Types: TypeScript

Below are the core types used by the @guardrails/guardrails-ts library.

GuardrailLLMContext

export interface GuardrailLLMContext {
  guardrailLlm: OpenAI;
}

Context interface providing access to the OpenAI client used by guardrails.

GuardrailLLMContextWithHistory

export interface GuardrailLLMContextWithHistory extends GuardrailLLMContext {
  getConversationHistory(): any[];
  getInjectionLastCheckedIndex(): number;
  updateInjectionLastCheckedIndex(index: number): void;
}

Extends the base context with helpers for conversation-history aware checks (e.g., prompt injection detection).

GuardrailResult

export interface GuardrailResult {
  tripwireTriggered: boolean;
  executionFailed?: boolean;
  originalException?: Error;
  info: {
    checked_text: string;
    [key: string]: any;
  };
}

Standard result returned by every guardrail check. The executionFailed field indicates if the guardrail itself failed to execute (e.g., invalid model name), and originalException contains the exception that caused the failure. These fields are used by the raise_guardrail_errors parameter to control error handling behavior.

CheckFn

export type CheckFn<TContext = object, TIn = unknown, TCfg = object> =
  (ctx: TContext, input: TIn, config: TCfg) => GuardrailResult | Promise<GuardrailResult>;

Callable signature implemented by all guardrails. May be sync or async.

Utility Types

export type MaybeAwaitableResult = GuardrailResult | Promise<GuardrailResult>;
export type TContext = object;
export type TIn = unknown;
export type TCfg = object;

For the full source, see guardrails-ts/src/types.ts in the repository.