The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
import { createContext, type ReactNode, useContext, useMemo } from 'react'
import type { ChatMessage } from '@/lib/chat-messages'
export interface TranscriptWindowValue {
/** Store holds older messages the runtime window has not materialized. */
olderAvailable: boolean
/** Direct row addressing replaces only the display page; null means no jump. */
revealRow?: (rowId: number, signal: AbortSignal) => Promise<string | null>
/** Cancel any pending jump and reselect the current live tail. */
returnToLatest?: () => void
isHistorical?: boolean
newerAvailable?: boolean
/** Selected historical page; live navigation reads the session store. */
currentMessages?: readonly ChatMessage[]
/** Pull a page, capturing the reader immediately before the prepend commits.
* A remote page resolves false on failure; callers can retry without growing
* an empty render window or holding the reader still during network I/O. */
expandWindow: (beforePrepend?: () => void) => void | Promise<boolean>
}
const DEFAULT_TRANSCRIPT_WINDOW: Required<TranscriptWindowValue> = {
olderAvailable: false,
expandWindow: () => {},
revealRow: async () => null,
returnToLatest: () => {},
isHistorical: false,
newerAvailable: false,
currentMessages: []
}
const TranscriptWindowContext = createContext<Required<TranscriptWindowValue>>(DEFAULT_TRANSCRIPT_WINDOW)
export function TranscriptWindowProvider({
children,
value
}: {
children: ReactNode
value: Pick<TranscriptWindowValue, 'olderAvailable' | 'expandWindow'> & Partial<TranscriptWindowValue>
}) {
const complete = useMemo(() => ({ ...DEFAULT_TRANSCRIPT_WINDOW, ...value }), [value])
return <TranscriptWindowContext.Provider value={complete}>{children}</TranscriptWindowContext.Provider>
}
export function useTranscriptWindow(): Required<TranscriptWindowValue> {
return useContext(TranscriptWindowContext)
}
/**
* "Show earlier" pages the DOM budget first and only then asks the store for
* more messages — the DOM page is already-materialized content, so spending it
* first keeps the click cheap and the store window as small as it can be.
*/
export function resolveShowEarlierAction(hiddenCount: number, olderAvailable: boolean): 'dom' | 'window' | null {
if (hiddenCount > 0) {
return 'dom'
}
return olderAvailable ? 'window' : null
}
/**
* Slack (px) within which a reader counts as "at the top edge". Wide enough
* that a wheel notch landing a few pixels short of 0 still pages; well under
* the RUN_START_SNAP-style thresholds so a mid-transcript reader never does.
*/
export const TOP_EDGE_PX = 48
export interface ShouldAutoShowEarlierInput {
action: 'dom' | 'window' | null
isAtBottom: boolean
loadSettled: boolean
restorePending: boolean
scrollTop: number
/** Present only for `wheel` events; omitted for `scroll`. */
wheelDeltaY?: number
}
/**
* Whether reading at the viewport top should page older turns through the
* same `showEarlier()` path as the button. An unsettled load, a prepend
* restore still pending, a reader following the bottom, or a mid-transcript
* scroll must never page on its own — each of those has scrollTop near 0 or
* changing for reasons that are not "I want to read earlier".
*/
export function shouldAutoShowEarlier({
action,
isAtBottom,
loadSettled,
restorePending,
scrollTop,
wheelDeltaY
}: ShouldAutoShowEarlierInput): boolean {
if (action == null || !loadSettled || restorePending || isAtBottom || scrollTop > TOP_EDGE_PX) {
return false
}
// A wheel at the clamped top is intent only when it points up.
return wheelDeltaY === undefined || wheelDeltaY < 0
}