Skip to content

Quick Start Part 2: Agentic Flows in 5 Minutes

Editions: OSS, Cloud, Enterprise. Unless stated otherwise, everything on this page ships in OSS.

Build an event-driven workflow that uses your protected tools with an AI agent.

Prerequisites

Complete Part 1: Safety Layer first to set up your account, policies, and protected tools.

What You'll Accomplish

  • Configure an AI model for your flows
  • Create an automated payment processing flow
  • Trigger the flow and see the approval workflow in action
  • Learn how to trigger flows via webhooks

Step 1: Create Your AI Model (1 minute)

Flows need an AI model to execute tasks. Let's add one.

  1. Open Flows in the sidebar
  2. Click Create flow (the presets further down the page can wait)
  3. Under AI model, click Add AI model

  4. Fill in:

    • Name: GPT-5.4
    • Type: Inference / chat
    • Provider: OpenAI
    • Model Name / ID: gpt-5.4
    • API key: Your OpenAI API key

  5. Click Save

You can also add models under Models > Add model.

Add AI Model The Add model dialog. Model Name / ID appears after Fetch models from provider, or as a text field if the fetch fails.

Don't Have an OpenAI Key?

Get one at platform.openai.com/api-keys, or add a model from any other supported provider.

✓ Checkpoint: You now have an AI model configured!

Step 2: Create a Payment Approval Flow (2 minutes)

Let's create a flow that processes contract payments. If the amount exceeds your threshold, it will require approval.

Flow Scenario:

When triggered (manually or via webhook), the flow will:

  1. Receive payment details (recipient, amount, contract_id)
  2. Use the prelooped pay tool to send payment
  3. If amount exceeds the condition, require approval first
  4. Report the result

Create the Flow:

  1. You're already on the Create Flow page. Fill in:

  2. Flow name: Contract Payment Processor

  3. Description: Process contract payments with approval for large amounts
  4. Trigger type: Webhook (the webhook URL is generated after creation)
  5. AI model: the model you just added

  6. Prompt template:

    You are a payment processor. Process the payment with these details:

Recipient: {{trigger_event.payload.recipient}}
Amount: ${{trigger_event.payload.amount}}
Contract ID: {{trigger_event.payload.contract_id}}

Use the pay tool to send the payment. The tool is configured with
an approval workflow - small amounts are auto-approved, larger amounts
require human approval.

After payment completes, report the status. Do not retry if declined.

Under MCP Server Tools, make sure pay is checked. You can uncheck other tools.

  1. Click Create flow at the bottom

Create Flow The tools section of the new flow, with pay checked under MCP Server Tools

✓ Checkpoint: Your flow is created!

Step 3: Test Your Flow (2 minutes)

Now let's trigger your flow and see the approval workflow in action!

Trigger the Flow:

  1. You should now be on the flow details page (flows you create start enabled)
  2. Click Run now
  3. The dialog Values for the trigger event asks for the template variables. The run is a real one and spends like any other

  4. Fill in:

    • trigger_event.payload.recipient: contractor@example.com
    • trigger_event.payload.amount: 150 (above the $100 auto-allow limit from Part 1)
    • trigger_event.payload.contract_id: CONTRACT-2026-001

  5. Click Run now

Test Flow Running the flow with sample payment data, which triggers the approval workflow

Watch the Execution:

You'll be redirected to the execution page where you can see the AI agent working:

  1. The agent will read your prompt
  2. It will attempt to call the pay tool with amount=$150
  3. Because of the rules you set on pay in Part 1, the Support workflow gets an approval request

Flow execution started Flow execution in progress: the AI agent is processing the request

Flow waiting for approval The run is waiting for the Support workflow to approve the payment

Approve the Payment:

The approver receives an email with a subject such as "Tool Approval Required: pay" (when the agent names itself or summarizes the request, the subject says so instead).

  1. Click Approve in the email, or:
  2. Approve from Audit > Approvals in the console
  3. Approve from the mobile app
  4. Go back to the flow execution page
  5. Watch the agent complete the payment after your approval!

✓ Checkpoint: You've successfully run an automated flow with human approval!

Triggering via Webhook

Your flow has a unique webhook URL. You can trigger it from external services!

  1. Go back to your flow details page
  2. Find the Webhook URL section
  3. Copy the URL (looks like: https://preloop.ai/api/v1/webhooks/flows/{flow_id}/{webhook_secret})

Test with curl:

curl -X POST 'YOUR_WEBHOOK_URL' \
  -H 'Content-Type: application/json' \
  -d '{
    "recipient": "vendor@example.com",
    "amount": 50,
    "contract_id": "CONTRACT-2026-002"
  }'

This payment runs without approval because it is at or below $100.

Try with a larger amount:

curl -X POST 'YOUR_WEBHOOK_URL' \
  -H 'Content-Type: application/json' \
  -d '{
    "recipient": "vendor@example.com",
    "amount": 500,
    "contract_id": "CONTRACT-2026-003"
  }'

This one needs the CFO workflow's approval. Anything above $1,000 is denied by the last rule.

Success! You've Built Safe Automation

You've just:

  • Set up the Safety Layer: protected a risky tool with policies (Part 1)
  • Built an automated workflow: created a flow with an AI agent
  • Tested end-to-end: triggered the flow, got approval request, watched it execute
  • Learned webhook triggers: you can now trigger flows from external services

What's Next?

Learn the flow building blocks:

Go deeper on how flows are created, triggered, and executed:

  • Creating Flows: build reusable AI-driven workflows
  • Flow Triggers: understand webhooks and other trigger types
  • Flow Execution: inspect status, history, and outcomes

Creating Flows → | Flow Triggers → | Flow Execution →

Advanced Approval Workflows:

Make approval smarter with conditional logic:

  • Approve only if environment == "production"
  • Different approvers for different amounts
  • Team-based approval with quorum (require 2 of 5 approvers), Cloud and Enterprise
  • Escalation chains, Cloud and Enterprise

Conditional Approval (CEL) → | Team-Based Approvals →

Connect Your Own MCP Servers:

Put tools from any MCP server behind Preloop:

  • Deployment tools
  • Database operations
  • Cloud provider APIs
  • Your custom MCP servers

External MCP Tools →

Learn More

Core Concepts:

Flows:

Advanced:

Need Help?