Struct UsageEntry 

pub struct UsageEntry {

    pub session_id: String,
    pub timestamp_ms: u64,
    pub model: String,
    pub input_tokens: u64,
    pub output_tokens: u64,
    pub cache_read_tokens: u64,
    pub cache_write_tokens: u64,
    pub cache_write_1h_tokens: u64,
    pub web_search_requests: u64,
    pub speed: Option<String>,
    pub tool_uses: Vec<String>,
    pub text_chars: u32,
    pub activity: Activity,
    pub entrypoint: Option<String>,
}

Flat row representing a single completed assistant turn.

One entry per unique message.id. Claude Code writes multiple JSONL entries per assistant message as the stream progresses (typically 2-3), all sharing the same id and final usage numbers — dedup keeps only the first occurrence.

Token field semantics (inherited from Anthropic’s API):

• input_tokens: fresh tokens in this turn’s prompt that were NOT served from cache. Small when conversation is reused, large on cache misses.
• cache_read_tokens: tokens served from the prompt cache (cheap).
• cache_write_tokens: tokens written into the prompt cache (~25% premium).
• output_tokens: tokens generated by the model.

“Total context sent” for a turn = input + cache_read + cache_write.

Fields

session_id: String

timestamp_ms: u64

Unix epoch ms from the JSONL timestamp field. 0 if unparseable.

model: String

input_tokens: u64

output_tokens: u64

cache_read_tokens: u64

cache_write_tokens: u64

Total cache-write (creation) tokens for this turn: max(legacy flat cache_creation_input_tokens, ephemeral_5m + ephemeral_1h). The 1-hour subset is broken out separately in cache_write_1h_tokens for pricing; this total still drives cache-hit % and token displays.

cache_write_1h_tokens: u64

The 1-hour-TTL portion of cache_write_tokens (from usage.cache_creation.ephemeral_1h_input_tokens). Billed at 2× input vs the 5-minute tier’s 1.25×. The 5-minute portion is the remainder.

web_search_requests: u64

Anthropic web-search server-tool requests this turn (usage.server_tool_use.web_search_requests). Billed flat at $0.01 each.

speed: Option<String>

usage.speed: "standard" or "fast". Some("fast") applies the fast-mode cost multiplier (see fast_multiplier_for_model). None on rows that predate the field.

tool_uses: Vec<String>

Names of tools this turn invoked (from tool_use content blocks). Empty for pure-text turns. Duplicates preserved when the model called the same tool multiple times in one turn.

text_chars: u32

Sum of text content block character counts for this turn. Zero means a silent tool-only turn; non-zero means the model said something.

activity: Activity

Classified activity category (Coding, Debugging, Exploration, …). See the crate::activity module for the decision tree.

entrypoint: Option<String>

JSONL entrypoint field captured from the first row of each message id. "cli" = interactive Claude Code session. "sdk-cli" = headless claude -p spawn (any Agent SDK consumer lands here). None for legacy rows that predate the field. Useful for distinguishing interactive vs. headless usage; also load-bearing once Anthropic splits Agent SDK billing from interactive plans (announced June 2026).

Trait Implementations

impl Clone for UsageEntry

fn clone(&self) -> UsageEntry

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for UsageEntry

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for UsageEntry

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for UsageEntry

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

where __S: Serializer,

Serialize this value into the given Serde serializer.