Skip to main content

Documentation Index

Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Build your first AI agent in under 5 minutes — no installation required.

​
Prerequisites

All you need is:

​
Step 1: Sign up and log in

Head to ablplatform.com and create your account. Once you verify your email, you’ll land in Studio — the browser-based IDE where you build, test, and manage your agents.

​
Step 2: Create your first project

Click New Project from the Studio dashboard. Give it a name like “My First Agents” and select a workspace. Projects organize your agents, supervisors, tools, and knowledge sources in one place.

​
Step 3: Create your first agent

Inside your project, click New Agent. Open the ABL editor and paste this definition: 
AGENT: Support_Assistant

EXECUTION:
  model: claude-sonnet-4-5-20250929

GOAL: |
  Help customers with product questions. Be concise
  and friendly. If you do not know the answer, say so.

PERSONA: |
  Helpful product support assistant. Answers questions
  clearly and concisely.

LIMITATIONS:
  - "Cannot process payments or refunds"
  - "Cannot access customer account information"

TOOLS:
  search_knowledge(query: string) -> {results: object[], totalCount: number}
    description: "Search the product knowledge base"

INSTRUCTIONS: |
  1. Understand the customer's question
  2. Search the knowledge base for relevant information
  3. Provide a clear, sourced answer
  4. If unsure, offer to connect with a human agent
This 20-line definition creates an agent that:
  • Uses an LLM to understand customer questions
  • Searches a knowledge base for answers
  • Responds with sourced information
  • Has clear boundaries on what it can and cannot do
Click Save to validate your ABL definition. Studio parses it in real time and flags any syntax issues inline.

​
Step 4: Test your agent

Open the Test panel on the right side of Studio and send a message: 
What is your return policy?
Your agent processes the message, searches for relevant knowledge, and responds. You can see the full execution trace — including LLM calls, tool invocations, and reasoning steps — in the trace viewer below the chat. Try a few more messages to see the agent handle different questions, maintain context across turns, and respect its defined limitations.

​
Step 5: Add a supervisor

Ready for multi-agent orchestration? Create a new Supervisor in your project and paste this definition: 
SUPERVISOR: Product_Supervisor

EXECUTION:
  model: claude-sonnet-4-5-20250929

GOAL: |
  Route customer queries to the right specialist agent.

HANDOFF:
  - TO: Support_Assistant
    WHEN: user asks about products, features, or general help
    PASS: query

  - TO: Billing_Agent
    WHEN: user asks about invoices, payments, or subscriptions
    PASS: query
The supervisor analyzes each incoming message and routes it to the appropriate agent, passing conversation context along the way. Test it the same way — open the Test panel and send messages that should route to different agents.

​
What you built

In just a few minutes, without installing anything, you created:
  • An agent that understands natural language, retrieves knowledge, and enforces boundaries
  • A supervisor that routes messages to the right specialist
  • Observable traces for every execution step, visible right in Studio

​
Next steps

Keep building:
Tip: Check out the Template Gallery in Studio for ready-made agent definitions across industries — airlines, retail, banking, telecom, travel, and more.