> ## Documentation Index
> Fetch the complete documentation index at: https://koreai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Best practices for building and operating agents

These guides show how to design, build, test, and release agents on Agent Platform. Each guide starts with a decision, then shows the practice. Read the guide that matches the work in front of you.

The guides explain practice. The reference pages explain each setting. Each guide links to the reference pages and the [Artemis how-to guides](https://koreai.mintlify.site/agent-platform/abl/how-to/overview.md) for syntax and step-by-step tasks.

## Ways to work on a project

Agent Platform gives you three ways to work. Use one, or combine them based on the task, the risk, and how your team works.

| Surface | What it is | Use it for |
| - | - | - |
| **Studio** | The web console for visual authoring, configuration, and interactive testing. | Credentials, models, channels, visual edits, and interactive checks. |
| **Arch** | The AI architect inside Studio. It turns an SOP into a project plan and proposes changes for your review. | A first version from an SOP, or one small change. |
| **Coding assistant through Arch MCP** | An MCP-capable assistant, such as Codex or Claude Code, connected to platform tools. | Investigations and changes that span agents, tools, workflows, and tests. |

ABL (Agent Blueprint Language) is the source format for agent behavior. Whatever surface you use, agent developers own the outcome and approve consequential decisions. See [Work with a coding assistant](https://koreai.mintlify.site/agent-platform/best-practices/work-with-coding-assistant.md).

## The running example: Voyager Airlines

The guides use one fictional example so that journeys and names stay consistent. You don't need access to it. Apply the patterns to your own project.

A traveler calls because a flight is cancelled or delayed. The agent verifies the traveler, checks the flight status, and explains the options the policy allows. It asks before it changes the booking, and connects a person when needed.

The project has one orchestrator, **Concierge**. It verifies the traveler and routes to six specialists: **Booking**, **Manage Trip**, **Disruption**, **Baggage**, **Loyalty**, and **Travel Policy**. Disruption owns flight status and rebooking. The `rebook_flight` workflow checks eligibility, takes the seat, and moves the booking.

## Guides

| Guide | Read it when you need to |
| - | - |
| [Plan an agent journey](https://koreai.mintlify.site/agent-platform/best-practices/plan-agent-journey.md) | Turn a business process into scenarios, targets, and a test set before you build. |
| [Agent design guidelines](https://koreai.mintlify.site/agent-platform/best-practices/agent-design-guidelines.md) | Choose the agent structure and put each rule where it is enforced. |
| [Design context, memory, and handoffs](https://koreai.mintlify.site/agent-platform/best-practices/context-memory-handoffs.md) | Decide where each value lives and what passes between agents. |
| [Collaborate with Git](https://koreai.mintlify.site/agent-platform/best-practices/collaborate-with-git.md) | Work on one project with several developers. |
| [Work with a coding assistant](https://koreai.mintlify.site/agent-platform/best-practices/work-with-coding-assistant.md) | Connect an assistant through Arch MCP and keep it inside safe limits. |
| [Design voice conversations](https://koreai.mintlify.site/agent-platform/best-practices/design-voice-conversations.md) | Shape an agent for phone calls. |
| [Test voice agents](https://koreai.mintlify.site/agent-platform/best-practices/test-voice-agents.md) | Test the full call path and diagnose a failed call. |
| [Build multilingual agents](https://koreai.mintlify.site/agent-platform/best-practices/multilingual-agents.md) | Support several languages end to end. |
| [Diagnose and review a change](https://koreai.mintlify.site/agent-platform/best-practices/diagnose-and-review.md) | Find the cause of a failure, make one focused change, and review it. |
| [Evaluate agents](https://koreai.mintlify.site/agent-platform/best-practices/evaluate-agents.md) | Use eval sets, baselines, and calibrated judges as a release gate. |
| [Release agents to production](https://koreai.mintlify.site/agent-platform/best-practices/release-to-production.md) | Publish, deploy, watch, and roll back. |
| [Silent failures and key limits](https://koreai.mintlify.site/agent-platform/silent-failures-and-limits.md) | Look up settings that fail without an error. |

## Key terms

| Term | Meaning |
| - | - |
| **Working copy** | What you edit. **Chat with Agent** and working-copy eval runs use it. |
| **Version** | An immutable copy of an agent and a snapshot of its tools, made with **Publish Version**. |
| **Deployment** | A set of published versions running in one environment. |
| **Environment** | Development, Staging, or Production. Each has at most one active deployment. |
| **Deployment Binding** | The channel setting that decides whether a channel runs the working copy or a deployment. |
| **Eval set** | Personas, scenarios, and evaluators run against an execution target. |
| **Trace** | Every step of a turn. An LLM step's **Request** shows what the model received. |
| **Workflow tool** | A workflow called as one tool. Use it for rules and writes. |
| **Behavior profile** | Channel-specific instructions that an agent attaches with `USE BEHAVIOR_PROFILE`. |

## Before you build

Open every guide with the evidence in hand. When a rule matters, check it in the trace, the evaluation run, or the deployment that proves it. Keep the source version and the evidence with each change.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.