Agent Platform Overview
FactVerse AI Agent is a governed runtime for industrial agents. It lets a customer environment expose multiple configured agents through one platform model, while keeping tenant boundaries, tool access, action risk, review gates, and audit records visible.
Use this page when you need to explain how Agent workflows are organized before choosing a specific module, MCP endpoint, or workflow guide.
Platform flow
What the platform controls
| Control | What it means for customers |
|---|---|
| Agent roster | A tenant sees only the agents configured for that environment. |
| Agent purpose | Each agent should have a named operating task, owner, and entry point. |
| Tool binding | An agent can call only the tools assigned to it and visible in that environment. |
| Runtime discovery | Tool availability is checked at runtime, keeping customers aligned with the live environment. |
| Risk boundary | Read, analysis, draft write, and operational action paths are handled differently. |
| Review gate | Higher-risk actions can require confirmation or human approval before execution. |
| Audit trail | Decisions, blocked actions, approvals, and final results should be traceable. |
Agent types
FactVerse AI Agent documentation describes agent types by their business role.
| Agent type | Typical role | Normal output |
|---|---|---|
| Capability agent | Runs a model, calculation, data pipeline, or specialized analysis that other workflows use. | Score, forecast, classification, simulation result, or enriched dataset. |
| Assistant agent | Helps a user investigate a situation, summarize context, or prepare a reversible record. | Explanation, recommendation, draft work request, or review note. |
| Worker agent | Carries an approved operational task through a controlled process. | Approved task update, dispatch step, or action record. |
| Orchestrator agent | Coordinates multiple agents, tools, or operating steps inside a governed workflow. | Multi-step decision package, routed approval, or coordinated execution plan. |
These names describe the role of the agent. The actual tools available to an agent still depend on the customer environment, enabled modules, scopes, and risk controls.
Runtime model
An Agent workflow should be designed in this order:
- Define the operating task and user role.
- Confirm tenant, site, asset, equipment, scene, or work-record boundary.
- Check whether source data, documents, and digital twin context are ready.
- Discover the runtime tools available to the agent.
- Run read-only validation before adding compute or write actions.
- Put controlled actions behind confirmation, approval, and audit.
- Record feedback so later runs can improve.
This model keeps Agent work connected to real operating records, review decisions, and accepted outcomes.
Relationship to MCP
MCP is the governed tool integration layer for external AI clients and enterprise agent runtimes. Within the broader Agent Platform, it provides the tool-access layer.
| Layer | Role |
|---|---|
| Agent Platform | Defines agents, data context, tool binding, risk gates, review behavior, and audit expectations. |
| Agent Hub | Gives users a tenant-specific place to find and enter configured agents. |
| MCP endpoints | Let approved clients discover and call tools through scoped endpoints. |
| Module tools | Expose product capabilities such as predictive maintenance, CMMS, facility operations, data center operations, and Physical AI workflows. |
Use Getting Started for client setup and Access and Scope Planning for endpoint and key planning.
Review boundaries
Agent outputs need handling that matches their risk and operating impact.
| Output type | Expected handling |
|---|---|
| Read-only answer | Return evidence, source timestamps, and unresolved gaps. |
| Analysis or compute result | Return assumptions, input boundaries, and reviewer notes. |
| Draft record | Create or prepare the record only when the user has the required scope and workflow owner. |
| Operational action | Route through the customer's approval and audit process before affecting a real system. |
Use Workflow Run Record to capture evidence and review results.
Public documentation map
| Need | Use |
|---|---|
| See configured agents in the product | Agent Hub |
| Connect an external client | Getting Started |
| Plan endpoint and scopes | Access and Scope Planning |
| Check source readiness | Data Readiness |
| Build a workflow | Workflow Guides |
| Look up tool and scope reference | MCP Tool Reference and Scope Reference |