glink-engine 0.5.0

Zero-dependency multi-agent workflow orchestration engine. YAML-defined pipelines, shared event bus, cross-agent routing.

Glink — The Bus That Makes Your Agents Work Together

I built three things:

1. GBase — an AI agent framework with a soul, that self-evolves and gets real work done.
2. Glink — the technology that lets GBase, OpenClaw, Hermes, Claude Code, and any AI agent truly collaborate on projects.
3. Opprime World — a metaverse built for AI, where agents work, live, meet, and communicate. — Gary Lin, 2026. Founder of the three.

---

Zero-dependency workflow orchestration. Define a pipeline in YAML. Glink routes every step to the right agent, logs every heartbeat onto a shared bus, and recovers when things go wrong.

---

Most workflow engines force you to choose: lightweight but dumb, or powerful but a database, a queue, and a team to maintain it.

Glink chose neither.

One Python process. One JSONL file. One YAML pipeline definition. That's all it takes to coordinate multiple AI agents across a complex, multi-step workflow — with automatic recovery, real-time dashboards, and per-step failure handling.

This is not an orchestrator you configure. This is the bus your agents ride on.

---

✨ What Makes Glink Different

🚌 Main Bus	Append-only JSONL timeline. Every step, every agent output, every failure — logged, timestamped, replayable. No database, no message queue
🔄 Automatic Recovery	PID-based watchdog + cron healthcheck. If the daemon stops, it restarts. If a step fails, it retries. You don't watch it — it watches itself
🧩 YAML Workflows	Define your pipeline in a single YAML file. Each step specifies an agent, a task prompt, input/output files, and retry policy
🧠 Smart Routing	Route each step to a specific agent by name or capability. Supports fallback agents if the primary is unavailable
📊 Commander Dashboard	Real-time web dashboard — see agent status, step progress, event timeline. SSE pushes live updates. No refresh required
🛡 Self-Recovery	self_restart() resumes from the last completed checkpoint step. Even crash recovery is intentional
🎯 Task Chaining	Each step receives the output of the previous step as context. Build on what came before — don't start from scratch
📬 Rich Status API	17 REST endpoints: /health, /status, /status/agents, /status/events, /intel/step, /intel/agents, /intel/timeline, /events/stream (SSE), and more
🧪 Zero Dependencies	Pure Python standard library. Nothing to install beyond Python 3.11+

---

🚀 Quick Start

# 1. Clone
git clone https://github.com/garyqlin/glink.git
cd glink

# 2. Run
python3 glink-daemon.py --serve

# 3. Submit a workflow
curl -X POST http://127.0.0.1:8426/run \
  -H "Content-Type: application/json" \
  -d '{"project": "hello-world"}'

# 4. Watch it live
curl http://127.0.0.1:8426/status

Commander Dashboard

Open http://127.0.0.1:8426/dashboard in your browser — see all agents, all steps, all events in real time.

One-Step Example

# workflows/hello-world.yaml
steps:
  - title: "Say Hello"
    agent: standard
    task: "Respond with 'Hello, World!' and count to 3."

Then run:

python3 glink-daemon.py --serve --project hello-world

---

🏗 Architecture

┌──────────────────────────────────────────────────────────┐
│                    Glink Daemon                           │
│                                                           │
│  ┌─────────────────────────────────────────────────────┐  │
│  │                   Main Bus                           │  │
│  │             Append-only JSONL Timeline               │  │
│  └──────┬──────────┬──────────┬──────────┬─────────────┘  │
│         │          │          │          │                │
│         ▼          ▼          ▼          ▼                │
│  ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐        │
│  │Agent A  │ │Agent B  │ │Agent C  │ │ Forge   │        │
│  │(port N) │ │(port M) │ │(port P) │ │(port Q) │        │
│  └─────────┘ └─────────┘ └─────────┘ └─────────┘        │
│                                                           │
│  ┌─────────────────────────────────────────────────────┐  │
│  │            Commander Dashboard (Web UI)              │  │
│  │     Real-time SSE push + per-agent status cards      │  │
│  └─────────────────────────────────────────────────────┘  │
│                                                           │
│  ┌─────────────────────────────────────────────────────┐  │
│  │          Auto-Recovery (PID + watchdog cron)         │  │
│  └─────────────────────────────────────────────────────┘  │
└──────────────────────────────────────────────────────────┘

Project Structure

glink/
├── glink-daemon.py        # Entry point
├── glink.py               # CLI runner
├── glink-config.yaml      # Server & agent configuration
├── daemon/
│   ├── __init__.py        # Public API
│   ├── core.py            # Workflow engine
│   ├── api.py             # REST API (17 endpoints + SSE)
│   ├── checks.py          # Healthcheck & self-recovery
│   ├── config.py          # Config loader
│   └── log.py             # Reporter integration
├── bus/
│   ├── main_bus.py        # Event bus (append-only JSONL)
│   └── agent_client.py    # Agent HTTP client
├── reporter/
│   └── reporter.py        # Multi-channel reporting (webhook/console)
├── dashboard/
│   ├── commander.html     # Real-time commander interface
│   └── index.html         # Step-by-step workflow viewer
├── workflows/             # YAML pipeline definitions
├── bin/                   # Launch & healthcheck scripts
└── docs/
    └── GLINK_WHITEPAPER.md

---

🛠 Configuration

glink-config.yaml

server:
  port: 8426
  retries: 2
  sse:
    heartbeat_interval: 60
    poll_interval: 2

project:
  default: hello-world

reporting:
  channels: []

Agent Ports

Agent	Default Port	Purpose
standard	8420	General-purpose agent
hammer	8431	Heavy engineering
ink	8432	Frontend & design
bumblebee	8434	Research & analysis
laser	8435	QA & documentation
forge	8436	Code art & review

Add your own agents by extending AGENT_PORTS in bus/agent_client.py — no limit.

---

📡 API Reference

Endpoint	Method	Description
/health	GET	Health check
/status	GET	Full workflow & agent status
/status/agents	GET	Per-agent online/offline
/status/events	GET	Recent event timeline
/intel/step	GET	Step-level intelligence (duration, agent, status)
/intel/agents	GET	Per-agent intelligence
/intel/timeline	GET	Timeline with timestamps
/events/stream	GET	SSE real-time push
/run	POST	Start/restart a workflow
/restart	POST	Restart daemon self
/reset	POST	Clear project bus events

---

❌ What Glink Is Not

• ❌ Not a task queue. Glink doesn't store pending tasks. It runs a pipeline and exits — every step is a live agent call.
• ❌ Not a database. The Main Bus is a log, not a query store. Use it for audit and recovery, not for ad-hoc analysis.
• ❌ Not a replacement for Airflow/Kubeflow. Glink is for agent orchestration, not data pipeline orchestration. If you're moving terabytes, look elsewhere.
• ❌ Not tied to any framework. Your agents can be anything — OpenClaw, Hermes, GBase, Claude Code, a shell script. If it speaks HTTP, Glink drives it.

✅ What Glink Is

The simplest way to make multiple AI agents collaborate on a complex task.

One daemon. One bus. One YAML file. Your agents talk to each other through the bus, and you watch it all happen on a real-time dashboard.

---

🔗 Related Projects

GBase	Recursive self-improvement agent framework. Give GBase agents a Glink pipeline and they orchestrate themselves.
Opprime World	The first metaverse where AI agents are natives. Glink workflows run across agents in Opprime World.

---

📄 License

MIT — free to use, modify, and distribute. No strings attached.

---

@garyqlin · 📦 Glink · 🧠 GBase