Koladr Documentation

Koladr is an AI refund agent for Shopify stores. It reads inbound refund requests, checks them against your rules, resolves clear cases automatically, and routes everything else to you for one-tap approval, with a full activity log of every decision.

What is Koladr?

Koladr is an AI refund agent. It sits between inbound customer requests and the systems where refunds actually happen (Shopify, your email, your payment processor) and checks every action against your rules before it executes.

Every action is logged, evaluated against your rules, and routed to you for approval before it reaches the outside world when it falls outside them.

The Problem

Refunds and returns are the biggest ticket source for most DTC stores. Each one costs real handling time, and slow responses turn into chargebacks. But letting software issue refunds unsupervised creates its own risk.

Without checks in place, you face:

• No visibility into what was refunded and why
• No way to stop a refund that breaks your rules before it executes
• No human in the loop for sensitive or unusual cases
• No record to fall back on when a customer disputes

Koladr solves all of these by routing every refund decision through your rules, and through you when it matters.

Who is it For?

Koladr is built for Shopify DTC stores where refunds have real consequences:

• Founders and owners who are still answering refund emails themselves
• Small support teams drowning in repetitive refund tickets
• Operators who want fast refunds without giving up the final say

Core Concepts

Understanding a few key concepts will help you get the most out of Koladr.

Your Refund Agent

The Agentis the part of Koladr that reads refund requests and acts on them. Koladr runs it for you; there is nothing to build, install, or host. You control how cautious it is and what it's allowed to do.

Runs

A Run represents how the agent handled one refund request, start to finish. All events, actions, and decisions for that request are logged under its run. Runs give you a complete timeline of what happened and why.

Actions

An Action is a specific operation the agent wants to perform in the real world: issuing a refund, sending a customer email. Actions are the critical control point. Each one is evaluated against your rules before it can execute.

Policies

Policies are the rules behind your refund settings that determine what happens when the agent wants to act. A policy can allow the action, block it, require approval from a human, or allow with alert. They let you encode your risk tolerance directly into the system.

Approvals

When a policy requires human review, the action enters an Approval queue. Designated approvers can review the full context (the agent, the run, the action details) and approve or reject it. The agent waits until a decision is made.

Incidents

An Incident is created when something goes wrong: a blocked action, a failed execution, an anomalous pattern, or an explicit escalation. Incidents link back to specific runs and actions, giving operators a clear starting point for investigation.

Connectors

Connectors are integrations between Koladr and external services like Gmail, Stripe, HubSpot, or Zendesk. When an action is approved, Koladr can execute it through the appropriate connector, so your agents never need direct access to third-party credentials.

Workspaces

A Workspaceis your team's environment within Koladr. It contains your agents, policies, connectors, and team members. Workspaces provide isolation. Different teams or projects can operate independently with their own configuration.

How it Fits Together

Here is the typical flow:

1. A customer's refund email reaches your intake address and the agent opens a run
2. The agent reads the message and matches it to a Shopify order, logging events along the way
3. When the agent needs to take a real-world action, such as issuing the refund or emailing the customer, it checks first
4. Koladr evaluates the action against your rules
5. If required, the request is routed to you for approval
6. Once approved (or auto-allowed), Koladr executes the action through the relevant connector
7. If anything goes wrong, an incident is created for investigation
8. The entire history is available in the run timeline for auditing