Struct AtomicTokenCounter 

pub struct AtomicTokenCounter { /* private fields */ }

Process-wide token counter. Cheap record on the hot path; richer per-model stats guarded by an async-compatible mutex and accessed best-effort via try_lock.

Prefer the super::super::TOKEN_USAGE singleton — constructing your own instance is only useful in tests.

Examples

use codetether_agent::telemetry::AtomicTokenCounter;

let c = AtomicTokenCounter::new();
c.record(100, 50);
let (p, o, t) = c.get();
assert_eq!((p, o, t), (100, 50, 150));

Implementations

impl AtomicTokenCounter

pub fn record_model_usage(&self, model: &str, prompt: u64, completion: u64)

Record per-model usage without prompt-cache data.

pub fn record_model_usage_with_cache( &self, model: &str, prompt: u64, completion: u64, cache_read: u64, cache_write: u64, )

Record a completion’s usage including prompt-cache read/write tokens.

prompt is the non-cached input tokens billed at full price. cache_read is billed at 10% of input price; cache_write at 125% on Anthropic / Bedrock. See crate::provider::pricing.

Examples

use codetether_agent::telemetry::AtomicTokenCounter;

let c = AtomicTokenCounter::new();
c.record_model_usage_with_cache("claude-sonnet-4", 500, 200, 1_000, 0);
assert_eq!(c.cache_usage_for("claude-sonnet-4"), (1_000, 0));
assert_eq!(c.last_prompt_tokens_for("claude-sonnet-4"), Some(1_500));

pub fn cache_usage_for(&self, model: &str) -> (u64, u64)

Cumulative (cache_read, cache_write) token counts for a model. Returns (0, 0) if the map is contended or the model is unknown.

pub fn last_prompt_tokens_for(&self, model: &str) -> Option<u64>

Most recent turn’s full prompt token count (including cache) for a model.

This reflects the current in-flight context size (what the next request will send), not cumulative lifetime tokens. Returns None if no completion has been recorded for the model yet.

impl AtomicTokenCounter

pub fn new() -> Self

Construct a zeroed counter.

pub fn record(&self, prompt: u64, completion: u64)

Record a single completion. Increments all four global atomics relaxed-ordering — correctness only requires eventual visibility.

pub fn get(&self) -> (u64, u64, u64)

Return the global (prompt, completion, total) triple.

pub fn global_snapshot(&self) -> GlobalTokenSnapshot

Snapshot the global counters into a plain-data struct.

pub fn model_snapshots(&self) -> Vec<TokenUsageSnapshot>

Snapshot every tracked model’s usage. Returns an empty Vec if the per-model map is currently contended.

Trait Implementations

impl Debug for AtomicTokenCounter

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

Formats the value using the given formatter.

impl Default for AtomicTokenCounter

fn default() -> Self

Returns the “default value” for a type.