Struct LlmAgentBuilder 

pub struct LlmAgentBuilder { /* private fields */ }

Available on crate feature agents only.

Builder for constructing an LlmAgent with all configuration options.

Implementations

impl LlmAgentBuilder

pub fn new(name: impl Into<String>) -> LlmAgentBuilder

Create a new builder with the given agent name.

pub fn description(self, desc: impl Into<String>) -> LlmAgentBuilder

Set the agent description.

pub fn model(self, model: Arc<dyn Llm>) -> LlmAgentBuilder

Set the LLM model for this agent.

pub fn instruction(self, instruction: impl Into<String>) -> LlmAgentBuilder

Set the system instruction for this agent.

pub fn instruction_provider( self, provider: Box<dyn Fn(Arc<dyn ReadonlyContext>) -> Pin<Box<dyn Future<Output = Result<String, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Set a dynamic instruction provider evaluated per invocation.

pub fn global_instruction( self, instruction: impl Into<String>, ) -> LlmAgentBuilder

Set a global instruction prepended to all requests.

pub fn global_instruction_provider( self, provider: Box<dyn Fn(Arc<dyn ReadonlyContext>) -> Pin<Box<dyn Future<Output = Result<String, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Set a dynamic global instruction provider evaluated per invocation.

pub fn with_skills(self, index: SkillIndex) -> LlmAgentBuilder

Available on crate feature skills only.

Set a preloaded skills index for this agent.

pub fn with_auto_skills(self) -> Result<LlmAgentBuilder, AdkError>

Available on crate feature skills only.

Auto-load skills from .skills/ in the current working directory.

pub fn with_skills_from_root( self, root: impl AsRef<Path>, ) -> Result<LlmAgentBuilder, AdkError>

Available on crate feature skills only.

Auto-load skills from .skills/ under a custom root directory.

pub fn with_skill_policy(self, policy: SelectionPolicy) -> LlmAgentBuilder

Available on crate feature skills only.

Customize skill selection behavior.

pub fn with_skill_budget(self, max_chars: usize) -> LlmAgentBuilder

Available on crate feature skills only.

Limit injected skill content length.

pub fn input_schema(self, schema: Value) -> LlmAgentBuilder

Set a JSON schema for validating user input.

pub fn output_schema(self, schema: Value) -> LlmAgentBuilder

Set a JSON schema for structured output from the LLM.

pub fn disallow_transfer_to_parent(self, disallow: bool) -> LlmAgentBuilder

Prevent this agent from transferring control back to its parent.

pub fn disallow_transfer_to_peers(self, disallow: bool) -> LlmAgentBuilder

Prevent this agent from transferring control to peer agents.

pub fn include_contents(self, include: IncludeContents) -> LlmAgentBuilder

Control which conversation history contents are included in LLM requests.

pub fn output_key(self, key: impl Into<String>) -> LlmAgentBuilder

Set a state key where the agent’s final output will be stored.

pub fn generate_content_config( self, config: GenerateContentConfig, ) -> LlmAgentBuilder

Set default generation parameters (temperature, top_p, top_k, max_output_tokens) applied to every LLM request made by this agent.

These defaults are merged with any per-request config. If output_schema is also set, the schema is preserved alongside these generation parameters.

Example

use adk_core::GenerateContentConfig;

let agent = LlmAgentBuilder::new("my-agent")
    .model(model)
    .generate_content_config(GenerateContentConfig {
        temperature: Some(0.7),
        max_output_tokens: Some(2048),
        ..Default::default()
    })
    .build()?;

pub fn temperature(self, temperature: f32) -> LlmAgentBuilder

Set the default temperature for LLM requests. Shorthand for setting just temperature without a full GenerateContentConfig.

pub fn top_p(self, top_p: f32) -> LlmAgentBuilder

Set the default top_p for LLM requests.

pub fn top_k(self, top_k: i32) -> LlmAgentBuilder

Set the default top_k for LLM requests.

pub fn max_output_tokens(self, max_tokens: i32) -> LlmAgentBuilder

Set the default max output tokens for LLM requests.

pub fn max_iterations(self, max: u32) -> LlmAgentBuilder

Set the maximum number of LLM round-trips (iterations) before the agent stops. Default is 100.

pub fn tool_timeout(self, timeout: Duration) -> LlmAgentBuilder

Set the timeout for individual tool executions. Default is 5 minutes. Tools that exceed this timeout will return an error.

pub fn tool(self, tool: Arc<dyn Tool>) -> LlmAgentBuilder

Add a tool to this agent’s toolbox.

pub fn toolset(self, toolset: Arc<dyn Toolset>) -> LlmAgentBuilder

Register a dynamic toolset for per-invocation tool resolution.

Toolsets are resolved at the start of each run() call using the invocation’s ReadonlyContext. This enables context-dependent tools like per-user browser sessions from a pool.

pub fn sub_agent(self, agent: Arc<dyn Agent>) -> LlmAgentBuilder

Add a sub-agent that this agent can delegate to.

pub fn before_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add a before-agent callback.

pub fn after_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add an after-agent callback.

pub fn before_model_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, LlmRequest) -> Pin<Box<dyn Future<Output = Result<BeforeModelResult, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add a before-model callback invoked before each LLM request.

pub fn after_model_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, LlmResponse) -> Pin<Box<dyn Future<Output = Result<Option<LlmResponse>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add an after-model callback invoked after each LLM response.

pub fn before_tool_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add a before-tool callback invoked before each tool execution.

pub fn after_tool_callback( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>) -> Pin<Box<dyn Future<Output = Result<Option<Content>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Add an after-tool callback invoked after each tool execution.

pub fn after_tool_callback_full( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, Arc<dyn Tool>, Value, Value) -> Pin<Box<dyn Future<Output = Result<Option<Value>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Register a rich after-tool callback that receives the tool, arguments, and response value.

This is the V2 callback surface aligned with the Python/Go ADK model where after_tool_callback receives the full tool execution context. Unlike after_tool_callback (which only receives CallbackContext), this callback can inspect and modify tool results directly.

Return Ok(None) to keep the original response, or Ok(Some(value)) to replace the function response sent to the LLM.

These callbacks run after the legacy after_tool_callback chain. ToolOutcome is available via ctx.tool_outcome().

pub fn on_tool_error( self, callback: Box<dyn Fn(Arc<dyn CallbackContext>, Arc<dyn Tool>, Value, String) -> Pin<Box<dyn Future<Output = Result<Option<Value>, AdkError>> + Send>> + Send + Sync>, ) -> LlmAgentBuilder

Register a callback invoked when a tool execution fails (after retries are exhausted).

If the callback returns Ok(Some(value)), the value is used as a fallback function response to the LLM. If it returns Ok(None), the next callback in the chain is tried. If no callback provides a fallback, the original error is reported to the LLM.

pub fn default_retry_budget(self, budget: RetryBudget) -> LlmAgentBuilder

Set a default retry budget applied to all tools that do not have a per-tool override.

When a tool execution fails and a retry budget applies, the agent retries up to budget.max_retries times with the configured delay between attempts.

pub fn tool_retry_budget( self, tool_name: impl Into<String>, budget: RetryBudget, ) -> LlmAgentBuilder

Set a per-tool retry budget that overrides the default for the named tool.

Per-tool budgets take precedence over the default retry budget.

pub fn circuit_breaker_threshold(self, threshold: u32) -> LlmAgentBuilder

Configure a circuit breaker that temporarily disables tools after threshold consecutive failures within a single invocation.

When a tool’s consecutive failure count reaches the threshold, subsequent calls to that tool are short-circuited with an immediate error response until the next invocation (which resets the state).

pub fn tool_confirmation_policy( self, policy: ToolConfirmationPolicy, ) -> LlmAgentBuilder

Configure tool confirmation requirements for this agent.

pub fn require_tool_confirmation( self, tool_name: impl Into<String>, ) -> LlmAgentBuilder

Require confirmation for a specific tool name.

pub fn require_tool_confirmation_for_all(self) -> LlmAgentBuilder

Require confirmation for all tool calls.

pub fn tool_execution_strategy( self, strategy: ToolExecutionStrategy, ) -> LlmAgentBuilder

Set the tool execution strategy for this agent.

When set, this overrides the RunConfig’s tool_execution_strategy for this agent’s dispatch loop. When None (the default), the RunConfig value is used.

pub fn input_guardrails(self, guardrails: GuardrailSet) -> LlmAgentBuilder

Set input guardrails to validate user input before processing.

Input guardrails run before the agent processes the request and can:

• Block harmful or off-topic content
• Redact PII from user input
• Enforce input length limits

Requires the guardrails feature.

pub fn output_guardrails(self, guardrails: GuardrailSet) -> LlmAgentBuilder

Set output guardrails to validate agent responses.

Output guardrails run after the agent generates a response and can:

• Enforce JSON schema compliance
• Redact PII from responses
• Block harmful content in responses

Requires the guardrails feature.

pub fn build(self) -> Result<LlmAgent, AdkError>

Build the LlmAgent, returning an error if no model was set.