETOR_MASTER_KEY) uses AES-256-GCM with HKDF-SHA-256 derivation. See Security and privacy.