:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
import { useValues } from 'kea'
import React from 'react'
import { IconSparkles } from '@posthog/icons'
import { LemonButton } from '@posthog/lemon-ui'
import {
ConversationDisplayOption,
ConversationMessagesDisplay,
} from '../ConversationDisplay/ConversationMessagesDisplay'
import { useCustomParserMaxTool } from '../customParser/useCustomParserMaxTool'
import { useAIData } from '../hooks/useAIData'
import { normalizeMessage, normalizeMessages } from '../messageNormalization'
import type { GenerationSentiment } from '../sentimentResults'
import { parserRecipesLogic } from '../settings/parserRecipesLogic'
import { AIDataLoading } from './AIDataLoading'
import { JSONValueDisplay } from './JSONValueDisplay'
interface EventContentConversationProps {
eventId: string
traceId?: string | null
timestamp?: string
rawInput: unknown
rawOutput: unknown
tools?: unknown
errorData?: unknown
httpStatus?: unknown
raisedError?: boolean
outputTokens?: unknown
reasoningTokens?: unknown
textOutputTokens?: unknown
stopReason?: unknown
searchQuery?: string
displayOption?: ConversationDisplayOption
/** Original $ai_input index to auto-expand (e.g. from sentiment tab deep link) */
highlightMessageIndex?: number | null
/** Generation id for per-message sentiment lookups; only generations have sentiment. */
generationEventId?: string
generationSentiment?: GenerationSentiment | null
}
// Renders an event's input/output, routing to one of two renderers: the chat UI
// when recipes recognize both sides as a conversation, or a raw JSON view
// otherwise. One path for generations, spans, and embeddings.
export function EventContentConversation({
eventId,
traceId,
timestamp,
rawInput,
rawOutput,
tools,
errorData,
httpStatus,
raisedError,
outputTokens,
reasoningTokens,
textOutputTokens,
stopReason,
searchQuery,
displayOption,
highlightMessageIndex,
generationEventId,
generationSentiment,
}: EventContentConversationProps): JSX.Element {
const {
input,
output,
tools: loadedTools,
isLoading,
} = useAIData({
uuid: eventId,
input: rawInput,
output: rawOutput,
tools,
traceId: traceId ?? undefined,
timestamp,
})
// The normalizer is a module singleton — recipesVersion signals the memos below are stale
const { recipesVersion } = useValues(parserRecipesLogic)
// Map each normalized input message back to its original index in $ai_input,
// a stable key for per-message sentiment lookups. Generations only.
const inputSourceIndices = React.useMemo(() => {
if (!generationEventId) {
return undefined
}
const indices: number[] = []
if (loadedTools) {
indices.push(-1) // tools message prepended by normalizeMessages
}
if (Array.isArray(input)) {
for (let i = 0; i < input.length; i++) {
const expanded = normalizeMessage(input[i], 'user')
for (let j = 0; j < expanded.length; j++) {
indices.push(i)
}
}
}
return indices
// oxlint-disable-next-line react-hooks/exhaustive-deps
}, [input, loadedTools, generationEventId, recipesVersion])
const { recognized: inputRecognized, messages: inputMessages } = React.useMemo(
() => normalizeMessages(input, 'user', loadedTools),
// oxlint-disable-next-line react-hooks/exhaustive-deps
[input, loadedTools, recipesVersion]
)
const { recognized: outputRecognized, messages: outputMessages } = React.useMemo(
() => normalizeMessages(output, 'assistant'),
// oxlint-disable-next-line react-hooks/exhaustive-deps
[output, recipesVersion]
)
const openCustomParserMax = useCustomParserMaxTool({
eventId,
traceId,
input,
output,
tools: loadedTools,
inputRecognized,
outputRecognized,
isLoading,
isGeneration: !!generationEventId,
})
if (isLoading) {
return <AIDataLoading variant="block" />
}
if (!inputRecognized || !outputRecognized) {
return (
<JsonInputOutput
input={input}
output={output}
errorData={errorData}
raisedError={raisedError}
onSetUpCustomParser={openCustomParserMax}
/>
)
}
return (
<ConversationMessagesDisplay
inputNormalized={inputMessages}
outputNormalized={outputMessages}
inputSourceIndices={inputSourceIndices}
errorData={errorData}
httpStatus={typeof httpStatus === 'number' ? httpStatus : undefined}
raisedError={raisedError}
outputTokens={outputTokens}
reasoningTokens={reasoningTokens}
textOutputTokens={textOutputTokens}
stopReason={stopReason}
searchQuery={searchQuery}
displayOption={displayOption}
traceId={traceId}
eventId={eventId}
generationSentiment={generationSentiment}
highlightMessageIndex={highlightMessageIndex}
/>
)
}
// Raw renderer for input/output that no recipe recognized as a conversation.
function JsonInputOutput({
input,
output,
errorData,
raisedError,
onSetUpCustomParser,
}: {
input: unknown
output: unknown
errorData?: unknown
raisedError?: boolean
onSetUpCustomParser?: (() => void) | null
}): JSX.Element {
// On error, surface the error payload when there's no output to show.
const outputValue = raisedError ? (output ?? errorData) : output
return (
<div className="space-y-4">
{onSetUpCustomParser && (
<div className="flex items-center justify-between gap-2 px-2 py-1 border border-primary rounded bg-surface-secondary">
<span className="text-xs text-muted">
Shown as raw JSON — no parser recognizes this event's shape.
</span>
<LemonButton
type="secondary"
size="xsmall"
icon={<IconSparkles />}
onClick={onSetUpCustomParser}
data-attr="llma-json-fallback-create-parser"
>
Set up custom parser with PostHog AI
</LemonButton>
</div>
)}
<div>
<h3 className="font-semibold mb-2">Input</h3>
<div className="p-2 bg-surface-secondary rounded text-xs overflow-auto">
<JSONValueDisplay value={input} />
</div>
</div>
<div>
<h3 className="font-semibold mb-2">Output</h3>
<div
className={`p-2 rounded text-xs overflow-auto ${
raisedError ? 'bg-danger-highlight' : 'bg-surface-secondary'
}`}
>
<JSONValueDisplay value={outputValue} />
</div>
</div>
</div>
)
}