config

Provider configuration shared by all provider implementations.

Model builders expose this through their configure function, e.g. anthropic.sonnet_5_5(key) |> anthropic.configure(config.timeout(_, 120_000)).

Types

ProviderConfig

</>

Configuration for creating a provider.

pub type ProviderConfig {
  ProviderConfig(
    api_key: String,
    base_url: String,
    model: String,
    headers: List(#(String, String)),
    timeout_ms: Int,
    max_tokens_field: String,
    max_retries: Int,
    initial_retry_delay_ms: Int,
    send_reasoning_content: Bool,
    stream_usage: Bool,
    extra_body: List(#(String, json.Json)),
  )
}

Constructors

```
ProviderConfig(
  api_key: String,
  base_url: String,
  model: String,
  headers: List(#(String, String)),
  timeout_ms: Int,
  max_tokens_field: String,
  max_retries: Int,
  initial_retry_delay_ms: Int,
  send_reasoning_content: Bool,
  stream_usage: Bool,
  extra_body: List(#(String, json.Json)),
)
```
Arguments

api_key

API key for authentication

base_url

Base URL for the API (no trailing slash)

model

Model ID to use

headers

Additional HTTP headers

timeout_ms

Request timeout in milliseconds

max_tokens_field

Field name for max tokens parameter (e.g. “max_completion_tokens” or “max_tokens”)

max_retries

Maximum retries for transient HTTP errors (429, 5xx)

initial_retry_delay_ms

Initial retry delay in milliseconds (doubles each retry, with jitter)

send_reasoning_content

OpenAI-compatible APIs only: send the model’s earlier reasoning back as reasoning_content (DeepSeek and Kimi require it in tool loops).

stream_usage

OpenAI-compatible APIs only: ask for token usage in streaming responses with stream_options.include_usage.

extra_body

OpenAI-compatible APIs only: extra top-level request fields for provider-specific options glean has no setting for, such as DeepSeek’s thinking or Qwen’s enable_thinking. They are added after glean’s own fields.

Values

base_url

</>

pub fn base_url(
  config: ProviderConfig,
  url: String,
) -> ProviderConfig

Set the base URL.

extra_body_field

</>

pub fn extra_body_field(
  config: ProviderConfig,
  name: String,
  value: json.Json,
) -> ProviderConfig

Add an extra top-level request field (OpenAI-compatible providers only). Setting the same name twice keeps the last value.

headers

</>

pub fn headers(
  config: ProviderConfig,
  headers: List(#(String, String)),
) -> ProviderConfig

Add extra headers.

initial_retry_delay

</>

pub fn initial_retry_delay(
  config: ProviderConfig,
  ms: Int,
) -> ProviderConfig

Set initial retry delay in milliseconds (doubles each retry).

max_retries

</>

pub fn max_retries(
  config: ProviderConfig,
  retries: Int,
) -> ProviderConfig

Set max retries for transient HTTP errors (429, 5xx).

max_tokens_field

</>

pub fn max_tokens_field(
  config: ProviderConfig,
  field: String,
) -> ProviderConfig

Set the field name used for max tokens in API requests. Defaults to “max_completion_tokens”. Use “max_tokens” for providers that don’t support the newer field name (e.g. Mistral).

new

</>

pub fn new(
  api_key api_key: String,
  model model: String,
) -> ProviderConfig

Create a config with defaults.

send_reasoning_content

</>

pub fn send_reasoning_content(
  config: ProviderConfig,
  enabled: Bool,
) -> ProviderConfig

Send earlier reasoning back as reasoning_content (OpenAI-compatible providers only).

stream_usage

</>

pub fn stream_usage(
  config: ProviderConfig,
  enabled: Bool,
) -> ProviderConfig

Ask for token usage in streaming responses (OpenAI-compatible providers only).

timeout

</>

pub fn timeout(config: ProviderConfig, ms: Int) -> ProviderConfig

Set the request timeout in milliseconds.

Constructors

Arguments