AgentFlow

AgentFlow is a lightweight workflow engine for multi-agent systems.

It allows developers to define task graphs, register agents with capabilities, and execute workflows with dependency-aware scheduling.

Why this exists

Most agent frameworks focus on prompts, tools, or messaging.

They do not provide a clear execution model for:

• task dependencies
• workflow scheduling
• agent capability matching
• execution order tracing

AgentFlow explores that missing layer.

Quick Start

git clone https://github.com/joshuamlamerton/agentflow
cd agentflow
python examples/demo.py

Demo

The demo shows:

• a workflow with dependent tasks
• two agents with different capabilities
• a scheduler assigning tasks in dependency order
• an execution trace printed to the console

Architecture

flowchart TB

A[Workflow Definition]
B[Scheduler]
C[Task Queue]
D[Agent Registry]
E[Worker Agent A]
F[Worker Agent B]

A --> B
B --> C
B --> D
D --> E
D --> F
C --> E
C --> F

Loading

Repository Structure

agentflow

README.md
LICENSE

docs
  architecture.md

core
  workflow.py
  agent.py
  scheduler.py

examples
  demo.py

tests
  test_basic.py

Roadmap

Phase 1
Basic workflow and dependency execution

Phase 2
DAG visualization and trace export

Phase 3
Distributed task routing

Phase 4
Framework integrations

License

Apache 2.0