Koladr Documentation

Koladr is the control plane for AI agents in production. Route every high-stakes action through policy checks, human approvals, and a full audit trail — so your agents can do real work without losing your trust.

What is Koladr?

Koladr is an AI Agent Control Room. It sits between your AI agents and the external systems they interact with — email providers, payment processors, CRMs, support tools — and governs every action before it executes.

Think of it as the operations layer your agents need before you can trust them in production. Every action is logged, evaluated against your policies, and routed through the appropriate approval workflow before it reaches the outside world.

The Problem

AI agents are increasingly capable of taking real actions: sending emails, processing refunds, updating CRM records, and managing tickets. But capability without control creates risk.

Without a governance layer, you face:

• No visibility into what agents are doing in real time
• No way to block high-risk actions before they execute
• No human-in-the-loop for sensitive decisions
• No audit trail for compliance and debugging
• No incident response when things go wrong

Koladr solves all of these by providing a single control plane that every agent action must pass through.

Who is it For?

Koladr is built for teams that are deploying AI agents in production environments where mistakes have real consequences:

• Engineering teams building AI-powered workflows that interact with external APIs
• Operations teams responsible for monitoring and approving agent actions
• Compliance teams that need audit trails and policy enforcement
• Customer success teams managing AI support agents that handle real customer requests

Core Concepts

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

Agents

An Agent is any AI system you register with Koladr. It could be a customer support bot, a sales assistant, a data processing pipeline, or any autonomous workflow. Each agent gets a unique identity, API key, and its own run history.

Runs

A Run represents a single execution session of an agent. When your agent starts processing a task — like handling a support ticket — it opens a run. All events, actions, and decisions within that task are logged under that run. Runs give you a complete timeline of what happened and why.

Actions

An Action is a specific operation your agent wants to perform in the real world: sending an email, issuing a refund, updating a record. Actions are the critical control point — each one is evaluated against your policies before it can execute.

Policies

Policies are rules you define that determine what happens when an agent requests an action. A policy can allow the action, block it, require approval from a human, or allow with alert. Policies let you encode your organization's 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. Your agent starts a run in Koladr
2. The agent processes its task and logs events along the way
3. When the agent needs to take a real-world action, it requests it through Koladr
4. Koladr evaluates the action against your policies
5. If required, the action is routed to a human 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