researcher

a cited answer, retrying until every claim is grounded.

The grounding loop is this agent's own contract: a grounded, cited answer is what it produces. Checking that answer against a wider task is the caller's job, via reviewAgent.

Types

ResearchEvalInput

What an eval hands the researcher: the question, and the conversation before it when the test has one. The shape the evals/researcher suite uses as its input.

/** What an eval hands the researcher: the question, and the conversation
  before it when the test has one. The shape the `evals/researcher` suite
  uses as its input. */
export type ResearchEvalInput = {
  task: string;
  history?: ThreadMessage[]
}

(source)

Functions

buildTools

buildTools(): any[]

Return the researcher's tools: the encyclopedia and fetch tools that need no key, a local file read for material the caller points at, and whichever search providers the environment has a key for.

Returns: any[]

(source)

researcherAgent

researcherAgent(
  task: string,
  context: string = "",
  maxAttempts: number = 2,
  maxCost: number = $50.00,
  maxTime: number = 30m,
  model: string = "",
  provider: string = "",
  session: string = "",
  extraTools: any[] = [],
  history: ThreadMessage[] = [],
): Result<string>

Answer a research question from web and Wikipedia sources and return a cited answer, retrying while a review finds ungrounded claims.

@param task - The research question @param context - Extra material folded into the prompt, or "" @param history - The conversation before the question, oldest first, so the question is read in its context; [] for none @param maxAttempts - Answer-and-fix attempts before returning @param maxCost - Hard spend cap @param maxTime - Hard wall-clock cap @param model - Model override, or "" for the ambient model @param provider - Provider for the model override @param session - Session name to share a thread across calls, or "" for isolated @param extraTools - Extra tools to offer the LLM, appended to the built-in set

Parameters:

Name	Type	Default
task	string
context	string	""
maxAttempts	number	2
maxCost	number	$50.00
maxTime	number	30m
model	string	""
provider	string	""
session	string	""
extraTools	any[]	[]
history	ThreadMessage[]	[]

Returns: Result<string>

Throws: std::guard

(source)