agentic-crew

agentic-crew

Python

Define AI crews in YAML. Run them on CrewAI, LangGraph, or Strands — whichever is installed. One format, any framework.

The Problem

The AI agent ecosystem is fragmented. CrewAI, LangGraph, and Strands each have different APIs for the same concepts — agents, tasks, and orchestration. Switching frameworks means rewriting your crew definitions. And single-agent CLI tools (aider, claude-code, ollama) each have their own invocation patterns.

The Solution

agentic-crew gives you:

1. Universal crew format — YAML definitions that work across all three frameworks
2. Runtime auto-detection — Automatically uses whichever framework is installed
3. Universal CLI runner — One interface for aider, claude-code, ollama, codex, kiro, and goose
4. Custom runner support — Define your own CLI tools in YAML

Installation

# Core (auto-detects framework at runtime)
pip install agentic-crew

# With a specific framework
pip install agentic-crew[crewai]      # CrewAI (recommended)
pip install agentic-crew[langgraph]   # LangGraph
pip install agentic-crew[strands]     # AWS Strands

# All frameworks
pip install agentic-crew[ai]

Quick Start

1. Define a Crew

.crewai/manifest.yaml

name: my-package
version: "1.0"
crews:
  analyzer:
    description: Analyze codebases
    agents: crews/analyzer/agents.yaml
    tasks: crews/analyzer/tasks.yaml

crews/analyzer/agents.yaml

code_reviewer:
  role: Senior Code Reviewer
  goal: Find bugs and improvements
  backstory: Expert at code analysis

crews/analyzer/tasks.yaml

review_code:
  description: Review the provided code for issues
  expected_output: List of findings with severity
  agent: code_reviewer

2. Run It

from agentic_crew import run_crew

# Auto-detects best framework (CrewAI > LangGraph > Strands)
result = run_crew("my-package", "analyzer", inputs={"code": "..."})

# Or from the CLI
agentic-crew run my-package analyzer --input "Review this code"

Single-Agent CLI Runners

For simpler tasks, run CLI-based coding tools through a unified interface:

Runner	Description	Local?
aider	AI pair programming	No
claude-code	Anthropic’s coding agent	No
codex	OpenAI’s coding agent	No
ollama	Free local LLM execution	Yes
kiro	AWS AI development CLI	No
goose	Block’s extensible agent	No

# Aider for quick fixes
agentic-crew run --runner aider --input "Add error handling to auth.py"

# Claude Code for refactoring
agentic-crew run --runner claude-code --input "Refactor the database module"

# Ollama for free local execution
agentic-crew run --runner ollama --input "Fix the bug" --model deepseek-coder

When to Use What

Scenario	Use
Complex multi-step tasks needing collaboration	Multi-agent crew
Quick edits, simple file changes	Single-agent CLI runner
Local development without API keys	Ollama runner
Tasks requiring planning and delegation	Multi-agent crew

Framework Auto-Detection

agentic-crew checks what’s installed and picks the best option:

from agentic_crew.core.decomposer import detect_framework, get_runner

framework = detect_framework()  # "crewai", "langgraph", or "strands"
runner = get_runner()           # Auto-selects best available
runner = get_runner("langgraph") # Force a specific framework

Priority order:

1. CrewAI — Most features, best for complex crews
2. LangGraph — Good for flow-based orchestration
3. Strands — Lightweight, minimal dependencies

Programmatic API

from agentic_crew import CrewConfig, AgentConfig, TaskConfig, run_crew_auto

config = CrewConfig(
    name="review-crew",
    agents=[
        AgentConfig(
            name="security_reviewer",
            role="Security Expert",
            goal="Identify security vulnerabilities",
            backstory="10+ years of security auditing",
        ),
        AgentConfig(
            name="performance_reviewer",
            role="Performance Engineer",
            goal="Identify performance bottlenecks",
            backstory="Expert in optimization and profiling",
        ),
    ],
    tasks=[
        TaskConfig(
            name="security_review",
            description="Review code for security issues",
            agent="security_reviewer",
            expected_output="Vulnerabilities with severity ratings",
        ),
        TaskConfig(
            name="performance_review",
            description="Analyze code for performance issues",
            agent="performance_reviewer",
            expected_output="Bottlenecks with recommendations",
        ),
    ],
)

result = run_crew_auto(config, inputs={"code_path": "./src"})

Package Structure

Any package can include crew definitions:

my-package/
├── .crewai/
│   ├── manifest.yaml
│   ├── knowledge/
│   │   └── domain_docs/
│   └── crews/
│       └── my_crew/
│           ├── agents.yaml
│           └── tasks.yaml
└── src/

Development

# Install with dev deps
uv sync --extra dev --extra tests --extra crewai

# Run tests
uv run pytest tests/ -v

# Lint
uvx ruff check src/ tests/ --fix

Related

• CrewAI — Original crew framework
• LangGraph — Graph-based agents
• Strands — AWS agent framework
• @jbcom/agentic — Fleet management and orchestration
• @jbcom/agentic (Triage) — Triage primitives

Links

• GitHub
• PyPI