Providers

The provider list is fixed in code, in src/vexy_co_model_catalog/core/provider.py. It holds 40 providers. You cannot add or remove providers from the command line; vmc providers add and vmc providers remove only print a notice.

Provider types

Each provider has one of three types. The type decides which URL is requested and how the API key is sent.

Type	Request URL	Authentication header
openai	<base URL>/models	Authorization: Bearer <key>
anthropic	<base URL>/models	x-api-key: <key> and anthropic-version: 2023-06-01
url	<base URL> itself	none

Every request also carries User-Agent: ModelDumper/1.0.

How the base URL is chosen

Each provider names two environment variables: one for the API key and one for the base URL.

1. When the package loads the provider table, it sets each base-URL variable to its default value, unless that variable is already set in your environment.
2. When a request is built, the tool reads the base-URL variable.

So you can point any provider at another endpoint by exporting its base-URL variable before you run vmc:

export LMSTUDIO_API_OPENAI="http://localhost:1234/v1"
vmc fetch lmstudio

The tool then requests http://localhost:1234/v1/models.

Provider table

Name	Type	API key variable	Base-URL variable	Default base URL
aihorde	openai	AIHORDE_KEY	AIHORDE_API_OPENAI	https://oai.aihorde.net/v1
anthropic	anthropic	ANTHROPIC_API_KEY	ANTHROPIC_API_OPENAI	https://api.anthropic.com/v1
atlascloud	openai	ATLASCLOUD_API_KEY	ATLASCLOUD_API_OPENAI	(none, see note)
avian	openai	AVIAN_API_KEY	AVIAN_API_OPENAI	https://api.avian.io/v1
baseten	openai	BASETEN_API_KEY	BASETEN_API_OPENAI	https://inference.baseten.co/v1
cerebras	openai	CEREBRAS_API_KEY	CEREBRAS_API_OPENAI	https://api.cerebras.ai/v1
chutes	openai	CHUTES_API_KEY	CHUTES_API_OPENAI	https://llm.chutes.ai/v1
deepinfra	openai	DEEPINFRA_API_KEY	DEEPINFRA_API_OPENAI	https://api.deepinfra.com/v1/openai
deepseek	openai	DEEPSEEK_API_KEY	DEEPSEEK_API_OPENAI	https://api.deepseek.com/v1
enfer	openai	ENFER_API_KEY	ENFER_API_OPENAI	https://api.enfer.ai/v1
featherless	openai	FEATHERLESS_API_KEY	FEATHERLESS_API_OPENAI	https://api.featherless.ai/v1
fireworks	openai	FIREWORKS_API_KEY	FIREWORKS_API_OPENAI	https://api.fireworks.ai/inference/v1
friendli	openai	FRIENDLI_TOKEN	FRIENDLI_API_OPENAI	https://api.friendli.ai/serverless/v1
gemini	openai	GOOGLE_API_KEY	GEMINI_API_OPENAI	https://generativelanguage.googleapis.com/v1beta/openai
groq	openai	GROQ_API_KEY	GROQ_API_OPENAI	https://api.groq.com/openai/v1
huggingface	openai	HUGGINGFACEHUB_API_TOKEN	HUGGINGFACE_API_OPENAI	https://router.huggingface.co/v1
hyperbolic	openai	HYPERBOLIC_API_KEY	HYPERBOLIC_API_OPENAI	https://api.hyperbolic.xyz/v1
inference	openai	INFERENCENET_API_KEY	INFERENCENET_API_OPENAI	https://api.inference.net/v1
infermatic	openai	INFERMATIC_API_KEY	INFERMATIC_API_OPENAI	https://api.totalgpt.ai/v1
litellm	url	(none)	LITELLM_API_OPENAI	https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json
llm7	openai	LLM7_API_KEY	LLM7_API_OPENAI	https://api.llm7.io/v1
lmstudio	openai	LMSTUDIO_API_KEY	LMSTUDIO_API_OPENAI	http://othello.local:1234/v1
mancer	openai	MANCER_API_KEY	MANCER_API_OPENAI	https://neuro.mancer.tech/oai/v1
mistral	openai	MISTRAL_API_KEY	MISTRAL_API_OPENAI	https://api.mistral.ai/v1
moonshot	openai	MOONSHOT_API_KEY	MOONSHOT_API_OPENAI	https://api.moonshot.ai/v1
morphllm	openai	MORPHLLM_API_KEY	MORPHLLM_API_OPENAI	https://api.morphllm.com/v1
nebius	openai	NEBIUS_API_KEY	NEBIUS_API_OPENAI	https://api.studio.nebius.com/v1
nineteenai	openai	NINETEENAI_API_KEY	NINETEENAI_API_OPENAI	https://api.nineteen.ai/v1
novita	openai	NOVITA_API_KEY	NOVITA_API_OPENAI	https://api.novita.ai/v3/openai
openai	openai	OPENAI_API_KEY	OPENAI_API_OPENAI	https://api.openai.com/v1
openrouter	openai	OPENROUTER_API_KEY	OPENROUTER_API_OPENAI	https://openrouter.ai/api/v1
parasail	openai	PARASAIL_API_KEY	PARASAIL_API_OPENAI	https://api.parasail.io/v1
poe	openai	POE_API_KEY	POE_API_OPENAI	https://api.poe.com/v1
pollinations	openai	POLLINATIONS_API_KEY	POLLINATIONS_API_OPENAI	https://text.pollinations.ai/openai
redpill	openai	REDPILL_API_KEY	REDPILL_API_OPENAI	https://api.redpill.ai/v1
sambanova	openai	SAMBANOVA_API_KEY	SAMBANOVA_API_OPENAI	https://api.sambanova.ai/v1
targon	openai	TARGON_API_KEY	TARGON_API_OPENAI	https://api.targon.com/v1
togetherai	openai	TOGETHERAI_API_KEY	TOGETHERAI_API_OPENAI	https://api.together.xyz/v1
xai	openai	XAI_API_KEY	XAI_API_OPENAI	https://api.x.ai/v1
zai	openai	ZAI_API_KEY	ZAI_API_OPENAI	https://api.z.ai/api/paas/v4

atlascloud has no default base URL

In the default URL table, the atlascloud entry is keyed ATLASCLOUD_API_KEY instead of ATLASCLOUD_API_OPENAI. As a result the package sets no default for ATLASCLOUD_API_OPENAI, and it also writes the URL into ATLASCLOUD_API_KEY when that variable is not already set. Set both variables yourself to use this provider:

export ATLASCLOUD_API_KEY="..."
export ATLASCLOUD_API_OPENAI="https://api.atlascloud.ai/v1"

lmstudio points at a local host name

The default for lmstudio is http://othello.local:1234/v1, a machine on the author's network. Plain http:// is only accepted for localhost and 127.0.0.1, so the default is rejected by the URL check. Set LMSTUDIO_API_OPENAI to your own server, for example http://localhost:1234/v1.

Providers with special handling

litellm

A url provider. Its base URL is the JSON file model_prices_and_context_window.json in the LiteLLM repository, so no key is needed. The response is an object keyed by model ID; the sample_spec key is left out of the model ID list.

chutes

After the normal /models request, the tool also requests https://api.chutes.ai/chutes/?include_public=true&include_schemas=false and merges the extra fields into the matching model entries. If the second request fails, the tool logs a warning and keeps the first response.

lmstudio

The only provider in the table allowed to use plain http://, and only for localhost or 127.0.0.1.

URL check

Before each request the tool checks the URL. It must use https://; plain http:// is accepted only for localhost and 127.0.0.1, and only for lmstudio. Host names that contain patterns such as metadata, internal, private, 169.254 or 10. are rejected. A rejected URL fails the provider with a message that starts with SECURITY: URL failed security validation.

Response formats

The tool accepts three response shapes and extracts model IDs from each:

Shape	Example	Model IDs taken from
OpenAI style	{"data": [{"id": "gpt-4o"}, …]}	each data[].id; data is sorted by id
Object keyed by ID	{"gpt-4o": {…}, …}	the keys, except sample_spec; keys are sorted
Plain array	[{"id": "gpt-4o"}, …]	each id; the array is sorted by id

Check a provider

vmc providers show groq      # type, key variable, base URL, failure stats
vmc validate groq            # is the key set? is a base URL available?
vmc provider_health groq     # does the endpoint answer?