Struct ProviderEntry 

pub struct ProviderEntry {

    pub provider_type: ProviderKind,
    pub name: Option<String>,
    pub model: Option<String>,
    pub base_url: Option<String>,
    pub max_tokens: Option<u32>,
    pub embedding_model: Option<String>,
    pub stt_model: Option<String>,
    pub embed: bool,
    pub default: bool,
    pub thinking: Option<ThinkingConfig>,
    pub server_compaction: bool,
    pub enable_extended_context: bool,
    pub prompt_cache_ttl: Option<CacheTtl>,
    pub reasoning_effort: Option<String>,
    pub thinking_level: Option<GeminiThinkingLevel>,
    pub thinking_budget: Option<i32>,
    pub include_thoughts: Option<bool>,
    pub api_key: Option<String>,
    pub candle: Option<CandleInlineConfig>,
    pub vision_model: Option<String>,
    pub gonka_nodes: Vec<GonkaNode>,
    pub gonka_chain_prefix: Option<String>,
    pub cocoon_client_url: Option<String>,
    pub cocoon_access_hash: Option<String>,
    pub cocoon_health_check: bool,
    pub cocoon_pricing: Option<CocoonPricing>,
    pub instruction_file: Option<PathBuf>,
    pub max_concurrent: Option<u32>,
}

Unified provider entry: one struct replaces CloudLlmConfig, OpenAiConfig, GeminiConfig, OllamaConfig, CompatibleConfig, and OrchestratorProviderConfig.

Provider-specific fields use #[serde(default)] and are ignored by backends that do not use them (flat-union pattern).

Fields

provider_type: ProviderKind

Required: provider backend type.

name: Option<String>

Optional name for multi-provider configs. Auto-generated from type if absent.

model: Option<String>

Model identifier. Required for most types.

base_url: Option<String>

API base URL. Each type has its own default.

max_tokens: Option<u32>

Max output tokens.

embedding_model: Option<String>

Embedding model. When set, this provider supports embed() calls.

stt_model: Option<String>

STT model. When set, this provider supports speech-to-text via the Whisper API or Candle-local inference.

embed: bool

Mark this entry as the embedding provider (handles embed() calls).

default: bool

Mark this entry as the default chat provider (overrides position-based default).

thinking: Option<ThinkingConfig>

server_compaction: bool

enable_extended_context: bool

prompt_cache_ttl: Option<CacheTtl>

Prompt cache TTL variant. None keeps the default ~5-minute ephemeral TTL. Set to "1h" to enable the extended 1-hour TTL (beta, ~2× write cost).

reasoning_effort: Option<String>

thinking_level: Option<GeminiThinkingLevel>

thinking_budget: Option<i32>

include_thoughts: Option<bool>

api_key: Option<String>

candle: Option<CandleInlineConfig>

vision_model: Option<String>

gonka_nodes: Vec<GonkaNode>

Gonka network node pool. Required (non-empty) when type = "gonka".

gonka_chain_prefix: Option<String>

bech32 chain prefix for address encoding. Defaults to "gonka" when omitted.

cocoon_client_url: Option<String>

Cocoon sidecar HTTP URL. Defaults to "http://localhost:10000" when absent.

cocoon_access_hash: Option<String>

Sentinel field for access hash. Leave empty in config; actual value is resolved from the age vault as ZEPH_COCOON_ACCESS_HASH.

cocoon_health_check: bool

Whether to perform a health check against /stats at provider construction time.

cocoon_pricing: Option<CocoonPricing>

Manual per-1K-token pricing for this Cocoon provider.

Cocoon model names (e.g. Qwen/Qwen3-0.6B) are not in the built-in pricing table. When this section is present, the values are registered with CostTracker at startup so that token costs are tracked accurately.

Example TOML:

[llm.providers.cocoon_pricing]
prompt_cents_per_1k = 0.01
completion_cents_per_1k = 0.03

instruction_file: Option<PathBuf>

Provider-specific instruction file.

max_concurrent: Option<u32>

Maximum concurrent LLM calls from orchestrated sub-agents to this provider.

When set, DagScheduler acquires a semaphore permit before dispatching a sub-agent that targets this provider. Dispatch is deferred (using the existing deferral_backoff mechanism) when the semaphore is saturated.

None (default) = unlimited — no admission control applied.

Example (TOML)

[[llm.providers]]
name = "quality"
type = "openai"
model = "gpt-5"
max_concurrent = 3

Implementations

impl ProviderEntry

pub fn effective_name(&self) -> String

Resolve the effective name: explicit name field or type string.

pub fn effective_model(&self) -> String

Resolve the effective model: explicit model field or the provider-type default.

Defaults mirror those used in build_provider_from_entry so that runtime.model_name always reflects the actual model being used rather than the provider type string.

pub fn validate(&self) -> Result<(), ConfigError>

Validate this entry for cross-field consistency.

Errors

Returns ConfigError when a fatal invariant is violated (e.g. compatible provider without a name).

pub fn effective_gonka_chain_prefix(&self) -> &str

Resolve the effective Gonka chain prefix: explicit value or "gonka" default.

Trait Implementations

impl Clone for ProviderEntry

fn clone(&self) -> ProviderEntry

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ProviderEntry

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

Formats the value using the given formatter.

impl Default for ProviderEntry

fn default() -> ProviderEntry

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ProviderEntry

fn deserialize<__D>( __deserializer: __D, ) -> Result<ProviderEntry, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ProviderEntry

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

where __S: Serializer,

Serialize this value into the given Serde serializer.