ZRT Vercel AI Systems Suite
Enterprise Agent Workflow Studio
DEMO LOCAL ONLY

Design enterprise-safe agent workflows before tools can act.

Build an auditable agent graph with tool scope, data classification, risk score, approval gates, launch checks, and exportable evidence.

Plain English

This is a planning board for AI agents. It shows which tools an AI would use, which steps need a human to approve them, and what evidence must be saved before anything risky happens.

DEMO LOCAL ONLYApproval gatesAudit exportEval fixtures

Readiness Console ReadinessA quick status view showing whether this workflow is safe to demo, needs human approval, or is blocked.

Readiness ReadinessThe launch state after checking selected tools, risk, and data sensitivity.APPROVAL REQUIRED
Risk score Risk scoreA simple 0 to 100 score estimating how careful the workflow needs to be.87/100

4 approval gate(s) inserted before risky tool use.

Workflow Definition WorkflowThe goal, risk level, data type, and tools the agent is allowed to use.

Tool Registry Tool registryThe list of pretend business systems the demo agent can use, such as lookup, search, ticket update, or refund preparation.

Generated Agent Graph Agent graphA map of the order the agents and tools would run in.

STEP 1 | agent
Intake Agent

READY

STEP 2 | agent
Scribe Agent

READY

STEP 3 | tool
CRM Lookup

APPROVAL REQUIRED

STEP 4 | tool
Policy Search

APPROVAL REQUIRED

STEP 5 | tool
Ticket Update

APPROVAL REQUIRED

STEP 6 | tool
Human Approval

APPROVAL REQUIRED

STEP 7 | evidence
Audit Report

READY

Cross-App Build Repair Workflow Patch suggestionThis step turns the Build Doctor diagnosis and Gateway trace into a guarded patch plan.

STEP 1 | Build Doctor
Paste build logs

Secret-redacted log input

STEP 2 | Build Doctor
Diagnose root cause

Root cause, confidence, evidence lines

STEP 3 | AI Gateway
Show trace

Trace timeline with status labels

STEP 4 | Enterprise Studio
Suggest patch

Patch checklist and audit gate

STEP 5 | Resume Auditor
Export report

Markdown report for reviewer handoff

Suggested patch handoff

Use the smallest code or configuration change that closes the diagnosed failure, keep write-capable actions behind approval, then rerun the verification commands before exporting the audit report.

Approval Gates + Launch Checks Approval gateA stop sign that requires a person to review the evidence before a risky tool runs.

CRM Lookup

Human approval required before execution

Evidence: request intent | retrieved context | proposed action | rollback path

Policy Search

Human approval required before execution

Evidence: request intent | retrieved context | proposed action | rollback path

Ticket Update

Human approval required before execution

Evidence: request intent | retrieved context | proposed action | rollback path

Human Approval

Human approval required before execution

Evidence: request intent | retrieved context | proposed action | rollback path

Confirm tool scopes before execution.

Capture retrieved evidence before write actions.

Require human approval for regulated or irreversible actions.

Export audit report after workflow simulation.

Audit Export Audit exportA copyable report explaining what the workflow would do and why it is safe or blocked.

# Enterprise Agent Workflow Audit Report

## Objective
Coordinate an enterprise customer escalation with evidence gathering, ticket drafting, and human approval before write actions.

## Mode
DEMO LOCAL ONLY

## Readiness
APPROVAL REQUIRED; risk score 87/100.

## Approval Gates
- CRM Lookup: Human approval required before execution
- Policy Search: Human approval required before execution
- Ticket Update: Human approval required before execution
- Human Approval: Human approval required before execution

## Agent Graph
- intake -> scribe
- scribe -> crm_lookup
- crm_lookup -> policy_search
- policy_search -> ticket_update
- ticket_update -> human_approval
- human_approval -> audit

## Safety Notes
- No real enterprise systems are called.
- Tool/API registry is mocked for deterministic evaluation.
- High-risk and write-capable tools are gated with explicit approval.