Conductor

Orchestrate AI coding agents like a conductor leads an orchestra.

Conductor coordinates Claude, Codex, Gemini, and other AI CLI tools to tackle complex development tasks through intelligent delegation, parallel execution, and result aggregation.

Repository | Neul Labs | Quick Start

---

Why Conductor?

AI CLI tools like Claude Code and Codex are not just text generators - they are agents that perform actions: editing files, running commands, making commits, and using tools.

Conductor acts as a supervisor that:

• Delegates tasks to the best-suited agent
• Orchestrates complex workflows across multiple agents
• Aggregates results and tracks all actions taken

                    ┌─────────────────┐
                    │   Supervisor    │
                    │   (Orchestrator)│
                    └────────┬────────┘
                             │
        ┌────────────────────┼────────────────────┐
        │                    │                    │
        ▼                    ▼                    ▼
┌───────────────┐    ┌───────────────┐    ┌───────────────┐
│  Claude Agent │    │  Codex Agent  │    │  Gemini Agent │
│  (Refactoring)│    │  (Testing)    │    │  (Research)   │
└───────────────┘    └───────────────┘    └───────────────┘

---

Quick Start

Installation

pip install neul-conductor

Installs the agent command-line tool.

Run Your First Task

# Best-matched agent handles the task
agent run "Fix the authentication bug in auth.py"

# Use a specific agent
agent run --agent codex "Write tests for the API"

Orchestrate Multi-Agent Workflows

# Sequential pipeline
agent orchestrate \
    "Research best practices for rate limiting" \
    "Implement rate limiting middleware" \
    "Write comprehensive tests"

# Parallel review from multiple perspectives
agent orchestrate --pattern parallel \
    "Review src/api.py for security, performance, and maintainability"

# Consensus validation
agent orchestrate --pattern consensus --threshold 0.8 \
    "Is this database migration safe for production?"

---

Features

Orchestration Patterns

Pattern	Description	Use Case
Sequential	Pipeline execution with context flow	Multi-step development
Parallel	Concurrent execution across agents	Code review, analysis
Consensus	Multi-agent validation	Safety-critical changes
Handoff	Dynamic task routing	Complex problems

Supported Agents

Agent	Strengths	Best For
Claude	Complex refactoring, multi-file changes	Architecture, synthesis
Codex	Broad knowledge, sandbox execution	Full-stack, testing
Gemini	Long context (1M tokens)	Research, large codebases

---

Python API

import asyncio
from agent.core.task import Task, TaskType
from agent.agents import ClaudeAgent, CodexAgent
from agent.orchestration import Supervisor, OrchestrationPattern

async def main():
    supervisor = Supervisor([ClaudeAgent(), CodexAgent()])

    result = await supervisor.orchestrate(
        tasks=[
            Task(prompt="Research auth patterns", task_type=TaskType.RESEARCH),
            Task(prompt="Implement JWT auth", task_type=TaskType.CODE_CREATE),
            Task(prompt="Write tests", task_type=TaskType.TEST_WRITE),
        ],
        pattern=OrchestrationPattern.SEQUENTIAL,
    )

    print(f"Success: {result.success}")
    print(f"Files modified: {result.all_files_modified}")

asyncio.run(main())

---

Documentation

• Getting Started - Installation and quick start
• Core Concepts - Agents, Tasks, Results
• CLI Reference - All commands with examples
• Python API - Full API documentation
• Orchestration Patterns - Sequential, Parallel, Consensus
• Creating Custom Agents - Extend with your own agents

Build Documentation Locally

pip install -e ".[docs]"
cd documentation && mkdocs serve

---

Requirements

• Python 3.11+
• At least one AI CLI tool:
  • Claude Code
  • Codex CLI
  • Gemini CLI

---

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Lint
ruff check src/

# Type check
mypy src/

---

Contributing

Contributions welcome! See Contributing Guide for guidelines.

Priority areas:

• Additional agent implementations (Gemini, Aider, Continue)
• Orchestration patterns (Handoff, Magentic)
• Documentation and examples

---

License

MIT License - see LICENSE for details.

---

Credits

Inspired by:

• Datasette LLM - Plugin architecture
• Microsoft Agent Orchestration Patterns

---

Part of the Neul Labs toolchain

Conductor is part of the Neul Labs orchestration toolchain:

Project	Description
brat	Multi-agent harness for AI coding tools — crash-safe state, parallel execution.
ringlet	One CLI to rule all your coding agents.
fastworker	Background tasks in Python with zero infrastructure — no Redis, no RabbitMQ.
m9m	The n8n alternative without the bugs — one Go binary.

Learn more at neullabs.com.