multi-agent-orchestrator

# 🚀 Multi-Agent AI Orchestrator (Enterprise Edition) Typing SVG

### **Open-source enterprise AI orchestration platform with visual workflow editor, RBAC, multi-provider LLM runtime, HITL, telemetry, and production-grade governance.** 8 specialized agents · competitor-grade feature parity · visual DAG editor + RBAC + enterprise controls [![Next.js](https://img.shields.io/badge/Next.js_16-black?style=for-the-badge&logo=next.js)](https://nextjs.org/) [![TypeScript](https://img.shields.io/badge/TypeScript_5-3178C6?style=for-the-badge&logo=typescript&logoColor=white)](https://www.typescriptlang.org/) [![Groq](https://img.shields.io/badge/Groq_Free_Tier-F55036?style=for-the-badge)](https://groq.com/) [![Vercel AI SDK](https://img.shields.io/badge/Vercel_AI_SDK_6-000000?style=for-the-badge&logo=vercel)](https://sdk.vercel.ai/) [![Zod](https://img.shields.io/badge/Zod_v4-3E67B1?style=for-the-badge)](https://zod.dev/) [![License: MIT](https://img.shields.io/badge/License-MIT-22C55E?style=for-the-badge)](LICENSE) [![GitHub stars](https://img.shields.io/github/stars/ParthivPandya/multi-agent-orchestrator?style=social)](https://github.com/ParthivPandya/multi-agent-orchestrator/stargazers) [![GitHub forks](https://img.shields.io/github/forks/ParthivPandya/multi-agent-orchestrator?style=social)](https://github.com/ParthivPandya/multi-agent-orchestrator/network) [![GitHub issues](https://img.shields.io/github/issues/ParthivPandya/multi-agent-orchestrator?color=d73a4a)](https://github.com/ParthivPandya/multi-agent-orchestrator/issues) [![GitHub last commit](https://img.shields.io/github/last-commit/ParthivPandya/multi-agent-orchestrator)](https://github.com/ParthivPandya/multi-agent-orchestrator/commits/main) [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](https://github.com/ParthivPandya/multi-agent-orchestrator/pulls) --- > 🌟 **If this project impresses you — and it will — please drop a ⭐ star. It's the best open-source AI orchestrator on GitHub.** 🌟 --- [🚀 Quick Start](#-quick-start) · [🏆 Competitor Parity](#-competitor-feature-parity--enterprise-readiness) · [🎨 Visual Pipeline Builder](#-visual-pipeline-builder-enterprise-ux) · [👥 RBAC & Team Controls](#-rbac--team-management-enterprise-controls) · [🤖 8 Agents](#-the-8-agents) · [🏗️ Architecture](#️-full-architecture) · [📡 API](#-api-reference) · [🗺️ Roadmap](#️-roadmap)

🔎 SEO Keywords

Multi-Agent AI Orchestrator, Enterprise AI Orchestration Platform, Visual Pipeline Builder, Drag and Drop AI Workflow Editor, LangFlow Alternative, Flowise Alternative, CrewAI Alternative, LangGraph Alternative, OpenDevin Alternative, RBAC for AI Agents, Human in the Loop AI, Multi-Provider LLM Orchestration, AI Agent Telemetry Dashboard, AI Workflow Automation, AI Governance Platform.

🏆 Why This Is the Best Repo on GitHub

“This is what enterprise tools promise but never deliver — and it’s completely free.”

🥇 What makes it extraordinary	Details
💸 Zero Cost	Runs 100% on Groq’s free tier. No OpenAI bills. No AWS. No subscriptions. Ever.
🤖 8 Specialized AI Agents	Each agent has a distinct role, skill set, and LLM — like a real software team
⚡ Real-time SSE Streaming	Watch every agent think and write code live — token by token
🛡️ Enterprise-Grade Security	OWASP Top 10 scan + GDPR/HIPAA/PCI-DSS/SOC 2 compliance built-in
🌊 14 Major Features	9 production gaps + 5 enterprise enhancements — more than most paid platforms
🌍 7 Languages Supported	TypeScript, Python, Go, Java, Rust, Ruby, PHP — with idiomatic prompt injection
🖼️ Vision-to-Code	Upload a Figma screenshot → get production component code in 30 seconds
📊 ROI Intelligence	Proves your AI investment with board-ready savings metrics — $savings, ROI multiple, sprint days
🏗️ Technical Debt Grading	Scores every codebase A–F across 6 debt categories with a prioritized backlog
⏸️ Human-in-the-Loop	Approve, reject, or request changes before code continues — like a real review gate
🧠 Learns Your Style	Remembers your preferred stack across sessions and injects it into every future run
💾 Never Lose Progress	Checkpoints every stage to disk — resume from exactly where you left off
🐙 Ships to GitHub	Pushes all generated files to a new GitHub repo in one click
🎨 Premium UI	Glassmorphism dark theme, micro-animations, and a design that puts enterprise tools to shame
🧩 Visual Pipeline Builder	Enterprise-grade drag-and-drop DAG editor with workflow catalog, node inspector, zoom controls, and branch visualization
👥 RBAC + Team Panel	Admin/member/viewer roles with permission-aware UI, team management panel, and enterprise policy visibility

🥇 Competitor Feature Parity + Enterprise Readiness

This repository now includes the missing capabilities commonly found in paid platforms:

✅ Visual / Low-Code Workflow Editor (LangFlow / Flowise style)
✅ Human-in-the-Loop approvals with pause/resume and decision endpoint
✅ Multi-provider LLM runtime (provider/model selection per agent at run time)
✅ Session memory + context persistence
✅ Observability + audit export dashboard
✅ Connectors + webhook-triggered orchestration
✅ Team & RBAC controls (admin, member, viewer) with enterprise UI
✅ Flexible DAG workflows with branching / parallel / merge nodes
✅ Agent diversity + tool execution (Product, UX, code execution sandbox)

In short: this is not just a linear demo pipeline anymore - it is a production-oriented, enterprise-ready orchestration workspace.

🎨 Visual Pipeline Builder (Enterprise UX)

Full-screen visual workflow experience with drag-and-drop node movement
Workflow catalog for standard / security-first / rapid-prototype DAGs
Node inspector panel (conditions, human checkpoints, branch targets)
Zoom controls and clean edge rendering for branch readability
Direct “Run Workflow” execution from the builder UI

This is designed to be familiar for teams using LangFlow, Flowise, or graph-based orchestration tools.

👥 RBAC + Team Management (Enterprise Controls)

Role-based permissions: admin, member, viewer
Team management panel with role assignment and member lifecycle controls
Permission-aware UI gating for pipeline execution, settings administration, and workflow access
Enterprise policy visibility in-app for governance clarity

📊 2026 Market Context — Why This Exists

Stat	Value	Source
AI-generated code worldwide	41% of all code	GitHub, 2026
Annual losses from technical debt	$370M (U.S. enterprise)	Synergy Labs
IT budget spent on legacy maintenance	Up to 80%	Synergy Labs
Developers concerned about AI security	56%	GitHub survey
AI programs missing goals in 2026	40% of organizations	IDC
Developer productivity gain from AI	20–45%	McKinsey
Enterprise apps with AI agents	40% by end of 2026	Gartner
Gartner projected AI code defects	+2,500% increase	Gartner, 2026

“AI-centric organizations are achieving 20–40% reductions in operating costs and 12–14 point increases in EBITDA margins.” — McKinsey via CIO.com, Feb 2026

This repo directly solves 8 of the biggest organizational pain points AI teams face in 2026.

✨ v3 — 9 Enterprise Gaps Closed

These are the features that turn a “cool AI demo” into a production-grade enterprise tool.

#	Feature	Impact	Status
1	⏸️ Human-in-the-Loop (HITL)	Approve / reject / request changes before pipeline continues. 10-min auto-approve fallback	✅ Live
2	🔑 Multi-Provider LLM	Groq (free), OpenAI, Anthropic, Ollama (local) — per-agent model selection	✅ Live
3	✅ Structured Output Validation	Zod schema validation at every agent handoff — non-blocking warnings shown in activity feed	✅ Live
4	🛡️ Security Reviewer Agent	Dedicated OWASP Top 10 scanner — CRITICAL/HIGH blocks deployment automatically	✅ Live
5	🐙 MCP GitHub Delivery	One-click push of all generated files to a new GitHub repo via GitHub REST API	✅ Live
6	⚡ Parallel Agent Execution	Testing Agent runs via `Promise.allSettled` — faster pipelines, non-blocking	✅ Live
7	📋 Full Audit Log Export	Every event timestamped. Export as JSON with one click	✅ Live
8	🧠 Agent Memory	Learns your tech preferences across sessions — injected into future prompts	✅ Live
9	🌊 Token Streaming	Developer Agent streams code character-by-character via `streamText`	✅ Live

⚡ Phase 1 Enterprise Features (New!)

Directly addressing the $370M technical debt crisis and the enterprise security gap.

🌍 Feature 5 — Multi-Language Pipeline (7 Languages)

The Developer Agent now generates idiomatic, production-ready code in any of these languages — auto-detected from your requirements:

Language	Icon	Frameworks	Test Framework	Auto-Detected Keywords
TypeScript	🟦	Next.js, Express, NestJS, Hono	Jest / Vitest	react, nextjs, typescript, web
Python	🐍	FastAPI, Django, Flask, Celery	pytest	python, fastapi, django, ml, ai
Go	🐹	Gin, Echo, Fiber, stdlib	testing + testify	go, golang, grpc, goroutine
Java	☕	Spring Boot 3, JPA, Security	JUnit 5 + Mockito	java, spring, enterprise, kafka
Rust	🦀	Axum, Tokio, SQLx, Serde	cargo test	rust, axum, wasm, systems
Ruby	💎	Rails 7, Sidekiq, Devise	RSpec + FactoryBot	ruby, rails, activerecord
PHP	🐘	Laravel 11, Symfony	Pest + PHPUnit	php, laravel, eloquent

How it works: Requirements Analyst output is scanned for language keywords → best match selected → language-specific idiom prompt injected into Developer Agent system prompt before code generation.

You: "Build a REST API for a todo app"
        ↓
Auto-detected: TypeScript (keywords: api, rest)
Language skill injected: "Use strict TypeScript, Zod validation,
                          async/await, separate route/service/repository layers..."
        ↓
Developer Agent writes idiomatic TypeScript code ✅

🏗️ Feature 2 — Technical Debt Scanner Agent

Runs automatically after every pipeline as a non-blocking parallel task. Scores your generated code across 6 debt categories and produces a prioritized business backlog.

🏗️ Grade: B  |  Score: 7.2/10  |  Est. 14 hours to remediate

6 Debt Categories Measured:

Category	What It Catches
🏛️ Architectural	God objects, tight coupling, missing abstractions, no DI
🧪 Testing	Missing tests, brittle assertions, no edge cases, no integration tests
📝 Documentation	No JSDoc, undocumented APIs, missing README
🔒 Security	Hardcoded secrets, missing validation, no rate limiting, SQL risks
📦 Dependency	Outdated packages, known CVEs, excessive deps
⚡ Performance	N+1 queries, missing caching, blocking async ops, memory leaks

Output:

{
  "debtScore": 7.2,
  "grade": "B",
  "summary": "Code is functional with good structure. Testing and documentation debt are the primary concerns.",
  "hotspots": [
    { "type": "testing", "severity": "high", "description": "No integration tests for auth routes", "effort": "M" }
  ],
  "backlogItems": [
    { "title": "Add Jest integration tests for /api/auth", "businessOutcome": "Prevents auth regressions in production", "storyPoints": 3 }
  ]
}

The enterprise blocker is gone. Every generated codebase is automatically checked against 6 major compliance frameworks.

Pipeline Flow:
Security Reviewer → Compliance Agent → Testing Agent → Deployment
                          ↓
              CRITICAL violation? → Block Pipeline 🔴
              HIGH violation?    → HITL Required 🟡  
              MEDIUM/LOW?        → Warning Only  🟢

Framework	What Gets Checked
OWASP Top 10	A01-Broken Access Control, A02-Crypto Failures, A03-Injection, A07-Auth Failures
GDPR	Data minimization, consent, right to erasure, cross-border transfers
HIPAA	PHI encryption, access controls, audit logs, minimum necessary standard
PCI-DSS	Cardholder data protection, network segmentation, key management
SOC 2	Security (CC6), Confidentiality (C1), Availability (A1), Privacy (P1-P8)
DPDP Act	Data fiduciary obligations, consent, data principal rights (India)

Unlocks fintech, healthcare, legal, and government sectors for your projects.

🖼️ Feature 3 — Vision-to-Code (Upload Design → Get Code)

The #1 sprint killer is building the wrong thing. Now you can show the AI exactly what you want.

You upload: Figma screenshot / wireframe / hand-drawn sketch / UI mockup
     ↓
POST /api/vision + your framework/language choices
     ↓
Llama 4 Maverick (vision model) analyzes layout, spacing, colors, typography
     ↓
You get: Complete, responsive, accessible component code ✅

Supported input types:

📸 Figma design exports (PNG, JPG, WEBP)
✏️ Hand-drawn wireframes (photographed)
🖥️ Existing UI screenshots (for replication or migration)
📄 PDF mockup pages

Options: Framework (React/Next.js/Vue/Svelte/Angular/Vanilla) · Language (7 supported) · Style Library (Tailwind/MUI/shadcn/Chakra/CSS Modules)

💰 Feature 7 — ROI Intelligence Dashboard

“We simply cannot afford more AI investments that operate in the dark.” — Flexera CIO, Jan 2026

Every pipeline run now comes with a board-ready ROI report to justify your AI investment.

Metrics Calculated Per Run:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  ROI Intelligence Dashboard
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Hours Saved:          23.5h    (93% time reduction)
  Net Savings:          $2,800   per feature
  LLM Cost:             $0.003   (Groq free tier)
  ROI Multiple:         933x
  Sprint Days Saved:    2.9 days
  Annualized Value:     $145,000 /yr projected

  Cumulative (all runs):
  Total Hours Saved:    187h
  Total Savings:        $22,400
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

History persists in localStorage. Cumulative savings grow with every run. Annualized value projected from your actual usage frequency.

🆕 On-Demand Analysis Endpoint

POST /api/analyze — run technical debt + compliance analysis on ANY code, anytime:

// Request
{
  "code": "your generated or existing code here",
  "language": "python",
  "runDebt": true,
  "runCompliance": true,
  "complianceFrameworks": ["OWASP_TOP10", "GDPR", "HIPAA"],
  "industry": "healthcare"
}

// Response
{
  "debtScan": { "report": { "grade": "B", "debtScore": 7.1, "hotspots": [...] } },
  "complianceScan": { "report": { "overallScore": 82, "overallStatus": "WARNING", "frameworks": [...] } }
}

🤖 The 8 Agents

#	Agent	Model	Role	Max Tokens
0	🧭 Router / Classifier	`llama-3.1-8b-instant`	Classifies intent → routes to optimal pipeline subset (4 modes)	512
1	🔍 Requirements Analyst	`llama-3.1-8b-instant`	Parses English → structured JSON specs + Zod validation	2,048
2	📋 Task Planner	`llama-4-scout-17b`	Breaks specs → prioritized task list with dependencies	2,048
3	💻 Developer Agent	`qwen/qwen3-32b`	Writes code · RAG-grounded · web search · 7-language support · live streaming	4,096
4	🔎 Code Reviewer	`llama-3.3-70b-versatile`	Reviews code · lint tool · HITL gate · up to 3 revision loops	2,048
5	🛡️ Security Reviewer	`llama-3.3-70b-versatile`	OWASP Top 10 scan · CRITICAL/HIGH blocks deployment	3,072
6	🧪 Testing Agent	`llama-3.3-70b-versatile`	Generates unit + integration tests · runs in parallel	3,072
7	🚀 Deployment Agent	`llama-3.1-8b-instant`	Dockerfile · docker-compose · GitHub Actions CI/CD	2,048

All 8 agents run with exponential backoff retry (3 attempts: 2s → 4s → 8s). Every agent call is logged to the audit log.

Pipeline Routing Modes

Mode	Agents	Tokens Saved	Best For
`FULL_PIPELINE`	All 8	—	New feature / app build
`QUICK_FIX`	Router + Dev + Reviewer	~60%	Bug fix, small tweak
`PLAN_ONLY`	Router + Analyst + Planner	~70%	Architecture planning
`CODE_REVIEW_ONLY`	Router + Reviewer	~80%	Paste code for review

🏗️ Full Architecture

flowchart TB
    subgraph Input["📥 Input — Multiple Modes"]
        TXT["📝 Text Requirement"]
        IMG["🖼️ Image / Wireframe (F3 NEW)"]
    end

    subgraph Frontend["🖥️ Frontend — Premium Dark UI"]
        Page["page.tsx — Main Orchestrator UI"]
        PV["PipelineView — 8 Agent Cards"]
        HITLM["⏸️ HITLModal (Gap #1)"]
        SP["⚙️ SettingsPanel (Gap #2)"]
        ST["🌊 Token Stream Preview (Gap #9)"]
        VU["🖼️ VisionUploader (F3 NEW)"]
        EAP["📊 EnhancedAnalyticsPanel (F7 NEW)"]
        WV["WorkspaceViewer"]
        HP["HistoryPanel"]
    end

    subgraph API["📡 API Routes"]
        OR["/api/orchestrate — SSE Pipeline"]
        HITLA["/api/hitl — HITL Decisions"]
        DEL["/api/deliver — GitHub Push"]
        VIS["/api/vision — Vision-to-Code (F3 NEW)"]
        ANA["/api/analyze — Debt + Compliance (F2/F4 NEW)"]
        WS["/api/workspace — File Save"]
    end

    subgraph Pipeline["🔄 Orchestration Pipeline v3.1"]
        O["orchestrator.ts"]

        R0["🧭 Router Agent\nclassifyIntent()"]

        subgraph LangDetect["🌍 Language Detection (F5 NEW)"]
            LD["detectLanguage()\ngetSkillPrompt()"]
        end

        subgraph Mem["🧠 Memory (Gap #8)"]
            MEM["memory.ts — localStorage prefs"]
        end

        subgraph Rag["📚 RAG"]
            RAG["retrieveRelevantChunks()\nTF-IDF cosine similarity"]
        end

        A1["🔍 Analyst\n+ Zod validation"]
        A2["📋 Planner\n+ Zod validation"]
        A3["💻 Developer\n+ language skill\n+ streamText"]
        A4["🔎 Code Reviewer\n+ lint tool\n+ Zod validation"]
        HITLG["⏸️ HITL Gate"]
        A5["🛡️ Security Reviewer\n+ OWASP scan"]
        A6["⚖️ Compliance Agent\nGDPR/HIPAA/PCI/SOC2 (F4 NEW)"]

        subgraph Parallel["⚡ Parallel Group (Gap #6)"]
            A7["🧪 Testing Agent"]
            DS["🏗️ Debt Scanner (F2 NEW)"]
        end

        A8["🚀 Deployment Agent"]

        subgraph ROI["💰 ROI (F7 NEW)"]
            ROIC["roi.ts — computeROI()"]
        end
    end

    subgraph Audit["📋 Audit Log (Gap #7)"]
        AL["Every event → AuditLog.log()"]
    end

    TXT --> Page
    IMG --> VU --> VIS
    Page --> OR
    OR --> O
    O --> R0 --> LangDetect --> Mem --> Rag
    Rag --> A1 --> A2 --> A3
    LangDetect -->|skill prompt| A3
    A3 --> A4
    A4 -->|CHANGES| A3
    A4 -->|APPROVED| HITLG
    HITLG --> A5 --> A6
    A6 -->|PASS| Parallel
    Parallel --> A8
    A8 --> ROI
    O --> AL
    OR --> HITLA
    OR --> DEL
    OR --> ANA

    style Input fill:#1a1225,color:#e0e0ff
    style Frontend fill:#1a1a2e,color:#e0e0ff
    style API fill:#16213e,color:#e0e0ff
    style Pipeline fill:#0f1e3d,color:#e0e0ff
    style Audit fill:#1e2a1a,color:#e0e0ff
    style ROI fill:#1e2a1a,color:#e0e0ff
    style Parallel fill:#1a2e1a,color:#e0e0ff
    style LangDetect fill:#1a1e2e,color:#e0e0ff
    style Mem fill:#2a1a2e,color:#e0e0ff
    style Rag fill:#1e1a2e,color:#e0e0ff

🚀 Quick Start

Prerequisites

Node.js 18+ — Download
Groq API Key — Free at console.groq.com

1. Clone & Install

git clone https://github.com/ParthivPandya/multi-agent-orchestrator.git
cd multi-agent-orchestrator/multi-agent-system
npm install

2. Configure

cp .env.example .env.local

# .env.local

# ✅ Required — get free key at https://console.groq.com
GROQ_API_KEY=gsk_your_api_key_here

# Optional — for additional LLM providers
# OPENAI_API_KEY=sk-...
# ANTHROPIC_API_KEY=sk-ant-...

Tip: OpenAI and Anthropic keys can also be entered in Settings → Providers tab at runtime. They stay in localStorage — never sent to any server.

3. Run

npm run dev
# → http://localhost:3000

4. Deploy (One Click)

📁 Project Structure

multi-agent-system/
├── src/
│   ├── app/
│   │   ├── page.tsx                          # Main UI — all 14 features integrated
│   │   ├── globals.css                       # Premium glassmorphism dark theme
│   │   └── api/
│   │       ├── orchestrate/route.ts          # SSE pipeline — all agents
│   │       ├── hitl/route.ts                 # HITL decision endpoint
│   │       ├── deliver/route.ts              # GitHub MCP push
│   │       ├── vision/route.ts               # 🆕 Vision-to-Code (F3)
│   │       ├── analyze/route.ts              # 🆕 Debt + Compliance on-demand (F2, F4)
│   │       └── workspace/route.ts            # File save/list
│   ├── lib/
│   │   ├── orchestrator.ts                   # 🎯 v3.1 pipeline controller
│   │   ├── hitl.ts                           # HITL promise resolver
│   │   ├── audit.ts                          # Audit log recorder
│   │   ├── memory.ts                         # Cross-session memory
│   │   ├── roi.ts                            # 🆕 ROI calculator (F7)
│   │   ├── agents/
│   │   │   ├── routerAgent.ts                # Intent classifier
│   │   │   ├── requirementsAnalyst.ts        # Agent 1
│   │   │   ├── taskPlanner.ts                # Agent 2
│   │   │   ├── developer.ts                  # Agent 3 + streaming + language skills
│   │   │   ├── codeReviewer.ts               # Agent 4 + lint
│   │   │   ├── securityReviewer.ts           # Agent 5 — OWASP
│   │   │   ├── testingAgent.ts               # Agent 6 — parallel
│   │   │   ├── deploymentAgent.ts            # Agent 7
│   │   │   ├── debtScanner.ts                # 🆕 Technical Debt Scanner (F2)
│   │   │   └── complianceAgent.ts            # 🆕 GDPR/HIPAA/PCI/SOC2 (F4)
│   │   ├── skills/
│   │   │   └── languages.ts                  # 🆕 7-language skill registry (F5)
│   │   ├── providers/index.ts                # Multi-provider LLM factory
│   │   ├── providers/runtime.ts              # Runtime provider/model resolver per agent
│   │   ├── validation/
│   │   │   ├── schemas.ts                    # Zod schemas (Zod v4 compatible)
│   │   │   └── handoff.ts                    # validateHandoff() utility
│   │   ├── prompts/                          # System prompts for all agents
│   │   ├── tools/                            # searchWeb, lintCode, readFile
│   │   ├── rag/                              # TF-IDF knowledge base
│   │   ├── flows/types.ts                    # Pipeline flow DSL
│   │   └── workspace/checkpoint.ts           # Save/load/resume
│   └── components/
│       ├── RequirementInput.tsx
│       ├── PipelineView.tsx                  # 8-agent pipeline view
│       ├── AgentCard.tsx                     # Status-aware agent card
│       ├── HITLModal.tsx                     # HITL approval modal
│       ├── SettingsPanel.tsx                 # Provider/model/delivery/memory settings
│       ├── EnterpriseTeamPanel.tsx           # RBAC + team management enterprise panel
│       ├── VisualEditor.tsx                  # Drag-and-drop visual DAG workflow builder
│       ├── VisionUploader.tsx                # 🆕 Vision-to-Code uploader (F3)
│       ├── EnhancedAnalyticsPanel.tsx        # 🆕 ROI + Debt + Compliance dashboard (F7)
│       ├── OutputPanel.tsx
│       ├── WorkspaceViewer.tsx               # File tree + ZIP export
│       ├── AnalyticsPanel.tsx
│       └── HistoryPanel.tsx

📡 API Reference

POST `/api/orchestrate`

Main pipeline. Returns Server-Sent Events stream.

// Request
{
  "requirement": "Build a FastAPI REST API for a todo app with JWT auth",
  "hitlEnabled": true,
  "resumeCheckpointId": "optional-for-resume",
  "workflowId": "standard-pipeline",
  "customModels": {
    "developer": { "provider": "openai", "model": "gpt-4o" }
  },
  "apiKeys": {
    "openai": "sk-..."
  },
  "ollamaUrl": "http://localhost:11434"
}

SSE Event Types:

Event	Description
`route_decision`	Router classified intent — mode, confidence, skipped agents
`stage_start` / `stage_complete` / `stage_error`	Agent lifecycle events
`stage_token`	Developer agent streaming token (Gap #9)
`hitl_requested`	Pipeline paused for human approval (Gap #1)
`pipeline_blocked`	Security blocked deployment (Gap #4)
`validation_error`	Zod schema validation warning (Gap #3)
`parallel_group_start/complete`	Parallel execution (Gap #6)
`iteration_info`	Language detection, memory-active context, debt/compliance scan, and dev/review loop info
`final_result`	Complete results + `debtReport` + `complianceReport` + `detectedLanguage` + `auditLog`

Runtime provider note: customModels, apiKeys, and ollamaUrl are passed through to agent runtime, enabling per-agent provider/model execution.

POST `/api/hitl`

Submit human decision to resume paused pipeline.

{ "requestId": "hitl_xxx", "decision": "approved", "feedback": "Looks great!" }

Decisions: approved | rejected | changes_requested

POST `/api/deliver`

Push generated files to GitHub.

{
  "target": "github",
  "config": { "token": "ghp_...", "owner": "you", "repoName": "my-app" },
  "files": [{ "path": "src/index.ts", "content": "..." }]
}

POST `/api/vision` (Feature 3)

Convert design image to code. Accepts multipart/form-data.

image         File      — PNG/JPG/WEBP (required)
framework     string    — react|nextjs|vue|svelte|angular|vanilla
language      string    — typescript|python|go|java|rust|ruby|php
styleLibrary  string    — tailwind|mui|shadcn|chakra|css-modules
context       string    — optional extra instructions

POST `/api/analyze` (Features 2 & 4)

On-demand technical debt + compliance analysis.

{
  "code": "your code here",
  "language": "python",
  "runDebt": true,
  "runCompliance": true,
  "complianceFrameworks": ["GDPR", "HIPAA", "OWASP_TOP10"],
  "industry": "healthcare"
}

🛡️ Resilience & Rate Limiting

Setting	Value
Inter-agent delay	1,500ms (prevents Groq rate limits)
Retry attempts per agent	3
Retry backoff	2s → 4s → 8s
Max review iterations	3
HITL timeout	10 minutes (then auto-approved)
Security block level	CRITICAL & HIGH only
Compliance block	Critical violations only
Debt scan	Non-blocking (pipeline continues always)
Vision model	Llama 4 Maverick (multimodal)
Router failure	Falls back to `FULL_PIPELINE`
Testing failure	Non-fatal — pipeline continues

🏆 Feature Comparison

Capability Matrix (Expanded)

Feature	This Project	LangFlow	Flowise	LangGraph	CrewAI	AutoGen	OpenDevin	n8n
🎨 Visual Drag-and-Drop Workflow Editor	✅ Enterprise UI	✅	✅	⚠️ Code-first	⚠️ Limited	⚠️ Limited	⚠️ Limited	✅
🔀 Flexible DAG / Branching Workflows	✅	✅	✅	✅	⚠️	⚠️	⚠️	✅
⏸️ Human-in-the-Loop Controls	✅	⚠️	✅	✅	⚠️	⚠️	⚠️	⚠️
👥 Team + RBAC Controls	✅	⚠️	⚠️	⚠️	✅	⚠️	⚠️	✅
🔑 Multi-Provider LLM Runtime	✅ Per-agent	✅	✅	✅	✅	✅	⚠️	✅
📊 Observability / Traces / Audit Export	✅	⚠️	✅	✅	⚠️	⚠️	⚠️	✅
🔗 Connectors + Webhook Triggers	✅	⚠️	✅	⚠️	✅	⚠️	⚠️	✅
🧠 Session Memory + Context	✅	⚠️	✅	✅	⚠️	✅	⚠️	⚠️
🛡️ Security Review Gate (OWASP)	✅	❌	❌	❌	❌	❌	❌	❌
🧪 Built-in Testing Agent Flow	✅	❌	❌	❌	⚠️	⚠️	✅	❌
💾 Checkpoint + Resume	✅	⚠️	⚠️	✅	⚠️	⚠️	⚠️	✅
🌊 Live Token Streaming in UI	✅	⚠️	✅	⚠️	❌	⚠️	⚠️	❌
🖼️ Vision-to-Code	✅	⚠️	⚠️	❌	❌	❌	⚠️	❌

Legend: ✅ Native support · ⚠️ Partial / plugin / code-heavy setup · ❌ Not a core built-in capability

Competitor Snapshot

Platform	Strength	Where this repo stands out
LangFlow	Great visual flow authoring	Adds stronger governance features: RBAC + HITL + security gate + audit export
Flowise	Strong low-code orchestration and integrations	Adds built-in agent roles, security reviewer, and end-to-end software pipeline focus
LangGraph	Robust graph + memory model	Adds richer no-code UX and productized pipeline UI for non-framework users
CrewAI	Team/task collaboration model	Adds visual DAG builder + enterprise telemetry dashboard in one product
AutoGen	Flexible agent conversations	Adds production UX, visual workflow builder, and governance controls
OpenDevin	Execution-centric engineering workflows	Adds orchestration governance, routing, and enterprise monitoring controls
n8n	Best-in-class automation/connectors	Adds specialized AI software-engineering pipeline with reviewer/security/testing agents

🛠️ Tech Stack

Technology	Purpose
Next.js 16	Full-stack React framework (App Router)
TypeScript 5	Type-safe development across all 60+ files
Vercel AI SDK v6	`generateText` + `streamText` — unified LLM interface
@ai-sdk/groq	Groq API provider (default, free)
Zod v4	Schema validation at every agent handoff
Groq Cloud	Ultra-fast LLM inference — free tier (30 RPM)
TF-IDF (custom)	In-memory RAG — zero extra dependencies
DuckDuckGo Instant API	Free web search for agentic tool calls
GitHub REST API	MCP code delivery (no Octokit required)
Llama 4 Maverick	Vision model for image-to-code conversion
Vanilla CSS	Custom glassmorphism design system + animations

🗺️ Roadmap

✅ Shipped

v1 — Foundation (6-agent pipeline, streaming UI, workspace)

- 6-agent automated pipeline with live SSE UI - Developer ↔ Reviewer feedback loop (3 iterations) - Workspace file manager with ZIP export - Premium glassmorphism dark UI - Analytics dashboard (tokens, latency, cost) - Pipeline history with localStorage persistence - Individual agent test API - Retry with exponential backoff

v2 — Intelligence Layer (routing, tools, RAG, flows, checkpoints)

- 🧭 Intelligent routing — 4-mode classifier - ⚙️ Agentic tools — searchWeb, lintCode, readFile - 📚 RAG knowledge base — TF-IDF over Next.js 15, React 19, TS 5 docs - 🔀 Flows DSL — typed pipeline definitions - 💾 Stateful checkpoints — resume from failure - 🧪 Testing Agent — auto-generates unit & integration tests

v3 — Enterprise Gaps (9 critical features)

- ⏸️ Human-in-the-Loop (HITL) modal approval gate - 🔑 Multi-Provider LLM (Groq + OpenAI + Anthropic + Ollama) - ✅ Structured Output Validation (Zod v4 schemas) - 🛡️ Security Reviewer Agent (OWASP Top 10) - 🐙 MCP GitHub Delivery - ⚡ Parallel Agent Execution - 📋 Full Audit Log Export (JSON) - 🧠 Agent Memory — cross-session learning - 🌊 Token Streaming — live code preview

Phase 1 — Enterprise Power (5 new features, current)

- 🌍 Multi-Language Pipeline (7 languages with auto-detection) - 🏗️ Technical Debt Scanner (A–F grade, 6 categories) - ⚖️ Compliance Agent (GDPR, HIPAA, PCI-DSS, SOC 2, DPDP, OWASP) - 🖼️ Vision-to-Code (image → component code with Llama 4 Maverick) - 💰 ROI Intelligence Dashboard (savings, ROI multiple, sprint impact)

🔭 Coming Next (Phase 2–3)

Feature	Phase	Impact
🏛️ Legacy Code Modernizer	Phase 2	COBOL/Java 8 → modern stack
🗺️ Knowledge Graph Builder	Phase 2	Semantic codebase map for onboarding
📋 Jira / Linear / Azure DevOps	Phase 2	AI tasks → PM tickets automatically
👥 Collaborative Multi-User	Phase 2	Real-time shared pipeline review
🧪 Synthetic Test Data	Phase 3	GDPR-safe realistic test datasets
🤖 Autonomous CI/CD Agent	Phase 3	Trigger → monitor → auto-fix build failures
🏪 Template Marketplace	Phase 3	Community starters, 40% faster runs
📱 Mobile-Responsive UI	Phase 3	Full mobile support

🤝 Contributing

Contributions are what make this the best repo on GitHub. Every PR is reviewed with care.

How to Contribute

# 1. Fork the repository
# 2. Create your feature branch
git checkout -b feature/amazing-feature

# 3. Make your changes
# 4. Run TypeScript check
npx tsc --noEmit --skipLibCheck

# 5. Commit
git commit -m 'feat: add amazing feature'

# 6. Push
git push origin feature/amazing-feature

# 7. Open a Pull Request

Good First Issues

Task	Difficulty	Where to Look
Add more RAG knowledge chunks	🟢 Easy	`src/lib/rag/knowledgeBase.ts`
Add a new language skill	🟢 Easy	`src/lib/skills/languages.ts`
Add a Jira delivery target	🟡 Medium	`src/app/api/deliver/route.ts`
Add email HITL notifications	🟡 Medium	`src/lib/hitl.ts`
Build the Legacy Modernizer Agent	🔴 Hard	`src/lib/agents/`
Real vector DB (pgvector) for RAG	🔴 Hard	`src/lib/rag/`

💡 Pro tip: Each new feature is self-contained in its own module. Adding a new agent takes ~150 lines. Adding a new delivery target takes ~50 lines.

📄 License

MIT License — see LICENSE for full details. Use it, fork it, ship it.

## ⭐ Star This Repository **If this project gave you value, taught you something, or just impressed you — drop a star.** **It helps more developers discover the best open-source AI orchestrator on GitHub.** [![Star History Chart](https://api.star-history.com/svg?repos=ParthivPandya/multi-agent-orchestrator&type=Date)](https://star-history.com/#ParthivPandya/multi-agent-orchestrator&Date) --- ### Built with ❤️ and extraordinary ambition by [Parthiv Pandya](https://github.com/ParthivPandya) *Next.js · TypeScript · Groq · Vercel AI SDK · Zod · RAG · HITL · MCP · Streaming · Multi-Language · Vision · Compliance · ROI* [![GitHub](https://img.shields.io/badge/GitHub-ParthivPandya-181717?style=for-the-badge&logo=github)](https://github.com/ParthivPandya) [![Email](https://img.shields.io/badge/Email-parthiv.pandya%40yahoo.co.in-6D4AFF?style=for-the-badge&logo=yahoo)](mailto:parthiv.pandya@yahoo.co.in) --- **[⬆ Back to Top](#-multi-agent-ai-orchestrator)** *v3.1 · March 2026 · 14 Enterprise Features · 8 Agents · 7 Languages · 100% Free*