SCAVENGERS ASSEMBLE
⭐ Star GitHub ⚡ Quickstart
NEXT-GEN AGENTIC AI CODING HARNESS

Turn Your AI Models into an
Autonomous Superhero Strike Team

Scavengers Assemble orchestrates your favorite LLMs (Gemini, Claude, GPT, Ollama) in an asynchronous Directed Acyclic Graph (DAG) with full interactive simulation, JARVIS voice comms, and direct disk code synthesis.

⚡ Quickstart Setup Guide ⚔️ See Why We're Different 🚀 View Source (MIT)
⚡ Install & Launch Dashboard Procedure
Follow the 3-step setup procedure below to build and initiate your local mission control dashboard:
# Step 1: Clone repo & install dependencies
$ git clone https://github.com/shubhransh-gupta/scavengers-assemble.git
$ cd scavengers-assemble && npm install
# Step 2: Compile TypeScript source
$ npm run build
# Step 3: Initiate and open the local app dashboard
$ npm start # ➔ Initiates dashboard at http://localhost:3000
scavengers-mission-control · 3D Simulation Engine · Taj Mahal & Stark Spire
ONLINE: 5 AGENTS
Scavengers Assemble Mission Control Simulation HUD

Explore Scavengers Assemble

Everything you need to configure your superhero harness, understand our DAG architecture, and see why we outperform single-agent chatbots.

How to Set Up & Launch Scavengers Assemble

Clone the repository, build TypeScript, and launch the autonomous coding war room in under 60 seconds.

Option A: Clone & Compile TypeScript (Recommended)

Installs all dependencies, compiles TypeScript, and boots the local Mission Control server.

Recommended
Terminal (bash / zsh / powershell)
# 1. Clone repository from GitHub
$ git clone https://github.com/shubhransh-gupta/scavengers-assemble.git
$ cd scavengers-assemble

# 2. Install dependencies & compile TypeScript
$ npm install
$ npm run build

# 3. Launch Mission Control HUD (opens http://localhost:3000)
$ npm start

💡 Next Step: Open your local browser at http://localhost:3000/ to enter Mission Control, enter your directive, or speak via JARVIS voice comms!

PREREQUISITE 1
Node.js >= 18.0.0
Compatible with Node 18, 20, 22 LTS on macOS, Linux, and Windows.
PREREQUISITE 2
Modern Web Browser
Chrome, Safari, Firefox, Edge or Brave with Canvas 2D support.
PREREQUISITE 3
At least 1 LLM Key or Ollama
Use free Gemini API keys, OpenAI, Claude, or run 100% offline with Ollama.

How Scavengers Assemble is Radically Different

Most AI tools force you into a single text chatbox or charge \$20-\$500/mo for a sequential single-agent loop. Scavengers turns your models into an asynchronous strike team with zero vendor lock-in.

🦾
Parallel DAG vs. Sequential Bottlenecks

Tools like Devin or Copilot run single-threaded loops that stall on errors. Tony Stark deconstructs your goal into an asynchronous DAG where Spider-Man (UI), Hulk (Logic), and Thor (DevOps) build concurrently in parallel.

🌍
Full Visual Simulation & Comms

Watch your agents execute tasks with real-time thought logs, multi-lingual audio synthesis, and interactive visual feedback instead of watching a blank spinner.

📁
Direct Local Disk Codebase Output

No sandboxed ephemeral VMs. Scavengers creates full, runnable projects directly on your local disk at ./workspace/<slug>/ complete with package manifests and copyable terminal run scripts.

💎
100% Free & Open-Source (MIT)

Zero subscription fees. Bring your own keys (Gemini, Claude, OpenAI) or use 100% free offline local models with Ollama (DeepSeek-R1, Llama 3.3). You own all generated code.

Comprehensive Capability Comparison Matrix

Feature / Dimension 🦾 Scavengers Assemble Cursor / Copilot Devin / AutoGPT ChatGPT / Claude Web
Agent Architecture ✓ 5-Hero Parallel DAG + 5 Reserve ✗ Single-turn Autocomplete ✗ Sequential Single Agent ✗ Single Chat Turn
Visual Interface ✓ 3D Isometric Simulation Engine ✗ Code Editor Sidebar ✗ Web Terminal Log ✗ Plain Text Markdown
Multilingual Voice Comms ✓ JARVIS Mic Translation (Hindi/Eng/Esp) ✗ None ✗ None Voice chat only (no code DAG)
Local Disk Code Emission ✓ Full runnable repos (./workspace/) Inline file edits only Remote Cloud VM Sandbox ✗ Copy-paste code blocks
Vibranium QA & AST Verifier ✓ Captain America & Dr. Strange Basic Linter check CLI error loop retry ✗ None
LLM Provider Freedom ✓ Gemini, Claude, OpenAI, Ollama (Zero Lock) Proprietary Proxy Fixed Model Provider Specific
Pricing & License ✓ 100% Free & MIT Open Source $20 / month $500 / month / Waitlist $20 / month

Goal-Oriented Decomposition (GOD) DAG Architecture

How Scavengers Assemble deconstructs high-level human directives into production-grade multi-file repositories with zero hallucination.

1
Directive Input
Voice comms (Hindi/English) or text prompt submitted to Stark Tower.
2
Tony Stark Blueprint
GOD Orchestrator parses requirements, structures file graph & task DAG.
3
Parallel Dispatch
Spider-Man (UI), Hulk (Logic) & Thor (Builds) execute simultaneously.
4
Vibranium QA Shield
Captain America verifies typing, boundary conditions & syntax integrity.
5
Disk Generation
Complete project files emitted to ./workspace/ with run scripts.

⚡ Asynchronous Inter-Agent Mesh Communication Bus

Heroes do not operate in silos. If The Hulk exposes an exported TypeScript interface (e.g., interface UserSession), the Stark Comms Bus broadcasts that contract in real-time to Spider-Man for frontend state binding and to Thor for container environment variables, preventing schema drift before code is even written.

Earth's Mightiest Agent Strike Team: 5 Core + 5 Multiverse Specialists

Each superhero agent is backed by specialized system prompts, tailored tools, and distinct cognitive tasks.

Tony Stark Iron Man CORE LEAD
Tony Stark
GOD Lead Architect
Breaks master requirements into modular file graphs and orchestrates parallel subagent execution.
DAG Planner Synthesis
Spider-Man CORE HERO
Spider-Man
Frontend UI/UX Specialist
Builds ultra-clean, reactive UI components in React, Tailwind, Next.js, and SwiftUI with fluid animations.
React Tailwind Glassmorphism
Thor CORE HERO
Thor
DevOps & Infrastructure
Forges Dockerfiles, package manifests (package.json), build pipelines, and terminal run scripts.
Docker CI/CD Scripts
The Hulk CORE HERO
The Hulk
Core Logic & Algorithms
Smashes computational bottlenecks, writes complex business logic, state machines, and high-speed services.
Algorithms State Logic Database
Captain America CORE HERO
Captain America
Vibranium QA Shield
Enforces strict TypeScript safety, boundary unit tests, and production code compliance before disk write.
Unit Tests Type Safety QA Shield
Doctor Strange MULTIVERSE
Doctor Strange
Temporal AST & Memory
Manipulates Abstract Syntax Trees and preserves cross-turn context memory across long coding sessions.
AST Memory Stone
Black Widow MULTIVERSE
Black Widow
Security & Vulnerabilities
Performs deep threat modeling, audits dependencies for CVEs, and prevents secret leaks or injection attacks.
Security CVE Patching
Doctor Doom MULTIVERSE
Doctor Doom
Compilers & Low-Level
Optimizes Rust, C++, WebAssembly, and native binary bindings for maximum silicon efficiency.
Rust / C++ Wasm
Thanos MULTIVERSE
Thanos
Rate Limiter & Balancer
Balances API quota utilization across models so token limits are never exceeded during large builds.
Load Balancing Quota Manager
Kang the Conqueror MULTIVERSE
Kang the Conqueror
Timeline Simulator
Simulates multiple architectural forks simultaneously to pick the fastest and cleanest implementation.
Forks Simulation

Universal Multi-LLM Provider Engine

Route tasks to the best model for the job with live health check ping latency and instant fallback support.

Google Gemini
Ultra Fast

Gemini 2.0 Flash / Pro & Gemini 1.5. Massive 1M-2M context window with near-instant sub-second token generation.

Default: gemini-2.0-flash
Anthropic Claude
Deep Reasoning

Claude 3.5 / 3.7 Sonnet & Opus. Outstanding precision for complex architectural design, AST logic, and code refactoring.

Default: claude-3-5-sonnet-latest
OpenAI
Versatile

GPT-4o, GPT-4o-mini & o3-mini. Reliable code structure, OpenAPI schema generators, and unit test compliance.

Default: gpt-4o
Local Ollama
100% Offline

DeepSeek-R1, Llama 3.3, Qwen 2.5 Coder, Mistral. Zero cloud requests, full privacy, and zero token costs.

Default: deepseek-r1:latest

Interactive Mission Command Generator

Pick your stack and desired project to generate the customized execution command for your terminal.

Generated Terminal Command
node bin/stark.js hud --port 3000

Engineered for Maximum Speed & Developer Delight

Built with modern web standards, HTML5 canvas, and rigorous type safety.

🌍
3D Simulation Engine
Pure HTML5 Canvas isometric world rendering landmarks, monuments, and bridges with zero third-party 3D bundle bloat.
🎙️
JARVIS Voice Comms
Speak in Hindi, English, Spanish, or your native language. Our live microphone pipeline translates your voice directly into production engineering directives.
📁
Workspace File Explorer
Browse generated codebases on local disk, inspect syntax-highlighted source code, copy file paths, and execute terminal launch scripts with 1 click.
🔨
Interactive Simulation
Inspect real-time agent execution with interactive ghost preview cursors, particle FX, and multi-threaded agent status.
🎨
Multi-Theme Engine
Seamlessly toggle between ☀️ Daylight Theme (crisp white & sky blue), ⛅ Modern Slate, and 🌙 Dark Stark HUD with instant localStorage sync.
🧠
Personal Skills Loader
Register your own custom prompt skills, architecture patterns, and tools into the Avengers harness to personalize code generation standards.

Frequently Asked Questions

Answers to common questions about setup, safety, and agent orchestration.

Is Scavengers Assemble completely free to use? ▼
Yes! Scavengers Assemble is 100% free and open-source under the permissive MIT License. There are no monthly paywalls or subscription tiers. You can use free Google Gemini API tiers, bring your own API keys, or run entirely offline with Ollama models.
Where does Scavengers Assemble save the generated code? ▼
Generated projects are written directly to your local filesystem inside the ./workspace/<project-slug>/ directory. Each workspace includes full source files, package manifests, and copyable terminal execution scripts (e.g. npm install && npm start).
Can I run Scavengers Assemble without an internet connection? ▼
Yes! With our local Ollama integration, you can connect models such as DeepSeek-R1, Llama 3.3, or Qwen 2.5 Coder running on http://localhost:11434. All multi-agent orchestration will run entirely locally on your machine.
How does JARVIS voice comms translate multilingual audio? ▼
When you speak into the microphone (in Hindi, English, Spanish, etc.), the audio is transcribed and synthesized via our language model pipeline to distill the spoken intent into a formal engineering task graph before Tony Stark initiates decomposition.
How do I add my own custom personal skills or hero agents? ▼
You can add custom hero classes in src/heroes/ extending BaseHero or drop prompt skill files into the registered skills directory. The orchestrator automatically discovers registered agents and delegates tasks matched to their declared competencies.