: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 { CyclotronJobInputSchemaType, CyclotronJobInputType, HogFunctionTypeType } from '~/types'
export type HogFunctionDeliveryType = 'batch' | 'realtime'
// Batch exports vs realtime destinations share `type: 'destination'`; the only signal is the id prefix.
export function getHogFunctionDeliveryType(item: { id: string }): HogFunctionDeliveryType {
return item.id.startsWith('batch-export-') ? 'batch' : 'realtime'
}
export function humanizeHogFunctionType(type: HogFunctionTypeType, plural: boolean = false): string {
if (type === 'source_webhook') {
return 'source' + (plural ? 's' : '')
}
if (type === 'internal_destination') {
return 'notification' + (plural ? 's' : '')
}
if (type === 'site_app') {
return 'Web script' + (plural ? 's' : '')
}
if (type === 'transformation_log') {
return 'log transformation' + (plural ? 's' : '')
}
return type.replaceAll('_', ' ') + (plural ? 's' : '')
}
/** Default char cap for a config blob attached to the PostHog AI agent as context. */
export const HOG_FUNCTION_CONTEXT_MAX_CHARS = 10_000
/**
* Caps a stringified config blob before it's registered as PostHog AI attached context, so a pathological
* hog source or inputs payload can't bloat the agent's context window. These are keyed entity-style items
* (not `type: 'text'`), so they're sent once per run rather than every turn; the cap is a safety ceiling.
*/
export function truncateHogFunctionContext(value: string, max: number = HOG_FUNCTION_CONTEXT_MAX_CHARS): string {
return value.length > max ? value.slice(0, max) + '… (truncated)' : value
}
/**
* Replaces secret input values with a placeholder before the live form config leaves the scene (agent
* context, approval-card diffs). A saved secret comes back masked from the API, but a value the user
* just typed sits in form state in cleartext and must never reach the LLM. An entry counts as secret
* when the schema marks its key secret or the entry itself carries `secret: true`.
*/
export function redactSecretHogFunctionInputs(
inputs: Record<string, CyclotronJobInputType>,
inputsSchema: CyclotronJobInputSchemaType[]
): Record<string, CyclotronJobInputType> {
const secretKeys = new Set(inputsSchema.filter((schema) => schema.secret).map((schema) => schema.key))
return Object.fromEntries(
Object.entries(inputs).map(([key, entry]) => {
const isSecret = secretKeys.has(key) || entry?.secret === true
return [key, isSecret && entry ? { ...entry, value: '[secret]' } : entry]
})
)
}