Agent Lifecycle and Configuration

FactVerse AI Agents should be managed as operating capabilities. Each configured agent needs a business purpose, owner, user entry point, data context, tool boundary, review path, and run record.

Use this page when preparing a new agent for a tenant, changing an existing agent, or deciding whether an agent is ready for regular use.

Lifecycle flow

Configuration package

Prepare a short configuration package before an agent appears in Agent Hub.

Field	Why it matters	Review question
Agent name	Gives users a recognizable entry point.	Would the target user understand the name from the card alone?
Purpose	Ties the agent to an operating task.	What decision, record, or workflow does the agent support?
Owner	Names the person or team accountable for operation and change.	Who accepts behavior changes and review outcomes?
Status	Shows whether the agent is available, limited, planned, or retired.	Should users enter this agent today?
Entry route	Sends users to the right module, dashboard, assistant, or workflow.	Does the route match the work users expect to do?
Data dependencies	Lists source systems, documents, scenes, and records the agent needs.	Are freshness and quality expectations known?
Tool boundary	Defines the tools the agent may discover and call.	Are visible tools aligned with the agent purpose and scope?
Review boundary	Defines read-only, analysis, draft, confirmation, and approval behavior.	Which outputs can be accepted directly, and which require review?
Run record	Preserves evidence, decisions, actions, and feedback.	Can a later reviewer reconstruct what happened?

Plan the agent

Start by defining the operating task, then choose the model, tools, and review path that fit the task. A useful agent plan should state:

• the user role or team that will use the agent;
• the tenant, site, asset group, equipment, scene, or work-record boundary;
• the normal question, task, or decision package the agent handles;
• the data sources and product modules involved;
• the expected output type;
• the reviewer or approval owner for higher-impact results.

Link the plan to Data Readiness when source quality affects the output.

Configure the Hub entry

Agent Hub should show a concise entry for each configured agent.

Hub item	Configuration guidance
Name	Use a business capability or operating role.
Summary	Explain the task in one sentence.
Status	Use available, limited, planned, or retired states consistently.
Entry point	Link to the module page, workflow, dashboard, assistant, or approval route.
Data health	Surface delayed, missing, or under-review data when it changes the answer.
Review note	Tell users whether the agent returns evidence, drafts, or actions needing approval.

Use Agent Hub for the full entry model.

Bind tools and access

Tool access should follow the agent purpose.

1. Confirm the agent's operating task.
2. Select the minimum tool set needed for that task.
3. Match MCP scopes and tenant permissions to the selected tools.
4. Confirm whether each tool is read-only, analysis-oriented, draft-generating, or action-capable.
5. Test runtime discovery with the same access package the target users will use.
6. Record missing tools, blocked tools, or scope gaps before launch.

Use Access and Scope Planning and MCP Tool Reference when preparing the access package.

Validate before launch

Before regular use, run a small validation set that covers normal, weak-data, and blocked-action cases.

Validation area	What to check
Identity and tenant boundary	The agent sees only the expected tenant, site, asset, and permission context.
Data readiness	Required records, signals, documents, and scenes are present or clearly marked as gaps.
Runtime discovery	The agent sees the intended tools in the live environment.
Evidence quality	Answers include source references, timestamps, assumptions, and unresolved gaps.
Review behavior	Drafts, confirmations, approvals, and blocked actions follow the configured review path.
Run record	Tool calls, outputs, reviewer decisions, and feedback are captured.

Use Workflow Run Record to structure the validation evidence.

Operate and improve

During regular operation, keep a lightweight review rhythm.

Operating signal	Action
Repeated missing data	Fix source mapping, freshness, or documentation before expanding usage.
Frequent blocked tools	Review scope, module enablement, and agent-tool binding.
Low-confidence answers	Add source evidence, examples, or reviewer feedback to the workflow.
Review bottlenecks	Clarify who approves which output type.
Accepted corrections	Feed approved corrections back into the workflow design and runbook.
Usage decline	Check whether the entry point, name, or task boundary matches real work.

Change or retire an agent

Treat changes as controlled updates. Before changing tool access, review boundary, or entry route, record the reason, expected user impact, and validation plan.

Retire an agent when its source systems, owner, entry point, or operating purpose no longer match current work. Preserve historical run records and remove the Hub entry from active use.

Launch checklist

Check	Ready when
Purpose	The agent has a named operating task and user role.
Owner	A product or operations owner accepts changes and review outcomes.
Data	Source systems, documents, scenes, and records have freshness expectations.
Access	MCP endpoint, scopes, tenant permissions, and tool discovery have been tested.
Review	Output types and approval paths are clear.
Evidence	The workflow can produce source references and run records.
Handover	Users know where to enter the agent and where to report issues.

Related pages

Need	Use
Understand the platform model	Agent Platform Overview
Design the user entry point	Agent Hub
Plan source readiness	Data Readiness
Plan access and scopes	Access and Scope Planning
Capture validation evidence	Workflow Run Record