providers.json Settings Reference
This page lists every key the schema accepts, grouped into one table per provider. See providers.json for what the file is, where it lives, how these keys fit together, and worked examples of the configurations that come up most often.
Each row lists a key in full, such as providers.bedrock.aws.region, so you can search for a key
from your providers.json and find it here verbatim.
A few keys use a placeholder instead:
{name}: the name you give a custom entry underproviders.custom.{header}: a header name incustomHeaders.{model-id}: a model id inmodels.overrides.
[] marks an array element, as in models.custom[].id.
Top-level keys
| Key | Description | Type | Default |
|---|---|---|---|
$schema | Path or URL of the JSON Schema for this file. Editors use it for validation and autocomplete. | string | "https://assistant.posit.co/schemas/providers.schema.json" |
providers | Provider configuration, keyed by built-in provider id, plus the reserved `default` and `custom` keys. | object | — |
version | Version of this config format. Always 1. | 1 | — |
anthropic
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in anthropic provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
bedrock
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in bedrock provider. | object | — |
providers. | AWS connection settings for Bedrock. Credentials are not stored here. | object | — |
providers. | Named profile from your AWS config and credentials files. | string | — |
providers. | AWS region to call, such as `us-east-1`. | string | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
copilot
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in copilot provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
custom
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Providers you declare yourself, keyed by a name of your choosing. | object | — |
providers. | — | object | — |
providers. | Whether an API key is optional for this provider. Set to true when the endpoint authenticates another way, such as a local proxy that adds SSO credentials, so no API key needs to be configured. | boolean | — |
providers. | AWS connection settings for Bedrock. Credentials are not stored here. | object | — |
providers. | Named profile from your AWS config and credentials files. | string | — |
providers. | AWS region to call, such as `us-east-1`. | string | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Google Cloud connection settings for Vertex AI. | object | — |
providers. | Google Cloud region for Vertex AI, such as `us-central1`. | string | — |
providers. | Google Cloud project id for Vertex AI. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Snowflake connection settings. Credentials are not stored here. | object | — |
providers. | Snowflake account identifier. | string | — |
providers. | Which connection from `connections.toml` to use. Ignored in Positron, which uses its own Snowflake sign-in. | string | — |
providers. | Directory holding `connections.toml`, the same thing `SNOWFLAKE_HOME` sets. Point this at a managed directory when your credentials are not in the default location. | string | — |
providers. | Snowflake host name. Set this when the account identifier alone does not resolve to the right host. | string | — |
providers. | Which provider client handles this entry. | "openai-compatible" | "anthropic" | "openai" | "gemini" | "aws" | "snowflake" | "google-vertex" | "ollama" | "lmstudio" | "deepseek" | "openrouter" | "ms-foundry" | "litellm" | "portkey" (required) | — |
databricks
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in databricks provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Databricks workspace settings. | object | — |
providers. | Databricks workspace host. This is not a chat base URL: the serving-endpoint URL is derived from it. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
deepseek
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in deepseek provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
default
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Baseline applied to every provider that does not set the same key itself. | object | — |
providers. | Default enablement for providers that do not set `enabled` themselves. | boolean | true |
gemini
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in gemini provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
google-vertex
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in google-vertex provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Google Cloud connection settings for Vertex AI. | object | — |
providers. | Google Cloud region for Vertex AI, such as `us-central1`. | string | "us-central1" |
providers. | Google Cloud project id for Vertex AI. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
litellm
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in litellm provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
lmstudio
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in lmstudio provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | "http://localhost:1234/v1" |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
ms-foundry
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in ms-foundry provider. | object | — |
providers. | Microsoft Entra ID connection settings for the built-in ms-foundry provider. | object | — |
providers. | How to authenticate: `apikey` (default, back-compat) or `entra` for Microsoft Entra ID sign-in. | "apikey" | "entra" | "apikey" |
providers. | OAuth scope requested when acquiring a Microsoft Entra ID token. | string | "https://cognitiveservices.azure.com/.default" |
providers. | Microsoft Entra ID tenant to authenticate against. Uses the default tenant if unset. | string | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
ollama
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in ollama provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | "http://localhost:11434" |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
openai
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in openai provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
openai-compatible
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in openai-compatible provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
opencode
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in opencode provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | "https://opencode.ai/zen/go/v1" |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Which OpenCode hosted product to use: `go` (subscription, the default) or `zen` (pay-per-request). Selects the API endpoint; an explicit `baseUrl` overrides the selection. | "go" | "zen" | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
openrouter
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in openrouter provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
portkey
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in portkey provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
posit-connect
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in posit-connect provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
positai
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in positai provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | "https://gateway.posit.ai" |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Posit sign-in settings for the built-in positai provider. | object | — |
providers. | OAuth client id used for the device-authorization sign-in flow. | string | "rstudio-ide" |
providers. | Posit host to sign in against. The device-authorization and token URLs are derived from it. | string | "login.posit.cloud" |
providers. | Space-separated OAuth scopes requested at sign-in. | string | "prism" |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
snowflake-cortex
| Key | Description | Type | Default |
|---|---|---|---|
providers. | Configuration for the built-in snowflake-cortex provider. | object | — |
providers. | Base URL for the provider's API. Some providers need a version segment: `https://api.anthropic.com/v1` for Anthropic, `https://api.openai.com/v1` for OpenAI, and `https://generativelanguage.googleapis.com/v1beta` for Gemini. A bare host will fail for those. | string | — |
providers. | Extra HTTP headers sent with each request to this provider, for proxy tenancy or routing markers, on providers whose transport supports custom headers. Do not put credentials or SDK-managed headers such as `Authorization`, `x-api-key`, or `anthropic-version` here. A `User-Agent` entry replaces the default Posit Assistant product identity; the SDK's library tokens are still appended. | object | — |
providers. | Header value. | string | — |
providers. | Whether this provider is available. Set to false to hide it from the model picker. | boolean | true |
providers. | Endpoint URL for a self-hosted provider such as Ollama or LM Studio, for example `http://localhost:11434`. | string | — |
providers. | Base URL overrides for one protocol at a time, keyed by protocol. Use this when every model on a protocol shares a path that the provider's other models do not. | object | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Base URL to use for this protocol. | string | — |
providers. | Which models this provider offers, and how their metadata is filled in. | object | — |
providers. | Model ids to offer. When this list is non-empty it is exclusive: no other model from this provider is offered. | string[] | — |
providers. | Models to declare yourself, for models the provider does not return from discovery. | object[] | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Model id sent to the provider's API. | string (required, non-empty) | — |
providers. | Total context window, in tokens. | integer (required, positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string (required, non-empty) | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean (required) | — |
providers. | Whether the model accepts images returned by a tool result. | boolean (required) | — |
providers. | Whether the model can call tools. | boolean (required) | — |
providers. | Whether the model supports the provider's own web search. | boolean (required) | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | Model ids to hide. Applied after `allow`, so a model listed in both is hidden. | string[] | — |
providers. | Whether to ask the provider for its model list. `auto` discovers models from the provider; `off` offers only the models declared under `models.custom`. | "auto" | "off" | "auto" |
providers. | Metadata patches for models the provider already returns, keyed by model id. Only the fields you set are changed. | object | — |
providers. | — | object | — |
providers. | Base URL for this model only, overriding the provider's base URL. | string | — |
providers. | Model family, used for grouping models and for selecting family-specific system prompt content. | string | — |
providers. | Total context window, in tokens. | integer (positive) | — |
providers. | Maximum number of input tokens accepted in one request. | integer (positive) | — |
providers. | Maximum number of tokens the model may generate in one response. | integer (positive) | — |
providers. | Display name shown in the model picker. | string | — |
providers. | API wire protocol for this model, overriding the provider's protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Media types the model accepts as input, such as `image/png`. | string[] | — |
providers. | Whether the model accepts images as input. | boolean | — |
providers. | Whether the model accepts images returned by a tool result. | boolean | — |
providers. | Whether the model can call tools. | boolean | — |
providers. | Whether the model supports the provider's own web search. | boolean | — |
providers. | Thinking effort levels this model accepts. The set varies by model, for example `off`, `low`, `medium`, `high`, `xhigh`. | string[] | — |
providers. | API wire protocol for this provider. Set this only when the provider does not use its default protocol. | "anthropic-messages" | "openai-chat" | "openai-responses" | "mlflow-responses" | "bedrock-converse" | "google-generative" | — |
providers. | Snowflake connection settings. Credentials are not stored here. | object | — |
providers. | Snowflake account identifier. | string | — |
providers. | Which connection from `connections.toml` to use. Ignored in Positron, which uses its own Snowflake sign-in. | string | — |
providers. | Directory holding `connections.toml`, the same thing `SNOWFLAKE_HOME` sets. Point this at a managed directory when your credentials are not in the default location. | string | — |
providers. | Snowflake host name. Set this when the account identifier alone does not resolve to the right host. | string | — |