@juspay/neurolink
Version:
The pipe layer of an AI nervous system: one interface connecting provider neurons to your application, across three inference types — generate, stream and decide. `decide` returns typed, calibrated judgements from a non-generative model (~400ms, ~$0.00002
27 lines (26 loc) • 872 B
TypeScript
/**
* Curator P3-6: shape of the sentinel chunk yielded by every provider's
* stream-transformation generator when AI SDK throws
* `NoOutputGeneratedError`. Built by `buildNoOutputSentinel` in
* `src/lib/utils/noOutputSentinel.ts`.
*/
export type StreamNoOutputSentinel = {
content: "";
metadata: {
noOutput: true;
errorType: "NoOutputGeneratedError";
finishReason: unknown;
usage: unknown;
providerError: string;
modelResponseRaw: string | undefined;
};
};
/**
* Subset of AI SDK's `StreamTextResult` that the sentinel builder reads.
* Both fields are Promises in production but typed loosely so callers
* can pass either the Promise or a resolved value.
*/
export type StreamNoOutputSentinelResultLike = {
finishReason?: Promise<unknown> | unknown;
totalUsage?: Promise<unknown> | unknown;
};