tokenTracker()

Wraps a stream with reactive token/cost tracking metadata (Tier 2).

Signature

function tokenTracker<A>(
	countTokens: (value: A) => TokenUsage,
	opts?: { name?: string },
): StoreOperator<A, A>

Parameters

Parameter	Type	Description
countTokens	(value: A) => TokenUsage	Function that extracts token usage from each emitted value.
opts	{ name?: string }	Optional configuration.

Returns

StoreOperator<A, A> — pipe-compatible. The returned store has a tokens property (Store<TokenMeta>).

Method	Signature	Description
get()	() => A \	undefined
tokens	Store\<TokenMeta\>	Reactive metadata: promptTokens, completionTokens, totalTokens, cost, count.
source	callbag	Underlying callbag source for subscriptions.

Basic Usage

import { state, pipe, effect } from 'callbag-recharge';
import { tokenTracker } from 'callbag-recharge/orchestrate';

const llmOutput = state({ text: 'Hello', usage: { promptTokens: 10, completionTokens: 5 } });
const tracked = pipe(llmOutput, tokenTracker(v => v.usage));
effect([tracked.tokens], () => {
    const t = tracked.tokens.get();
    console.log(`${t.totalTokens} tokens`); // "15 tokens"
  });

Options / Behavior Details

• Tier 2: Cycle boundary — each forwarded value starts a new DIRTY+value cycle.
• Accumulative: Token counts and cost accumulate across all values. Resets on reconnect.
• Flexible extraction: countTokens can return partial usage — missing fields default to 0.

See Also

• track — lifecycle tracking
• fromLLM — LLM adapter