Trait BaseAgent

Source

pub trait BaseAgent: Send + Sync {
    // Required method
    fn plan<'life0, 'life1, 'life2, 'life3, 'async_trait>(
        &'life0 self,
        intermediate_steps: &'life1 [AgentStep],
        inputs: &'life2 HashMap<String, String>,
        config: Option<&'life3 RunnableConfig>,
    ) -> Pin<Box<dyn Future<Output = Result<AgentOutput, AgentError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait,
             'life2: 'async_trait,
             'life3: 'async_trait;

    // Provided methods
    fn plan_stream<'life0, 'life1, 'life2, 'life3, 'life4, 'async_trait>(
        &'life0 self,
        intermediate_steps: &'life1 [AgentStep],
        inputs: &'life2 HashMap<String, String>,
        on_token: &'life3 mut (dyn FnMut(String) -> Pin<Box<dyn Future<Output = ()> + Send>> + Send),
        config: Option<&'life4 RunnableConfig>,
    ) -> Pin<Box<dyn Future<Output = Result<AgentOutput, AgentError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait,
             'life2: 'async_trait,
             'life3: 'async_trait,
             'life4: 'async_trait { ... }
    fn input_keys(&self) -> Vec<&str> { ... }
    fn get_allowed_tools(&self) -> Option<Vec<&str>> { ... }
    fn return_stopped_response(
        &self,
        _intermediate_steps: &[AgentStep],
    ) -> AgentFinish { ... }
    fn last_token_usage(&self) -> Option<TokenUsage> { ... }
}

Expand description

Base Agent trait.

Defines the core interface for agents. Agent is responsible for planning, not execution. Execution is handled by AgentExecutor.

Required Methods§

Source

fn plan<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, intermediate_steps: &'life1 [AgentStep], inputs: &'life2 HashMap<String, String>, config: Option<&'life3 RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<AgentOutput, AgentError>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait,

Plans the next action.

§Arguments

intermediate_steps - History of executed steps.
inputs - User input.
config - Per-run config for this planning round. The executor supplies a config carrying its callback manager plus trace-linkage metadata, so the LLM provider fires on_llm_* for every planning round and the LLM run joins the agent chain’s trace tree; None means “no observers configured” and providers dispatch nothing (the pre-0.22.4 behavior). Implementations forward it verbatim to chat / stream_chat and never inspect or strip it.

§Returns

AgentOutput::Action - Action to execute.
AgentOutput::Finish - Final answer.

Provided Methods§

Source

fn plan_stream<'life0, 'life1, 'life2, 'life3, 'life4, 'async_trait>( &'life0 self, intermediate_steps: &'life1 [AgentStep], inputs: &'life2 HashMap<String, String>, on_token: &'life3 mut (dyn FnMut(String) -> Pin<Box<dyn Future<Output = ()> + Send>> + Send), config: Option<&'life4 RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<AgentOutput, AgentError>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait, 'life4: 'async_trait,

Plans the next action, streaming any model text through on_token as it is produced.

§Arguments

intermediate_steps - History of executed steps.
inputs - User input.
on_token - Called with each chunk of model text as it becomes available, taking ownership of the chunk (so the returned future never borrows the token and stays 'static). Streaming-capable agents (e.g. ReAct, function-calling) emit free text per token; steps that produce no free text (e.g. a function-calling step invoking a tool) emit nothing. Non-streaming agents deliver the whole answer in one call.

§Returns

AgentOutput::Action - Action to execute.
AgentOutput::Finish - Final answer.

The default implementation delegates to BaseAgent::plan and forwards the whole final-answer text as a single chunk — behaviorally identical to calling plan() directly. Streaming-capable agents (e.g. ReAct) override this to emit per-token chunks from the model’s streaming chat API; callers must still call BaseAgent::plan for the non-streaming path (e.g. invoke) so that path is unaffected.

Source