@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
17 lines (16 loc) • 897 B
JavaScript
/**
* Size/count ceilings for the Codex-outbound fallback, centralized so request
* ingestion (`codexOutboundFallback.ts`) and response translation
* (`codexToAnthropicFallback.ts`) share one source instead of re-declaring the
* numbers `codexFallback.ts` already applies on the reverse
* (Claude-into-Codex) leg (`:763`, `:784`).
*/
/** Caps how much unparsed upstream SSE text a translator buffers, and how
* long one tool call's argument, input or output text may be, mirroring
* codexFallback.ts's own ceiling for the reverse leg. */
export const CODEX_STREAM_SIZE_CEILING_BYTES = 16 * 1024 * 1024;
/** Caps how many tool calls one turn may open or replay, mirroring
* codexFallback.ts's own ceiling for the reverse leg — a runaway or
* adversarial upstream cannot make a translator accumulate unbounded
* open-item state. */
export const CODEX_STREAM_MAX_TOOL_CALLS = 4096;