Struct Agent 

pub struct Agent { /* private fields */ }

High-level agent facade.

Holds the LLM client and agent config. Workspace-independent. Use Agent::session() to bind to a workspace.

Implementations

impl Agent

pub async fn new(config_source: impl Into<String>) -> Result<Self>

Create from a config file path or inline config string.

Auto-detects: file path (.hcl/.json) vs inline JSON vs inline HCL.

pub async fn create(config_source: impl Into<String>) -> Result<Self>

Create from a config file path or inline config string.

Alias for Agent::new() — provides a consistent API with the Python and Node.js SDKs.

pub async fn from_config(config: CodeConfig) -> Result<Self>

Create from a CodeConfig struct.

pub async fn refresh_mcp_tools(&self) -> Result<()>

Re-fetch tool definitions from all connected global MCP servers and update the internal cache.

Call this when an MCP server has added or removed tools since the agent was created. The refreshed tools will be visible to all new sessions created after this call; existing sessions are unaffected (their ToolExecutor snapshot is already built).

pub fn session( &self, workspace: impl Into<String>, options: Option<SessionOptions>, ) -> Result<AgentSession>

Bind to a workspace directory, returning an AgentSession.

Pass None for defaults, or Some(SessionOptions) to override the model, agent directories for this session.

pub fn session_for_agent( &self, workspace: impl Into<String>, def: &AgentDefinition, extra: Option<SessionOptions>, ) -> Result<AgentSession>

Create a session pre-configured from an [AgentDefinition].

Maps the definition’s permissions, prompt, model, and max_steps directly into SessionOptions, so markdown/YAML-defined subagents can be used as crate::agent_teams::TeamRunner members without manual wiring.

The mapping follows the same logic as the built-in task tool:

• permissions → permission_checker
• prompt → prompt_slots.extra
• max_steps → max_tool_rounds
• model → model (as "provider/model" string)

extra can supply additional overrides (e.g. planning_enabled) that take precedence over the definition’s values.

pub fn resume_session( &self, session_id: &str, options: SessionOptions, ) -> Result<AgentSession>

Resume a previously saved session by ID.

Loads the session data from the store, rebuilds the AgentSession with the saved conversation history, and returns it ready for continued use.

The options must include a session_store (or with_file_session_store) that contains the saved session.

Trait Implementations

impl Debug for Agent

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

Formats the value using the given formatter.