metabase/docs/ai/providers.md

title, summary

title	summary
Supported AI providers	The AI providers Metabase can connect to, the credentials each one needs, and the models each one offers.

Supported AI providers

This documentation was generated from source by running:

clojure -M:ee:doc ai-providers-documentation

To power Metabot, you can connect Metabase to one of the providers below with your own credentials, or let Metabase manage the AI for you. For how to set up a connection, check out AI settings.

You set up a provider in Admin > AI. If you self-host Metabase, you can set a provider's credentials with environment variables instead; each credential below lists the variable that sets it. You don't have to use environment variables. On Metabase Cloud, contact support if you want environment variables set for your instance.

If you want Metabase to support a provider or model that isn't listed here, let us know by submitting a feature request.

Anthropic

• Provider key: anthropic
• Default model: claude-sonnet-4-6
• Model for short tasks like naming a conversation: claude-haiku-4-5-20251001

Supported models:

Model	Model ID	Context window (tokens)
Claude Fable 5	claude-fable-5	1,000,000
Claude Haiku 4.5	claude-haiku-4-5-20251001	200,000
Claude Opus 4.1	claude-opus-4-1-20250805	200,000
Claude Opus 4.5	claude-opus-4-5-20251101	200,000
Claude Opus 4.6	claude-opus-4-6	1,000,000
Claude Opus 4.7	claude-opus-4-7	1,000,000
Claude Opus 4.8	claude-opus-4-8	1,000,000
Claude Opus 5	claude-opus-5	1,000,000
Claude Opus 5.5	claude-opus-5-5	1,000,000
Claude Sonnet 4.5	claude-sonnet-4-5-20250929	200,000
Claude Sonnet 4.6	claude-sonnet-4-6	1,000,000
Claude Sonnet 5	claude-sonnet-5	1,000,000
Claude Sonnet 5.5	claude-sonnet-5-5	1,000,000

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_ANTHROPIC_API_KEY.
• API base URL (advanced). Defaults to https://api.anthropic.com. You can also set it with the environment variable MB_LLM_ANTHROPIC_API_BASE_URL.

OpenAI

• Provider key: openai
• Default model: gpt-5.4
• Model for short tasks like naming a conversation: gpt-5.4-mini

Supported models:

Model	Model ID	Context window (tokens)
GPT-5.4	gpt-5.4	922,000
GPT-5.4 Mini	gpt-5.4-mini	272,000
GPT-5.4 Pro	gpt-5.4-pro	922,000
GPT-5.5	gpt-5.5	922,000
GPT-5.5 Pro	gpt-5.5-pro	922,000
GPT-5.6 Luna	gpt-5.6-luna	922,000
GPT-5.6 Sol	gpt-5.6-sol	922,000
GPT-5.6 Terra	gpt-5.6-terra	922,000
GPT-6 Astra	gpt-6-astra	922,000

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_OPENAI_API_KEY.
• API base URL (advanced). Defaults to https://api.openai.com. You can also set it with the environment variable MB_LLM_OPENAI_API_BASE_URL.

OpenRouter

• Provider key: openrouter
• Default model: anthropic/claude-sonnet-4.6
• Model for short tasks like naming a conversation: anthropic/claude-haiku-4.5

Supported models:

Model	Model ID	Context window (tokens)
Claude Fable 5	anthropic/claude-fable-5	1,000,000
Claude Haiku 4.5	anthropic/claude-haiku-4.5	200,000
Claude Opus 4.1	anthropic/claude-opus-4.1	200,000
Claude Opus 4.5	anthropic/claude-opus-4.5	200,000
Claude Opus 4.6	anthropic/claude-opus-4.6	1,000,000
Claude Opus 4.7	anthropic/claude-opus-4.7	1,000,000
Claude Opus 4.8	anthropic/claude-opus-4.8	1,000,000
Claude Opus 5	anthropic/claude-opus-5	1,000,000
Claude Sonnet 4.5	anthropic/claude-sonnet-4.5	1,000,000
Claude Sonnet 4.6	anthropic/claude-sonnet-4.6	1,000,000
Claude Sonnet 5	anthropic/claude-sonnet-5	1,000,000
DeepSeek V4 Flash 0731	deepseek/deepseek-v4-flash-0731	1,048,576
DeepSeek V4 Pro 0423	deepseek/deepseek-v4-pro	1,048,576
DeepSeek V4 Pro 0813	deepseek/deepseek-v4-pro-0813	1,048,575
Mistral Medium 3.5	mistralai/mistral-medium-3-5	262,144
Kimi K3	moonshotai/kimi-k3	1,048,576
GPT-5.4	openai/gpt-5.4	922,000
GPT-5.4 Mini	openai/gpt-5.4-mini	272,000
GPT-5.4 Pro	openai/gpt-5.4-pro	922,000
GPT-5.5	openai/gpt-5.5	922,000
GPT-5.5 Pro	openai/gpt-5.5-pro	922,000
GPT-5.6 Luna	openai/gpt-5.6-luna	922,000
GPT-5.6 Sol	openai/gpt-5.6-sol	922,000
GPT-5.6 Terra	openai/gpt-5.6-terra	922,000
Qwen3.8 Max 0902	qwen/qwen3.8-max-0902	1,000,000
GLM-5.2	z-ai/glm-5.2	1,048,576
GLM-5.3	z-ai/glm-5.3	1,048,576

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_OPENROUTER_API_KEY.
• API base URL (advanced). Defaults to https://openrouter.ai/api. You can also set it with the environment variable MB_LLM_OPENROUTER_API_BASE_URL.

Mistral

• Provider key: mistral
• Default model: mistral-medium-3-5
• Model for short tasks like naming a conversation: mistral-medium-3-5

Supported models:

Model	Model ID	Context window (tokens)
Mistral Medium 3.5	mistral-medium-3-5	262,144

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_MISTRAL_API_KEY.
• API base URL (advanced). Defaults to https://api.mistral.ai/v1. You can also set it with the environment variable MB_LLM_MISTRAL_API_BASE_URL.

Z.AI

• Provider key: zai
• Default model: glm-5.2
• Model for short tasks like naming a conversation: glm-5.2

Supported models:

Model	Model ID	Context window (tokens)
GLM-5.2	glm-5.2	1,048,576
GLM-5.3	glm-5.3	1,048,576

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_ZAI_API_KEY.
• API base URL (advanced). Defaults to https://api.z.ai/api/paas/v4. You can also set it with the environment variable MB_LLM_ZAI_API_BASE_URL.

Moonshot AI

• Provider key: moonshot
• Default model: kimi-k3
• Model for short tasks like naming a conversation: kimi-k3

Supported models:

Model	Model ID	Context window (tokens)
Kimi K2.6	kimi-k2.6	262,144
Kimi K3	kimi-k3	1,048,576

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_MOONSHOT_API_KEY.
• API base URL (advanced). Point this at the .cn platform to use it instead; keys are not interchangeable between the two. Defaults to https://api.moonshot.ai/v1. You can also set it with the environment variable MB_LLM_MOONSHOT_API_BASE_URL.

DeepSeek

• Provider key: deepseek
• Default model: deepseek-v4-pro
• Model for short tasks like naming a conversation: deepseek-flash

Supported models:

Model	Model ID
DeepSeek Flash	deepseek-flash
DeepSeek V4 Pro	deepseek-v4-pro

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_DEEPSEEK_API_KEY.
• API base URL (advanced). The root both surfaces hang off; leave off any /anthropic or /v1 path. Defaults to https://api.deepseek.com. You can also set it with the environment variable MB_LLM_DEEPSEEK_API_BASE_URL.

xAI

• Provider key: xai
• Default model: grok-4.7
• Model for short tasks like naming a conversation: grok-4.3

Supported models:

Model	Model ID	Context window (tokens)
Grok 4.3	grok-4.3	1,000,000
Grok 4.7	grok-4.7	500,000

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_XAI_API_KEY.
• API base URL (advanced). Defaults to https://api.x.ai/v1. You can also set it with the environment variable MB_LLM_XAI_API_BASE_URL.

Google Gemini Enterprise

• Provider key: google
• Default model: google/gemini-3.5-flash

Supported models:

Model	Model ID
Claude Fable 5	anthropic/claude-fable-5
Claude Haiku 4.5	anthropic/claude-haiku-4-5@20251001
Claude Opus 4.6	anthropic/claude-opus-4-6
Claude Opus 5	anthropic/claude-opus-5
Claude Opus 5.5	anthropic/claude-opus-5-5
Claude Sonnet 4.6	anthropic/claude-sonnet-4-6
Claude Sonnet 5	anthropic/claude-sonnet-5
Claude Sonnet 5.5	anthropic/claude-sonnet-5-5
Gemini 3.5 Flash	google/gemini-3.5-flash
Gemini 3.6 Flash	google/gemini-3.6-flash
Gemini 3.7 Flash	google/gemini-3.7-flash

Credentials:

• Project ID. The Google Cloud project to use. Optional if the service account key provides it. Where do I find this? You can also set it with the environment variable MB_LLM_GOOGLE_PROJECT_ID.
• Location. Optional. Defaults to global. You can also set it with the environment variable MB_LLM_GOOGLE_LOCATION.
• Authentication method (required). Authenticate with a service account key or an OAuth access token. One of: Service account key, OAuth token. Defaults to Service account key.
• Service account key file. Only when Authentication method is Service account key. Upload a service account key file to authenticate with. Where do I find this? You can also set it with the environment variable MB_LLM_GOOGLE_SERVICE_ACCOUNT_KEY.
• OAuth access token. Only when Authentication method is OAuth token. A short-lived token, e.g. the output of gcloud auth print-access-token. Useful for testing. You can also set it with the environment variable MB_LLM_GOOGLE_OAUTH_ACCESS_TOKEN.
• Model Garden endpoint ID. Optional. Use an open model you deployed from Model Garden instead of one Google hosts. Set the location to the region you deployed it to.
• API base URL (advanced). Derived from the location when left at the global host. Defaults to https://aiplatform.googleapis.com. You can also set it with the environment variable MB_LLM_GOOGLE_API_BASE_URL.

Google Gemini Enterprise needs either Service account key file, or OAuth access token and Project ID.

Connect an open model from Model Garden

You can also run Metabot on an open model, like GLM or Llama, that you deployed from Model Garden. Metabase talks to the endpoint the deployment created through its OpenAI-compatible Chat Completions API, with the credentials above, and finds a dedicated endpoint's own DNS name for you.

To connect it, add a Google Gemini Enterprise provider in Admin > AI, enter the endpoint's ID in Model Garden endpoint ID, and set Location to the region you deployed to. That connection serves the endpoint instead of the models above: connecting checks the endpoint, and the model picker offers the endpoint as the connection's only model. To use the models above too, add a second Google Gemini Enterprise provider without an endpoint ID, and pick between them in the model picker.

If you configure the Google connection with environment variables, it has no endpoint ID. Set MB_LLM_METABOT_PROVIDER to google/endpoints/ followed by the endpoint's ID instead.

The credentials need aiplatform.endpoints.get to look up the endpoint and aiplatform.endpoints.predict to run it.

Metabot calls tools and sends long prompts, so deploy the model with tool calling turned on (for vLLM, --enable-auto-tool-choice and a --tool-call-parser that matches the model) and a context length of at least 16,384 tokens, the minimum Metabase also requires of a vLLM connection. Metabase doesn't check either one when you connect.

Microsoft Azure

• Provider key: azure

Supported models:

Whichever model your deployment serves. Microsoft Azure serves the deployments you create, not a fixed catalog, so there's no list to pick from — Metabase works out the model from Model provider and Deployment name instead.

Credentials:

• API key (required). Where do I find this? You can also set it with the environment variable MB_LLM_AZURE_API_KEY.
• API base URL (required). You can also set it with the environment variable MB_LLM_AZURE_API_BASE_URL.
• Model provider (required). Whether your deployment serves an Anthropic or an OpenAI model. One of: OpenAI, Anthropic. Defaults to OpenAI. You can also set it with the environment variable MB_LLM_AZURE_MODEL_FAMILY.
• Deployment name (required). The name of the model deployment on your Azure resource. We recommend naming deployments after the model they serve. You can also set it with the environment variable MB_LLM_AZURE_DEPLOYMENT_NAME.

Amazon Bedrock

• Provider key: bedrock
• Default model: anthropic.claude-opus-4-8
• Model for short tasks like naming a conversation: anthropic.claude-haiku-4-5

Supported models:

Model	Model ID	Context window (tokens)
Claude Fable 5	anthropic.claude-fable-5	1,000,000
Claude Haiku 4.5	anthropic.claude-haiku-4-5	200,000
Claude Opus 4.7	anthropic.claude-opus-4-7	1,000,000
Claude Opus 4.8	anthropic.claude-opus-4-8	1,000,000
Claude Opus 5	anthropic.claude-opus-5	1,000,000
Claude Opus 5.5	anthropic.claude-opus-5-5	1,000,000
Claude Sonnet 5	anthropic.claude-sonnet-5	1,000,000
Claude Sonnet 5.5	anthropic.claude-sonnet-5-5	1,000,000
GPT-5.4	openai.gpt-5.4	272,000
GPT-5.4 (2026-03-05)	openai.gpt-5.4-2026-03-05	272,000
GPT-5.5	openai.gpt-5.5	272,000
GPT-5.5 (2026-04-23)	openai.gpt-5.5-2026-04-23	272,000
GPT-6 Astra	openai.gpt-6-astra	922,000

Credentials:

• Access key ID. Only together with Secret access key. Leave the keys blank to authenticate with the AWS default credentials chain (IRSA, EKS Pod Identity, or instance profile). On Metabase Cloud, Bedrock always authenticates with your own AWS keys. Where do I find this? You can also set it with the environment variable MB_LLM_BEDROCK_ACCESS_KEY_ID.
• Secret access key. Only together with Access key ID. Required on Metabase Cloud. You can also set it with the environment variable MB_LLM_BEDROCK_SECRET_ACCESS_KEY.
• Region. Pick one from the dropdown in Admin > AI. Defaults to us-east-1. You can also set it with the environment variable MB_LLM_BEDROCK_REGION.
• Model ID. Optional. Use an inference profile, or a model that isn't listed for this region, by its ID or ARN.
• Session token (advanced). Only together with Access key ID and Secret access key. Only needed for temporary credentials. You can also set it with the environment variable MB_LLM_BEDROCK_SESSION_TOKEN.

IAM permissions for Bedrock

Metabase talks to Bedrock through the mantle endpoint, https://bedrock-mantle.{region}.api.aws, unless the connection's Model ID sends it to bedrock-runtime (see Use an inference profile). Mantle is a separate IAM namespace with its own actions, so a policy written against the bedrock prefix won't grant access. Metabase lists models and runs conversations, so it needs bedrock-mantle:ListModels and bedrock-mantle:CreateInference.

Here's a least-privilege policy that grants both:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": ["bedrock-mantle:ListModels", "bedrock-mantle:CreateInference"],
      "Resource": "arn:aws:bedrock-mantle:*:*:project/*"
    }
  ]
}

The AWS managed policy AmazonBedrockMantleInferenceAccess also covers both actions (along with permissions Metabase doesn't use).

If Metabase reports "AWS Bedrock credentials lack permission for this model or action", check that your policy uses the bedrock-mantle prefix.

The Bedrock models you can pick depend on the region

The table above lists the models Metabase can use. The Models card only offers the ones Bedrock serves in the connection's region, so what you can pick depends on the region. The mantle catalog has no cross-region inference profiles, so to reach a model that isn't served in your region, use an inference profile.

If the model list is empty or shorter than you expect after connecting:

• Check the region: the Region dropdown lists every AWS region, including regions where Bedrock serves none of these models. The AWS model cards show where each model is available. For example, the GPT models are only served in US regions.
• Check your account's data retention setting: Bedrock marks a model unavailable when your account's data retention mode doesn't meet what that model requires. For example, Claude Fable 5 requires the aws_review data retention mode.

Use an inference profile

To use an inference profile, or a Claude model that isn't listed for your region, enter its ID or ARN in Model ID, like eu.anthropic.claude-sonnet-4-6 or us.anthropic.claude-haiku-4-5-20251001-v1:0. That connection serves this model instead of the models above: connecting checks it by generating a single token, and the model picker offers it as the connection's only model. To use the models above too, add a second Amazon Bedrock provider without a model ID.

Metabase sends inference profile IDs, ARNs, and model IDs with a version suffix like -v1:0 to bedrock-runtime, https://bedrock-runtime.{region}.amazonaws.com, instead of mantle. Through bedrock-runtime, Metabase only talks to Claude models, and it needs bedrock:InvokeModelWithResponseStream on the inference profile and on the foundation model in every region the profile routes to:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": "bedrock:InvokeModelWithResponseStream",
      "Resource": [
        "arn:aws:bedrock:*:*:inference-profile/*",
        "arn:aws:bedrock:*:*:application-inference-profile/*",
        "arn:aws:bedrock:*::foundation-model/*"
      ]
    }
  ]
}

If you configure the Bedrock connection with environment variables, it has no model ID. Set MB_LLM_METABOT_PROVIDER to bedrock/ followed by the ID instead, and set MB_LLM_MINI_MODEL the same way, since short tasks otherwise run on anthropic.claude-haiku-4-5 through mantle.

vLLM

• Provider key: vllm

Supported models:

Metabase lists whichever models your vLLM server is serving, so what you can pick depends on how you started it.

Credentials:

• API base URL (required). Your server's OpenAI-compatible API. It should end in /v1. Metabase must be able to reach it: self-hosted, a server on your private network or on this machine needs MB_LLM_ALLOWED_NETWORKS. You can also set it with the environment variable MB_LLM_VLLM_API_BASE_URL.
• API key. Only needed if you started your server with --api-key. You can also set it with the environment variable MB_LLM_VLLM_API_KEY.

Metabase AI service

• Provider key: metabase
• Default model: anthropic/claude-sonnet-4-6
• Managed by Metabase, so there's nothing to configure. You can only connect one.

Supported models:

Model	Model ID
Claude Sonnet 4.6	anthropic/claude-sonnet-4-6

On Metabase Cloud, you can have Metabase manage the AI for you. Metabase selects benchmarked, cost-effective models, so this is a good option if you don't have a preferred AI provider, or if you want to manage your AI costs through Metabase. You'll be charged based on token usage, on top of your Metabase Cloud subscription. See Pricing.

Metabase authenticates this connection with your instance's license token, so there's no API key to enter.

For how to connect and disconnect, see AI settings.