App Agents

Every app you install on IronFlock can bring its own AI agents. These agents extend the IronFlock AI assistant with specialized knowledge and device capabilities — automatically, without any configuration on your part.

How It Works

When you install an app that includes AI agents, those agents register with the IronFlock AI assistant. From that point on, when you ask a question that falls within the agent’s domain, the AI automatically delegates to the right agent.

You don’t select agents manually. You don’t configure routing rules. You just ask your question, and the AI figures out which agent — or which combination of agents — can answer it.

You: "What machines are on the local network?"
         │
         ▼
   IronFlock AI recognizes this as a network discovery question
         │
         ▼
   Delegates to the OPC UA app's network scanning agent
         │
         ▼
   Agent calls a tool on the edge device to scan the network
         │
         ▼
   Results flow back to you through the AI

What App Agents Can Do

App agents have access to tools — functions that run on your edge devices or in the cloud. These tools can read sensor data, send commands to machines, query databases, or trigger any operation the app supports.

Some examples of what app agents might provide:

The AI Grows with Your Project

This is the key idea: installing an app teaches the AI new skills.

A fresh project with no apps installed has access to the built-in tools — data exploration, charts, and dashboard generation. As you add apps, the AI becomes more capable:

1. You install an energy monitoring app → the AI can now answer questions about power consumption, load profiles, and energy costs.
2. You add a predictive maintenance app → the AI can now explain vibration trends, predict bearing failures, and suggest maintenance windows.
3. You add a quality inspection app → the AI can now link defects to specific batches, show reject rates, and compare quality across shifts.

Each app’s agents work independently but coexist within the same conversation. You can ask the AI a question that spans multiple apps — for example, “Did the power spike on Line 3 correlate with the quality drop we saw yesterday?” — and the AI can pull from energy and quality agents to answer.

Example Conversation

Here’s what a conversation might look like with multiple app agents available:

You: What’s the current status of production line 2?

IronFlock AI: Production Line 2 is currently running. Here’s a summary:

• OEE: 87.3% (above target of 85%)
• Current speed: 142 units/hour
• Active alarms: None
• Last quality check: 2 minutes ago — 0 defects in last batch

[Data pulled from the production monitoring app’s agent]

You: Have there been any vibration anomalies on the main drive motor this week?

IronFlock AI: Yes. The predictive maintenance agent reports elevated vibration on Motor M-201 starting Wednesday:

• Baseline: 0.8 mm/s RMS
• Current: 1.6 mm/s RMS (2× baseline)
• Trend: Gradually increasing since April 2

The model estimates a 72% probability of bearing degradation. Recommended action: schedule inspection within the next 5 operating days.

[Data pulled from the predictive maintenance app’s agent]

You: Can you scan the OT network for any new devices?

IronFlock AI: Scanning the local network via the OPC UA agent…

Found 2 new devices that aren’t currently registered:

1. Siemens S7-1500 at 192.168.1.45 — PLC with 340 available tags
2. Beckhoff CX5130 at 192.168.1.52 — EtherCAT controller with 128 available tags

Would you like me to add either of these to your project and start collecting data?

[Network scan performed by the OPC UA app’s agent]

For Developers

If you’re building an app and want to add AI agents, see the developer documentation:

• Defining Agents & Tools — How to define agents in ai-template.yml