Crate llm_orchestrator_core 

LLM Orchestrator Core - Workflow orchestration engine for LLM pipelines.

This crate provides the core functionality for defining, validating, and executing multi-step LLM workflows with support for DAG-based execution, parallel processing, and fault tolerance.

Example

use llm_orchestrator_core::{Workflow, ExecutionContext};
use serde_json::json;

// Load workflow from YAML
let yaml = r#"
name: "simple-workflow"
steps:
  - id: "step1"
    type: "llm"
    provider: "openai"
    model: "gpt-4"
    prompt: "Say hello to {{ name }}"
    output: ["greeting"]
"#;

let workflow = Workflow::from_yaml(yaml)?;

// Validate workflow
workflow.validate()?;

// Create execution context
let mut inputs = std::collections::HashMap::new();
inputs.insert("name".to_string(), json!("World"));
let ctx = ExecutionContext::new(inputs);

// Render template with context
let prompt = ctx.render_template("Say hello to {{ name }}")?;
assert_eq!(prompt, "Say hello to World");

Re-exports

pub use context::ExecutionContext;

pub use dag::WorkflowDAG;

pub use error::OrchestratorError;

pub use error::Result;

pub use executor::StepResult;

pub use executor::StepStatus;

pub use executor::WorkflowExecutor;

pub use retry::RetryExecutor;

pub use retry::RetryPolicy;

pub use workflow::Workflow;

pub use workflow::Step;

pub use workflow::StepType;

pub use workflow::StepConfig;

pub use workflow::LlmStepConfig;

pub use workflow::EmbedStepConfig;

pub use workflow::VectorSearchConfig;

pub use workflow::TransformConfig;

pub use workflow::ActionConfig;

pub use workflow::ParallelConfig;

pub use workflow::BranchConfig;

pub use workflow::RetryConfig;

pub use workflow::BackoffStrategy;

Modules

context

Execution context management for workflows.

dag

DAG (Directed Acyclic Graph) builder and validator for workflows.

error

Error types for the LLM Orchestrator core.

executor

Workflow execution engine with async Tokio runtime.

executor_state

State persistence integration for WorkflowExecutor.

health

Health check functionality for monitoring system status.

metrics

Prometheus metrics instrumentation for workflow orchestration.

providers

Provider trait definitions (re-exported from llm-orchestrator-providers).

retry

Retry logic with exponential backoff.

workflow

Workflow definition types.

Structs

CompletionRequest

Completion request.

CompletionResponse

Completion response.

Enums

ProviderError

Provider error.

Constants

NAME

Library name.

VERSION

Library version.

Traits

LLMProvider

LLM provider trait.