Type Alias: GenerateTextFn()
GenerateTextFn = (
promptOrOptions) =>Promise<string>
Defined in: apps/paracosm/src/engine/compiler/types.ts:19
Function signature for LLM text generation calls used by compile hooks.
Supports two call shapes:
- Legacy: pass a raw prompt string (no caching)
- Cache-aware: pass
{ system, prompt }to route the stable prefix through cacheBreakpoint-tagged system blocks
Wrappers in llm-invocations/ always use the cache-aware form; direct callers (CLI scripts, tests) can still use the string form.
Parameters
promptOrOptions
string |
{ maxTokens?: number; prompt: string; system?: object[]; }
maxTokens?
number
Upper bound on completion tokens for this call. Caps tail spend when a model misbehaves and yaps beyond the intended output size (provider defaults sit at 4-8k tokens). Use ~2× the typical response size so well-behaved calls finish naturally and only runaway generations hit the cap.
prompt
string
system?
object[]
Returns
Promise<string>