Struct ProviderConfig 

pub struct ProviderConfig {

    pub provider_type: String,
    pub api_key: Option<String>,
    pub model: String,
    pub base_url: Option<String>,
    pub system_prompt: Option<String>,
    pub user_agent: Option<String>,
    pub context_window: usize,
    pub max_tokens: Option<usize>,
    pub thinking_type: Option<String>,
    pub thinking_keep: Option<String>,
    pub reasoning_history: Option<String>,
    pub thinking_enabled: Option<bool>,
    pub thinking_budget: Option<u32>,
    pub skip_tls_verify: bool,
    pub ephemeral: bool,
}

Fields

provider_type: String

api_key: Option<String>

model: String

base_url: Option<String>

system_prompt: Option<String>

user_agent: Option<String>

Override User-Agent for this provider (useful when the upstream blocks generic UAs). Defaults to atomcode/<version> if not set.

context_window: usize

Maximum tokens to use for context (system prompt + messages). The windowing algorithm fits messages within this budget, condensing old tool results to save space. Defaults vary by provider type; use default_context_window_for after deserialization.

max_tokens: Option<usize>

Maximum tokens the model can output per response. Larger values allow batching multiple write_file calls in one turn. If not set, defaults to context_window / 4.

thinking_type: Option<String>

Kimi K2.5 / K2.6 thinking control — emitted as thinking.type in the request body. "enabled" | "disabled". K2-thinking is always on and ignores this. Unset = don’t forward the field.

thinking_keep: Option<String>

Kimi K2.6 Preserved Thinking — emitted as thinking.keep in the request body. "all" to have the server reprocess historical reasoning_content (more expensive). Unset = default behavior.

reasoning_history: Option<String>

Override the history-echo policy for reasoning_content on historical assistant tool_call messages. "include" = always echo the stored reasoning back (required by Moonshot Kimi K2 thinking, DeepSeek V4 thinking mode); "exclude" = never echo (required by DeepSeek V3 R1, safe default for plain OpenAI). Unset = use the built-in auto-detect heuristic based on model name / base_url. Lets users work around new provider quirks without a code change.

thinking_enabled: Option<bool>

Whether extended thinking is enabled for this provider. For Claude: sends thinking.type = "enabled" in request body. Default: not set (thinking disabled).

thinking_budget: Option<u32>

Maximum tokens allocated to the thinking phase. Claude: sent as thinking.budget_tokens. Default: 10000 when thinking is enabled.

skip_tls_verify: bool

Skip TLS certificate verification for this provider. Useful for self-signed certificates in enterprise/internal environments. Default: false (TLS verification enabled). WARNING: Setting this to true reduces security by accepting any certificate.

ephemeral: bool

If true, this provider was added at runtime (e.g. OAuth /login) and should NOT be persisted to config.toml on save.

Implementations

impl ProviderConfig

pub fn accepts_images(&self) -> bool

True if this provider’s active model can accept image inputs. Driven entirely by the model-name heuristic in provider::model_name_suggests_vision — if a future model isn’t recognised, extend the heuristic rather than threading a per-provider config flag (no user-facing knob to discover).

Trait Implementations

impl Clone for ProviderConfig

fn clone(&self) -> ProviderConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ProviderConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ProviderConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ProviderConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.