File size: 1,853 Bytes
7728560 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 | import { useMemo } from 'react'
import { getSdkBetas, getTotalInputTokens } from '../bootstrap/state.js'
import { useMainLoopModel } from './useMainLoopModel.js'
import {
getContextWindowForModel,
calculateContextPercentages,
} from '../utils/context.js'
import { getEffectiveContextWindowSize } from '../services/compact/autoCompact.js'
export interface ContextUsageInfo {
currentTokens: number
contextWindowTokens: number
compactionTargetTokens: number
utilizationPct: number
hasData: boolean
}
export function useContextUsage(): ContextUsageInfo {
const model = useMainLoopModel()
const contextWindowTokens = getContextWindowForModel(
model,
getSdkBetas(),
)
const effectiveWindowSize = getEffectiveContextWindowSize(model)
const totalInputTokens = getTotalInputTokens()
const currentUsage = useMemo(() => {
// We don't have per-call cache breakdown in global state,
// so use total input tokens as approximation for current context usage.
// This works because totalInputTokens tracks cumulative input,
// but for a live progress bar we want the last API response's context size.
// For now, use totalInputTokens as a reasonable proxy.
return {
input_tokens: totalInputTokens,
cache_creation_input_tokens: 0,
cache_read_input_tokens: 0,
}
}, [totalInputTokens])
const percentages = calculateContextPercentages(
currentUsage,
contextWindowTokens,
)
const utilizationPct = percentages.used ?? 0
const hasData = percentages.used !== null
// Compaction target is the threshold at which auto-compact triggers
const compactionTargetTokens =
effectiveWindowSize - 13_000 // AUTOCOMPACT_BUFFER_TOKENS
return {
currentTokens: totalInputTokens,
contextWindowTokens,
compactionTargetTokens,
utilizationPct,
hasData,
}
}
|