EngineX

EngineX

Open-source goal-driven agent runtime — workflow graphs, event-loop nodes, human-in-the-loop, checkpoints, credentials, MCP tools, CLI, and web dashboard.

Introduction

EngineX is a framework for building and running goal-driven agent workflows in production. Define multi-step graphs (LLM steps, tools, validation loops, human approval gates), run them headless or via a web dashboard, and deploy in a client VPC with local credential storage.

Typical uses: document review with HITL, support triage, invoice validation, log monitoring, reconciliation, calendar/CRM automation, and custom client workflows built on the same runtime.

Key Features

• Workflow graphs — nodes, edges, fan-out/fan-in, subagents, supervisors
• Event-loop nodes — LLM + tool calling with automatic output validation and retry
• Human-in-the-loop (HITL) — pause nodes, dashboard review, checkpoint resume
• Checkpoints & session state — fault tolerance; resume from paused steps
• Credential vault — encrypted local store (~/.engine/credentials/) + OAuth Connect
• MCP tools — Slack, browser automation, terminal, files, and custom tools.py
• CLI & TUI — ./engine run, validate, setup-credentials, interactive terminal UI
• Web dashboard — sessions, graph view, HITL panels, ops console, credentials OAuth

Quick Start

git clone https://github.com/EngineXV/engineX.git
cd engineX
./quickstart.sh    # uv sync + validate example agent
./engine run examples/templates/agreement_analysis --tui

Or manually:

uv sync
./engine validate examples/templates/agreement_analysis
./engine run examples/templates/agreement_analysis --tui

Web Dashboard

cd core/frontend && npm install && npm run build
cd ../..
./engine serve

Open http://127.0.0.1:8787 for sessions, HITL review, checkpoints, credentials, and ops metrics.

Workflow Templates

15 reference templates ship in examples/templates/ — 8 workflow agents + 7 department supervisors. These are demos and accelerators; production client workflows typically live in exports/ (gitignored) or a private repo.

See the full list: [examples/templates/README.md](examples/templates/README.md)

| Category | Templates | |----------|-----------| | HITL review | agreementanalysis, invoicereview, supporttriage, medicalbillingauditor | | Automation | logmonitor (daemon), hourlytracking, meetingscheduler, deepresearch | | Supervisors | technology, legal, finance, marketing, operations, growth, branddesign | | Composite | supervisedagreementanalysis |

./engine validate examples/templates/log_monitor
./engine run examples/templates/medical_billing_auditor --tui
./engine run examples/templates/log_monitor --daemon --require-live

Integration patterns (Slack, HubSpot, Google Calendar): [examples/templates/integrations/README.md](examples/templates/integrations/README.md)

Client Deployments

EngineX OSS is designed for self-hosted installs in a client VPC:

• One install = one isolated environment; data under ~/.engine/
• Headless (./engine run, cron, systemd) and/or dashboard (./engine serve)
• Custom agents call client backend APIs via tools.py and accept JSON via --input / --input-file
• Client-specific agents are not committed to this public repo

See open INFO tickets for deployment and architecture notes: #13, #14, #15. Future multi-tenant SaaS control plane: #31 Engine Cloud design spec.

LLM Setup

EngineX requires an LLM for agent nodes.

Option A — Cloud API Key (Anthropic, OpenAI, etc.)

cp .env.example .env
# Edit .env — set ANTHROPIC_API_KEY or OPENAI_API_KEY
./engine run examples/templates/agreement_analysis --tui
./engine setup-credentials examples/templates/agreement_analysis

Option B — Local LLM (Ollama, no API key)

ollama serve
ollama pull qwen2.5:7b

Set default in ~/.engine/configuration.json:

{
  "llm": {
    "provider": "ollama",
    "model": "qwen2.5:7b"
  }
}

Or per run: ./engine run --model ollama/qwen2.5:7b --tui

Project Layout

engineX/
├── engine                 # CLI wrapper
├── core/engine/           # Runtime (import: engine)
├── core/frontend/         # Web dashboard (React)
├── tools/                 # MCP tool servers (import: engine_tools)
├── examples/templates/    # 15 reference workflow templates
└── exports/               # Local client agents (gitignored)

Commands

./engine run  [--tui] [--input '{...}'] [--input-file file.json]
./engine validate 
./engine info 
./engine serve [--host 0.0.0.0] [--port 8787]
./engine setup-credentials 
./engine tui

Configuration, sessions, and credentials: ~/.engine/

Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md).

License

[MIT License](LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: EngineXV
• Source: EngineXV/engineX
• License: MIT

Install and usage instructions live in the source repository linked above.