RuntimeAgent

RuntimeAgent is the concrete in-memory agent implementation. It stores a system prompt, a ToolRegistry, optional provider/model overrides, and the per-provider override fields applied by AgentBuilder.

Construction

Most code should use AgentBuilder::build() or Agent::quick(...), but direct construction is available:

pub fn new(
    name: impl Into<String>,
    system_prompt: impl Into<String>,
    registry: Arc<ToolRegistry>,
) -> Self

Public methods

pub fn with_provider(self, provider: LlmProvider) -> Self
pub fn with_model(self, model: impl Into<String>) -> Self
pub fn provider(&self) -> Option<LlmProvider>
pub fn model(&self) -> String
pub fn registry(&self) -> &Arc<ToolRegistry>
pub fn set_system_prompt(&mut self, prompt: impl Into<String>)
pub fn add_tool(&mut self, tool: Arc<dyn Tool>)
pub fn stream(&self, message: impl Into<String>) -> StreamBuilder<'_>

Notes:

• model() falls back to "openai/gpt-5" when no override is set.
• provider() returns only the agent-level override, not the resolved runtime provider after config/env merging.
• stream(...) is just a convenience wrapper around the trait-based streaming runtime.

Trait behavior

RuntimeAgent implements Agent and overrides:

• name()
• provider()
• apply_config_overrides(&mut AppConfig)
• required_completion_tools()
• max_continuations()
• continuation_message()
• system_prompt()
• available_tools()
• execute_tool(...)

The run/continue methods use the default runtime implementations from src/agent/runtime.rs.

Example

use appam::prelude::*;

let mut agent = Agent::quick("openai/gpt-5", "You are helpful.", vec![])?;
agent.set_system_prompt("You are a careful reviewer.");

agent
    .stream("Review this function")
    .on_content(|text| print!("{}", text))
    .run()
    .await?;