Struct DeepResearchAgent 

pub struct DeepResearchAgent<M: BaseChatModel> { /* private fields */ }

Multi-round deep research agent.

Decomposes a topic into sub-topics, searches in parallel, synthesizes a comprehensive report with citations, and iterates if gaps remain.

Implementations

impl<M: BaseChatModel> DeepResearchAgent<M>

pub fn new(llm: M) -> Self

Creates a new DeepResearchAgent with the given LLM.

At least one search tool must be added via with_searcher before calling research.

pub fn with_searcher(self, tool: Box<dyn BaseTool>) -> Self

Adds a search tool to the agent.

Multiple search tools can be added; all are queried in parallel.

pub fn with_max_rounds(self, n: usize) -> Self

Sets the maximum number of research rounds (default: 2).

pub fn with_max_subtopics(self, n: usize) -> Self

Sets the maximum number of sub-topics to decompose (default: 5).

pub fn with_max_source_tokens(self, tokens: usize) -> Self

Sets the maximum number of tokens for source text in synthesis prompts.

When set, source snippets are truncated to fit within this budget.

pub async fn research( &self, topic: &str, ) -> Result<ResearchReport, ResearchError>

Runs the full deep research pipeline on the given topic.

Returns a ResearchReport containing the markdown report, citations, sub-topics, and round count.

pub async fn stream_research( &self, topic: &str, ) -> Result<Pin<Box<dyn Stream<Item = AgentStreamEvent> + Send>>, ResearchError>

Streams the deep research execution, emitting pipeline step events.

Emits AgentStreamEvent::PipelineStep events for planning, searching, and synthesizing, and AgentStreamEvent::FinalAnswer when the report is ready.

Trait Implementations

impl<M: BaseChatModel> Debug for DeepResearchAgent<M>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<M> Orchestrator for DeepResearchAgent<M>

where M: BaseChatModel + Send + Sync, M::Error: Send + Sync,

type Input = String

Input type (usually a String objective/question).

type Output = ResearchReport

Output type.

fn run_with_context<'life0, 'life1, 'async_trait>( &'life0 self, input: Self::Input, ctx: &'life1 RunContext, ) -> Pin<Box<dyn Future<Output = Result<Self::Output, AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Execution entry point carrying the run context.