Agentic Workflow Engine

An operating system for agents — a general-purpose engine that decomposes ANY complex goal into a dynamic DAG (Directed Acyclic Graph) of subtasks, executes them with parallelism, handles failures through replanning, tracks costs, and includes human-in-the-loop checkpoints.

Features

• DAG-based Execution: Accepts natural language goals, decomposes them into atomic parallel subtasks using a DeepSeek-based planner.
• Parallel Subtasks: Runs independent tasks concurrently using AsyncIO.
• Adaptive Execution: When failure occurs, chooses between RETRY, REPLAN, or ESCALATE to adapt at runtime.
• Cost Awareness: Tracks model costs continuously to enforce budget constraints across execution phases.
• Human In The Loop: Requests human approval before high-stakes operations or when a budget is exceeded.
• DeepSeek Integration: Uses the reliable deepseek-chat model for robust planning and execution operations.

Technology Stack

• Python 3.10+
• DeepSeek API: deepseek-chat model replacing expensive legacy models.
• NetworkX: Verifies DAG topological bounds and detects cycles.
• Pydantic: Heavily utilized for data validation and parsing JSON outputs.

Setup & Run

1. Clone the repository
2. Set up the Environment (Requires Python 3.10+)

   pip install -r requirements.txt

3. Copy .env.example to .env and assign your API key. Make sure DEEPSEEK_API_KEY is present.
4. Run the Engine CLI:

   python main.py run "Evaluate three top web scraping libraries for python, compare their features and write a summary."

Engine Architecture

• Planner: GoalDecomposer, DAGValidator, CostEstimator, DynamicReplanner
• Executor: ParallelExecutor, WorkflowScheduler, DependencyResolver
• Failure Engine: Error Classification and Strategies (RETRY, REPLAN, SKIP, ESCALATE)
• Monitoring: Live event traces with Cost Analysis.

Subtask Node Execution Example (CLI Trace Representation)

▶ Executing...
[0.0s] 🚀 Started: research_competitors, research_market, analyze_strengths (parallel)
[3.2s] ✅ research_competitors completed ($0.04, 823 tokens)
[4.1s] ✅ analyze_strengths completed ($0.03, 612 tokens)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.