ai-agent-engine is a lightweight orchestration engine for AI agents. Workflows are defined declaratively in YAML and executed through a multi-LLM router that selects the optimal model for each step.

graph TB
    Client[Client / API]
    API[FastAPI Gateway]
    Config[YAML Config Loader]
    Router[LLM Router]
    Orchestrator[Workflow Orchestrator]

    Planner[Planner Agent]
    Executor[Executor Agent]
    Validator[Validator Agent]

    OpenAI[OpenAI API]
    Anthropic[Anthropic API]
    Local[Local LLM]

    Safety[Safety Layer]
    Observe[Observability]

    Client --> API
    API --> Config
    Config --> Orchestrator
    Orchestrator --> Router
    Router --> OpenAI
    Router --> Anthropic
    Router --> Local
    Orchestrator --> Planner
    Orchestrator --> Executor
    Orchestrator --> Validator
    Safety --> Orchestrator
    Observe --> Orchestrator

• FastAPI Gateway — HTTP entry point, request validation, routing to orchestrator
• YAML Config Loader — Parses and validates workflow definitions
• Workflow Orchestrator — Resolves step dependencies (DAG), executes steps sequentially or in parallel
• LLM Router — Selects optimal model per task (cost-aware, fallback chains)
• Agent Patterns — Built-in patterns: Router, Planner, Executor, Validator
• Safety Layer — Rate limiting, cost controls, content filtering
• Observability — Structured logging, metrics, trace ID propagation

• src/agent_engine/api/ — FastAPI routes
• src/agent_engine/core/ — Orchestrator, Router, Agent patterns
• src/agent_engine/config/ — YAML loader, validation
• src/agent_engine/llm/ — LLM providers (OpenAI, Anthropic, Local)
• src/agent_engine/safety/ — Rate limiting, cost controls
• src/agent_engine/observe/ — Logging, metrics, tracing
• src/agent_engine/models/ — Pydantic models
• src/agent_engine/services/ — Business logic services
• workflows/ — Example YAML workflow configs