Skip to main content

​
What You’ll Build

In this tutorial, you’ll create a Customer Support Agent that can:
  • Answer common questions about your product
  • Route complex issues to the right team
  • Maintain context across conversations
Time Required: 5 minutes Prerequisites: Google account (for login)

​
Step 1: Sign In

1

Go to Junis

Navigate to https://junis.ai and click “Sign in with Google”
2

Authorize Access

Grant Junis permission to access your Google account. We only use this for authentication.
Junis never accesses your Gmail, Drive, or other Google services without explicit permission.
3

Complete Onboarding

You’ll be prompted to:
  • Create or join an AI Team
  • Set your display name
  • Choose your role
For this tutorial, select “Create New AI Team” and name it anything you like (e.g., “Acme Support”).

​
Step 2: Create Your First Agent

1

Navigate to Orchestration

After logging in, you’ll see the main dashboard. Click “Team Management” dropdown in the Navigation Bar, then select “Agent Orchestration”.
2

View Agent Configuration

In the Orchestration tab, you’ll see a pre-configured Orchestrator agent (don’t modify this yet).
The Orchestrator is the “brain” that routes user messages to the right agents. Every AI Team has one.
3

Create New Agent

Click the “New Agent” button.

​
Step 3: Configure Your Agent

Fill in the agent creation form: 
Name: Customer Support Agent
Description: Helps customers with common questions and routes complex issues
Agent Type: LLM Agent
Status: Active ✅
  • Agent Type: LLM Agent uses a language model to generate responses
  • Model: Claude Sonnet 4.5 is fast and cost-effective for most tasks
  • Temperature: 0.7 balances creativity and consistency (0 = deterministic, 1 = creative)
  • Max Output Tokens: Limits response length to control costs
  • Instructions: The “system prompt” that defines the agent’s behavior
Click “Save” to create your agent.

​
Step 4: Connect Agent to Orchestrator

1

Edit the Orchestrator

Go back to the Agents list and click “Edit” on the Orchestrator agent.
2

Add Your Agent as a Sub-Agent

Scroll to the “Sub-Agents” section and click ”+ Add Sub-Agent”.Select “Customer Support Agent” from the dropdown and set:
  • Order: 1 (execution priority)
  • Description: “Handles customer support questions”
3

Update Orchestrator Instructions

In the Instructions field, add:
You are the Orchestrator for Acme Corp.

Route user messages to the appropriate agent:
- Customer support questions → Customer Support Agent
- General conversation → Respond directly

Always be helpful and professional.
Click “Save”.

​
Step 5: Test Your Agent

1

Start a New Chat

Click “Chat” in the sidebar, then click ”+ New Chat”.
2

Send a Test Message

Try these example messages:
How much does Acme Widgets cost?
3

Observe Agent Routing

Watch the Agent Pipeline Viewer at the bottom of the chat:
  1. Orchestrator receives your message
  2. Routes to Customer Support Agent
  3. Response streams back in real-time
The agent should correctly answer pricing and trial questions based on the instructions you provided.

​
Congratulations! 🎉

You’ve created your first AI agent! Here’s what you learned:
✅ How to create and configure an LLM Agent ✅ How to write effective agent instructions ✅ How to connect agents to the Orchestrator ✅ How to test agents in the chat interface

​
Next Steps

Add Tools to Your Agent

Connect to APIs, databases, and external services

Create a Sequential Workflow

Chain multiple agents together

Connect MCP Platforms

Integrate GitHub, Firecrawl, and custom MCP servers

​
Tips for Better Agents

  • Be specific about the agent’s role and responsibilities
  • Provide examples of expected inputs and outputs
  • Include edge cases (e.g., “If you cannot help, say X”)
  • Use bullet points for readability
  • Begin with basic functionality
  • Test thoroughly with real user messages
  • Add complexity (tools, RAG, sub-agents) gradually
  • Monitor agent performance and refine instructions
  • Claude Haiku: Fast, cheap, good for simple tasks ($0.25/1M tokens)
  • Claude Sonnet: Balanced speed and quality ($3/1M tokens)
  • Claude Opus: Most capable, best reasoning ($15/1M tokens)
  • GPT-4o: Best for structured outputs and function calling
Try these scenarios:
  • Ambiguous questions
  • Out-of-scope requests
  • Multi-turn conversations
  • Requests in different languages
  • Adversarial inputs (jailbreaks, prompt injections)

​
Common Issues

Possible causes:
  • Agent is inactive (check Status toggle)
  • Orchestrator didn’t route to your agent (check Agent Pipeline Viewer)
  • API key missing for the selected model
Solution:
  • Ensure agent Status is Active (green toggle)
  • Check Orchestrator instructions include your agent
  • Verify API keys in Settings > Organization
Possible causes:
  • Instructions are too vague
  • Temperature is too high (> 0.9)
  • Agent doesn’t have enough context
Solution:
  • Rewrite instructions with specific examples
  • Lower temperature to 0.5-0.7 for consistency
  • Add relevant information to the system prompt
Possible causes:
  • Using a slow model (Opus, GPT-4)
  • Agent is calling multiple tools/sub-agents
  • Network latency
Solution:
  • Switch to Haiku or Sonnet for faster responses
  • Optimize tool calls (reduce number of API requests)
  • Use streaming for better perceived performance

​
Need Help?

Email Support

Contact our team directly

API Reference

Integrate with OpenAI-compatible API
Pro Tip: Explore Public Agents in Agent Orchestration > New Agent > Public Agent to discover pre-built templates you can clone and customize for your AI Team.