Struct StreamHandlingOutput 

pub struct StreamHandlingOutput {
    pub response_id: Option<String>,
    pub content: String,
    pub reasoning_content: String,
    pub reasoning_signature: Option<String>,
    pub token_count: usize,
    pub tool_calls: Vec<ToolCall>,
    pub output_tokens: u64,
    pub thinking_tokens: u64,
    pub cache_creation_input_tokens: u64,
    pub cache_read_input_tokens: u64,
    pub provider_usage: Option<ProviderUsageSnapshot>,
    pub input_tokens: u64,
}

Fields

response_id: Option<String>

content: String

reasoning_content: String

reasoning_signature: Option<String>

Provider-minted signature covering reasoning_content, present only when the turn’s thinking arrived as exactly one signed Anthropic thinking block — see bamboo_llm::LLMChunk::ReasoningSignature (#520).

token_count: usize

tool_calls: Vec<ToolCall>

output_tokens: u64

thinking_tokens: u64

cache_creation_input_tokens: u64

cache_read_input_tokens: u64

provider_usage: Option<ProviderUsageSnapshot>

Merged authoritative provider snapshot, when at least one provider-usage chunk was observed. Repeated cumulative snapshots are idempotent, absent fields do not erase known values, and explicit zeros remain Some(0).

input_tokens: u64

Normalized non-cached (“fresh”) input, disjoint from the adjacent cache counters. When a provider total is available this is derived from that total with a saturating, cache-subset policy; the raw total remains in Self::provider_usage.