jj.bettong.usage — org.clojars.jj/bettong 0.1.0-SNAPSHOT

Token accounting. Every model response carries a usage block; the agent adds them up across the run so you can see what a task cost.

add clj

(add a b)

Sums two usage maps field by field.

cache-hit-rate clj

(cache-hit-rate {:keys [prompt-tokens cache-hit-tokens]})

The share of prompt tokens the provider served from its cache, 0.0 to 1.0. nil when nothing has been sent yet.

cost clj

(cost usage prices)

What a run cost, given prices in currency per million tokens:

{:input 0.28 :output 0.42} ; flat {:input 0.28 :output 0.42 :cache-hit 0.028} ; cheaper cached prompt tokens

Prices change and differ per model, so the harness holds none - pass your provider's current numbers. Returns nil when no prices are given.

normalize clj

(normalize usage)

Turns an API usage block into our keys. DeepSeek reports cache hits and misses separately (a hit is billed cheaper); other OpenAI-compatible providers omit them, and missing fields simply stay zero.

of-message clj

(of-message message)

The usage jj.bettong.llm/chat stashed on the message it returned. A stubbed chat-fn carries none, which still counts as one request.

summary clj

(summary {:keys [requests prompt-tokens completion-tokens total-tokens
                 cache-hit-tokens]
          :as usage})

One line worth printing at the end of a run.

zero clj