Entourage

An open-source multi-agent orchestrator for autonomous coding tasks

Entourage is a transparent, interactive coding assistant that solves complex software engineering problems through multi-agent orchestration — think of it as an open-source alternative to Warp's Oz.

What makes Entourage different

🤖 Multi-Agent Orchestration

Instead of a single monolithic AI, Entourage coordinates multiple specialized agents that collaborate to solve coding problems. Each agent focuses on specific tasks (analysis, implementation, testing, refactoring) and works together systematically.

👁️ Real-Time Transparency

Watch every decision unfold on an interactive dashboard. See what the agents are thinking, which tools they're using, and how they're approaching your problem — all in real-time.

🎯 Human-in-the-Loop Control

Step in anytime to guide the agents if they're going off track. Provide feedback, redirect efforts, or approve critical decisions before they execute.

🔒 Sandboxed Execution

All code runs in isolated environments (Docker/Firecracker), so experimentation is safe by default. No risk to your local system.

Core Features

• Multi-agent collaboration — Specialized agents working together on complex tasks
• Live decision tracking — Real-time dashboard showing agent reasoning and actions
• Interactive intervention — Guide agents mid-execution to stay on track
• Safe execution — Sandboxed environments prevent unintended consequences
• Tool-use transparency — See exactly which tools and APIs agents invoke
• Progressive enhancement — Agents learn from corrections and adapt

Architecture

User Request
    ↓
Orchestrator (coordinates agent workflow)
    ↓
┌─────────────┬──────────────┬─────────────┬──────────────┐
│  Analyzer   │ Implementer  │   Tester    │  Refactorer  │
│   Agent     │    Agent     │    Agent    │    Agent     │
└─────────────┴──────────────┴─────────────┴──────────────┘
    ↓
Sandbox Environment (Docker/Firecracker)
    ↓
Execution Results + Decision Log

Sandbox Strategy

Entourage progressively hardens execution environments:

1. subprocess — Local process isolation (development)
2. Docker — Containerized execution (current)
3. Firecracker — Lightweight microVM sandboxes (planned)

Interfaces

• CLI — Terminal-based interaction for quick requests
• API — HTTP server for programmatic access and integration
• Dashboard — Real-time web UI for monitoring agent decisions

Getting Started

# Requires Python 3.14+
git clone https://github.com/yourusername/entourage
cd entourage
pip install -e .

# Start the orchestrator
python main.py

# Or run via API server
python api.py

Example Usage

# CLI mode
entourage "Refactor the authentication module to use JWT tokens"

# Watch agents:
# 1. Analyzer examines current auth implementation
# 2. Implementer proposes JWT integration strategy
# 3. You intervene: "Use the jose library, not pyjwt"
# 4. Implementer adapts and generates code
# 5. Tester creates test cases
# 6. Refactorer optimizes final implementation

Why Open Source?

Warp's Oz is powerful but proprietary. Entourage brings similar multi-agent orchestration capabilities to the open-source community with full transparency:

• See the reasoning — No black box decisions
• Control the process — Intervene and guide at any point
• Extend the system — Add your own specialized agents
• Own your data — Everything runs locally or in your infrastructure

Roadmap

• Multi-agent orchestration framework
• Real-time decision logging
• Subprocess sandbox isolation
• Docker-based sandboxing
• Interactive web dashboard
• Human intervention checkpoints
• Firecracker microVM support
• Custom agent plugin system
• Multi-language support (beyond Python)
• Cloud deployment options

Status

Early development — core orchestration and sandboxing functional, dashboard in progress.