Assistant operations
- Command workspace
- Action drafts
- Tool registry
- Safe mode
Intelligence And Governance
AI
AI layer for governed assistants, provider controls, tool registry, safe execution, diagnostics, evaluations, schedules, training review, and auditability.
已实现
Evidence-first module page
Primary users
Platform admin
AI operator
Department owner
Security reviewer
App-derived visual
AI operations workspace
Tools
Confirmed
Providers
Policy scoped
Training
Guarded
| 区域 | 状态 | 证据 |
|---|---|---|
| 操作 | Approval aware | Execution audit |
| Diagnostics | Runtime visible | Health snapshot |
| Training | Review gated | Poisoning control |
核心流程
What the module helps operators complete.
Each workflow is written as an operational path, not a marketing claim, so buyers can understand the day-to-day module fit.
Step 1
Configure providers, tools, policies, and safe-mode execution
Step 2
Run governed assistant actions with confirmation and audit evidence
Step 3
Review diagnostics, evaluations, schedules, and runtime health
Step 4
Control training-signal review and prompt/data boundaries
Feature checklist
Capabilities grouped the way operators look for them.
Provider governance
- Provider settings
- Prompt boundaries
- Data redaction
- Policy checks
Evaluation and training
- Evaluation runs
- Training suggestions
- Review workflow
- Diagnostics
Connected module map
How this module exchanges business context with the rest of ERPStack.
| Module | 关系 |
|---|---|
| Platform Settings | Provider policy, credentials, and runtime controls |
| 安全运营 | Training-signal poisoning and unsafe tool incidents |
| BI | Analytics explanation, KPI context, and report assistance |
| All modules | Governed tools for department-specific workflows |
Governance and reporting
Controls and management visibility expected from a serious ERP module.
治理
- Tool confirmation
- Prompt/data boundary enforcement
- Training-signal poisoning controls
- AI action audit trail
报表
- Provider health
- Agent runs
- Evaluation results
- Training review
AI coverage policy
What ERPStack can honestly claim for this module today.
AI claims are tied to the official Agent Tool Registry, evaluation evidence, safe-mode controls, and scheduled-agent monitor readiness. Roadmap modules only show planned coverage until module-native tools are implemented.
AI L1/L4 gap| Coverage claim | Current AI coverage: AI Core L1/L4 below target level through the official Agent Tool Registry. |
|---|---|
| Governance evidence | Governance evidence: 1 registered tools, 0 evaluation cases, tenant/RBAC/safe-mode registry controls, and AI log evidence. |
| Scheduled monitor | Scheduled monitor readiness: not in the current proactive-agent monitor wave. |
| L4/L5 boundary | Raise AI Core AI coverage from L1 to L4. |
Evidence-first standard
Module pages explain value, workflow, control, and proof in one place.
This page structure supports the current public module catalog with app-derived visuals, connected-module evidence, governance proof, and report context that can be compared against real product routes.
Feature clarity
Buyers can see concrete function groups instead of vague capability claims.
Operational fit
Workflows, users, and connected modules show where the module belongs.
Governance proof
RBAC, audit, SIEM, approvals, and reporting are surfaced as first-class concerns.
Review path
Public pages link back to app routes so reviewers can compare page claims to product surfaces.