About IndustrialClaw

What is IndustrialClaw?

IndustrialClaw is an AI Agent platform for industrial edge environments. It helps teams build, deploy, manage, and scale AI Agents so they work with models, tools, and existing industrial systems instead of remaining isolated prototypes.

A device runs a Main Agent plus up to ten Sub-agents. Each Sub-agent is installed from a package, receives work as Tasks, and produces files that the platform verifies.

Does IndustrialClaw replace MES, SCADA, PLCs, or machine-control systems?

No. IndustrialClaw does not replace MES, SCADA, PLCs, AOI and vision systems, or other OT and IT systems. It provides an Agent layer that uses their data and capabilities through defined tools, APIs, and connectors.

It is also not a functional safety system or a deterministic machine-control system. Safety-critical control logic stays in the systems designed for it.

Can Agents take actions automatically?

Yes, within the Skills installed for that Agent and the commands a Task declares. A Task can call only Skills installed for its own Agent, so the package is the permission boundary.

Different use cases justify different levels of autonomy:

Read / analyse → recommend → human approval → execute

For actions that affect equipment or production, keep a human in the loop appropriate to the risk.

What is a Sub-agent?

A Sub-agent is an independent Agent with its own workspace, Skills, model, and Tasks—for example Evidence Analysis, SOP Retrieval, Planning, and Result Coordination as four separate Agents.

In 1.0.0 a Sub-agent does not call another Sub-agent. Roles are chained by running Tasks in sequence and passing files: one Agent's Outputs become the next Agent's Material upload, so every handoff has its own Log, Session, and Audit record. See Sub-agents.

What is a good first use case?

Something focused, easy to validate, and operationally useful: AOI anomaly analysis, SOP retrieval, equipment alarm summaries, maintenance recommendations, quality-review preparation, or an operational knowledge assistant.

Start with one Agent, validate it against a normal case, an uncertain case, and a failure case, and expand from there. Quick Start runs the whole loop once in about 30 minutes.