Providers

parseforge speaks to eighteen LLM providers through a common interface (RegexBuilder for naming, and the same provider name/model/api-key shape for generation). Install the extra for whichever one you use — see Installation.

At a glance

Provider	Extra	Auth	Default model
Anthropic	anthropic	ANTHROPIC_API_KEY	claude-haiku-4-5-20251001
OpenAI	openai	OPENAI_API_KEY	gpt-5.4-mini
DeepSeek	deepseek	DEEPSEEK_API_KEY	deepseek-v4-flash
Groq	groq	GROQ_API_KEY	llama-3.1-8b-instant
xAI	xai	XAI_API_KEY	grok-3-mini
Together	together	TOGETHER_API_KEY	meta-llama/Llama-3.1-8B-Instruct-Turbo
Fireworks	fireworks	FIREWORKS_API_KEY	accounts/fireworks/models/llama-v3p1-8b-instruct
Perplexity	perplexity	PERPLEXITY_API_KEY	sonar
OpenRouter	openrouter	OPENROUTER_API_KEY	google/gemini-2.5-flash-lite
Moonshot	moonshot	MOONSHOT_API_KEY	moonshot-v1-8k
Cerebras	cerebras	CEREBRAS_API_KEY	llama3.1-8b
Mistral	mistral	MISTRAL_API_KEY	mistral-small-latest
Cohere	cohere	COHERE_API_KEY	command-light
Gemini	gemini	GEMINI_API_KEY	gemini-2.5-flash
Azure OpenAI	azure	see below	(no fixed catalog — deployment name instead)
Vertex AI	vertexai	see below	gemini-2.5-flash
Amazon Bedrock	bedrock	see below	anthropic.claude-haiku-4-5-v1:0
Oracle Cloud (OCI)	oci	see below	meta.llama-3.3-70b-instruct

--api-key/api_key= always falls back to that provider's own env var — only needed on a naming cache miss, since a cache hit never touches the LLM at all. --model/model= defaults to the table above if omitted.

Non-standard auth providers

Fourteen providers fit the plain --api-key/--model shape. Four don't:

Azure OpenAI

No public base_url (it's your own Azure resource) and no fixed model catalog (an account-specific deployment name replaces it):

parseforge name --provider azure --api-key $AZURE_API_KEY \
  --endpoint https://my-resource.openai.azure.com \
  --deployment my-gpt4-deployment \
  show version

Flag	Env var fallback
--api-key	AZURE_API_KEY
--endpoint	AZURE_ENDPOINT
--api-version	AZURE_API_VERSION
--deployment	AZURE_DEPLOYMENT (replaces --model)

Vertex AI

No API key at all — authenticates through GCP's own Application Default Credentials — and needs a project and region instead:

parseforge name --provider vertexai \
  --gcp-project my-project --gcp-location us-central1 \
  show version

Flag	Env var fallback
--gcp-project	VERTEXAI_PROJECT
--gcp-location	VERTEXAI_REGION

Serves the same Gemini model catalog as the native Gemini provider above, just through GCP's separate enterprise billing.

Amazon Bedrock

No API key — authenticates via AWS's own credential chain (env vars, ~/.aws/credentials, or an IAM role) — and only needs a region:

parseforge name --provider bedrock --region us-east-1 show version

Flag	Env var fallback
--region	BEDROCK_REGION, then BEDROCK_DEFAULT_REGION

Oracle Cloud Infrastructure (OCI)

No API key — signs each request cryptographically against local credentials in ~/.oci/config (DEFAULT profile) — and needs both a region and a compartment ID (the OCID of the compartment/tenancy to bill and scope requests to):

parseforge name --provider oci \
  --region us-ashburn-1 --compartment-id ocid1.compartment.oc1..xxxx \
  show version

Flag	Env var fallback
--region	OCI_REGION, then whatever region is already set in ~/.oci/config
--compartment-id	OCI_COMPARTMENT_ID

Naming vs. generation providers

run/trial split naming and generation into separate provider configs (--naming-* vs. plain --*/generation flags) — they default to the same provider, but you can point them at two different ones, e.g. a cheap fast model for naming (cached after the first call, so cost matters less) and a stronger model for generation. Every flag above has a --naming- equivalent. See the CLI Guide for the full flag list.