oris-runtime 0.61.0

mod agent;
pub use agent::*;

mod hitl;
pub use hitl::*;

mod checkpoint;
pub use checkpoint::*;

mod message_repair;

mod executor;
pub use executor::*;

mod utils;
#[allow(unused_imports)]
pub use utils::*;

mod chat;
pub use chat::*;

mod open_ai_tools;
pub use open_ai_tools::*;

mod error;
pub use error::*;

mod model_detector;
pub use model_detector::*;

mod unified_agent;
pub use unified_agent::*;

mod state;
pub use state::*;

mod step_adapter;
pub use step_adapter::*;

mod structured_output;
pub use structured_output::*;

pub mod middleware;
pub use middleware::*;

pub mod multi_agent;
#[allow(ambiguous_glob_reexports)]
pub use multi_agent::*;

mod runtime;
pub use runtime::*;

pub mod context_engineering;
pub use context_engineering::*;

pub mod deep_agent;
#[allow(ambiguous_glob_reexports)]
pub use deep_agent::*;

// Integration tests module (requires external services/API keys)
#[cfg(test)]
mod integration_tests;

use std::sync::Arc;

use crate::{
    language_models::llm::LLM,
    schemas::StructuredOutputStrategy,
    tools::{Tool, ToolContext, ToolStore},
};

// Re-export for internal use
// ConversationalAgentBuilder is already exported via `pub use chat::*;` above
// Middleware trait is already exported via `pub use middleware::*;` above

/// Create an agent with a simplified API, similar to LangChain Python's `create_agent`.
///
/// This function provides a simple way to create an agent with:
/// - Model specified as a string (e.g., "gpt-4o-mini", "claude-3-sonnet")
/// - Tools as a slice
/// - Optional system prompt
/// - Optional middleware
///
/// # Arguments
///
/// * `model` - Model identifier string (e.g., "gpt-4o-mini", "claude-3-sonnet")
/// * `tools` - Slice of tools available to the agent
/// * `system_prompt` - Optional system prompt to guide agent behavior
/// * `middleware` - Optional list of middleware to apply during execution
///
/// # Returns
///
/// Returns a `Result<UnifiedAgent, AgentError>` containing the created agent or an error.
///
/// # Example
///
/// ```ignore
/// use std::sync::Arc;
/// use oris_runtime::agent::create_agent;
/// use oris_runtime::agent::middleware::LoggingMiddleware;
/// let agent = create_agent("gpt-4o-mini", &[], Some("You are a helpful assistant"), None).unwrap();
/// let result = agent.invoke_messages(vec![oris_runtime::schemas::Message::new_human_message("Hello")]).await.unwrap();
/// ```
pub fn create_agent(
    model: &str,
    tools: &[Arc<dyn Tool>],
    system_prompt: Option<&str>,
    middleware: Option<Vec<Arc<dyn Middleware>>>,
) -> Result<UnifiedAgent, AgentError> {
    create_agent_with_runtime(
        model,
        tools,
        system_prompt,
        None,
        None,
        None,
        middleware,
        None,
    )
}

/// Create an agent with structured output support.
pub fn create_agent_with_structured_output(
    model: &str,
    tools: &[Arc<dyn Tool>],
    system_prompt: Option<&str>,
    response_format: Option<Box<dyn StructuredOutputStrategy>>,
    middleware: Option<Vec<Arc<dyn Middleware>>>,
) -> Result<UnifiedAgent, AgentError> {
    create_agent_with_runtime(
        model,
        tools,
        system_prompt,
        None,
        None,
        response_format,
        middleware,
        None,
    )
}

/// Create an agent with runtime support (context, store, and optional file backend).
pub fn create_agent_with_runtime(
    model: &str,
    tools: &[Arc<dyn Tool>],
    system_prompt: Option<&str>,
    context: Option<Arc<dyn ToolContext>>,
    store: Option<Arc<dyn ToolStore>>,
    response_format: Option<Box<dyn StructuredOutputStrategy>>,
    middleware: Option<Vec<Arc<dyn Middleware>>>,
    file_backend: Option<Arc<dyn crate::tools::FileBackend>>,
) -> Result<UnifiedAgent, AgentError> {
    // Detect and create LLM from model string
    let llm = detect_and_create_llm(model)?;

    // Create agent with system prompt if provided
    let prefix = system_prompt
        .map(|s| s.to_string())
        .unwrap_or_else(|| "You are a helpful assistant.".to_string());

    let agent: Box<dyn Agent> = Box::new(
        ConversationalAgentBuilder::new()
            .tools(tools)
            .prefix(prefix)
            .build(llm)?,
    );

    let mut unified_agent = UnifiedAgent::new(agent);

    if let Some(ctx) = context {
        unified_agent = unified_agent.with_context(ctx);
    }

    if let Some(s) = store {
        unified_agent = unified_agent.with_store(s);
    }

    if let Some(rf) = response_format {
        unified_agent = unified_agent.with_response_format(rf);
    }

    if let Some(mw) = middleware {
        unified_agent = unified_agent.with_middleware(mw);
    }

    if let Some(fb) = file_backend {
        unified_agent = unified_agent.with_file_backend(Some(fb));
    }

    Ok(unified_agent)
}

/// Create an agent from an existing LLM instance.
///
/// This function is useful when you want to use a custom LLM configuration
/// instead of the model string detection.
pub fn create_agent_from_llm<L: Into<Box<dyn LLM>>>(
    llm: L,
    tools: &[Arc<dyn Tool>],
    system_prompt: Option<&str>,
) -> Result<UnifiedAgent, AgentError> {
    let prefix = system_prompt
        .map(|s| s.to_string())
        .unwrap_or_else(|| "You are a helpful assistant.".to_string());

    let agent: Box<dyn Agent> = Box::new(
        ConversationalAgentBuilder::new()
            .tools(tools)
            .prefix(prefix)
            .build(llm)?,
    );

    Ok(UnifiedAgent::new(agent))
}

/// Create an agent from an LLM with full runtime (context, store, middleware, response format).
///
/// Used by Deep Agent when creating from a custom LLM instance.
pub fn create_agent_with_runtime_from_llm<L: Into<Box<dyn LLM>>>(
    llm: L,
    tools: &[Arc<dyn Tool>],
    system_prompt: Option<&str>,
    context: Option<Arc<dyn ToolContext>>,
    store: Option<Arc<dyn ToolStore>>,
    response_format: Option<Box<dyn StructuredOutputStrategy>>,
    middleware: Option<Vec<Arc<dyn Middleware>>>,
    file_backend: Option<Arc<dyn crate::tools::FileBackend>>,
) -> Result<UnifiedAgent, AgentError> {
    let prefix = system_prompt
        .map(|s| s.to_string())
        .unwrap_or_else(|| "You are a helpful assistant.".to_string());

    let agent: Box<dyn Agent> = Box::new(
        ConversationalAgentBuilder::new()
            .tools(tools)
            .prefix(prefix)
            .build(llm)?,
    );

    let mut unified_agent = UnifiedAgent::new(agent);

    if let Some(ctx) = context {
        unified_agent = unified_agent.with_context(ctx);
    }

    if let Some(s) = store {
        unified_agent = unified_agent.with_store(s);
    }

    if let Some(rf) = response_format {
        unified_agent = unified_agent.with_response_format(rf);
    }

    if let Some(mw) = middleware {
        unified_agent = unified_agent.with_middleware(mw);
    }

    if let Some(fb) = file_backend {
        unified_agent = unified_agent.with_file_backend(Some(fb));
    }

    Ok(unified_agent)
}