System Agents

Router Agent

The Router Agent is Junet's "traffic controller" - it analyzes every user question and decides which specialist agent should handle it.

What It Does

The Router Agent is responsible for:

• Analyzing user questions to understand intent and context
• Reviewing all available agents and their capabilities
• Selecting the best agent for each specific query
• Routing the question to the chosen specialist agent

How It Works

Step-by-Step Process:

1. User asks a question

   • Example: "What's our CI/CD deployment process?"

2. Router Agent analyzes the question

   • Identifies keywords: "CI/CD", "deployment", "process"
   • Understands intent: User wants technical documentation about deployment

3. Router Agent reviews available agents

   • Reads each agent's name and description
   • Evaluates which agent is the best match

4. Router Agent makes a decision

   • "This is about deployment and DevOps..."
   • "The 'DevOps Documentation Agent' is the expert for this!"

5. Question is routed to the specialist agent

   • DevOps Documentation Agent receives the query
   • Agent uses its connections to find the answer
   • User receives relevant, accurate information

Real-World Example

Scenario: Your company has multiple agents set up:

• DevOps Documentation Agent - Expert in CI/CD from Confluence
• HR Policy Agent - Expert in company policies from HR Wiki
• Support Ticket Agent - Expert in issue tracking from Jira
• Sales Agent - Expert in customer data from CRM

User asks: "What's the process for deploying to production?"

Router Agent thinks:

• "This question mentions 'deploying' and 'production'"
• "The DevOps Documentation Agent knows about deployment processes"
• "Not relevant for HR, Support, or Sales agents"
• Decision: Route to DevOps Documentation Agent ✅

Why You Can't Delete It

What happens without the Router Agent:

• ❌ System can't determine which agent to use
• ❌ Questions might go to the wrong specialist
• ❌ Users would have to manually select agents for every question
• ❌ The intelligent routing system breaks completely

Synthesizer Agent

The Synthesizer Agent combines information from multiple sources into one coherent, comprehensive answer.

What It Does

The Synthesizer Agent is responsible for:

• Collecting information from multiple specialist agents
• Analyzing and combining different pieces of information
• Creating unified answers that are coherent and well-structured
• Presenting complete responses instead of disconnected fragments

How It Works

Step-by-Step Process:

1. Multiple agents gather information

   • Different specialist agents retrieve relevant data
   • Each agent contributes specific knowledge from their domain

2. Synthesizer Agent receives all pieces

   • Collects responses from all contributing agents
   • Has a complete picture of available information

3. Synthesizer Agent analyzes the information

   • Identifies connections between different pieces
   • Detects contradictions or redundancies
   • Determines the most important points

4. Synthesizer Agent combines everything

   • Creates a logical narrative structure
   • Integrates all relevant information
   • Ensures consistency and clarity

5. User receives a complete answer

   • One coherent response instead of separate fragments
   • All information properly contextualized
   • Easy to understand and act upon

Real-World Example

User asks: "What's the status of our authentication service deployment?"

This complex question requires information from multiple sources:

Information Gathering Phase:

1. Jira Agent checks tickets

   • Found: "2 open tickets related to auth service deployment"
   • Found: "1 ticket marked as 'Ready for Production'"

2. Confluence Agent searches documentation

   • Found: "Deployment procedure updated yesterday"
   • Found: "New security requirements added"

3. GitHub Agent reviews code

   • Found: "3 commits merged to main branch today"
   • Found: "All CI/CD checks passed"

4. Monitoring Agent checks system health

   • Found: "Current auth service: 99.9% uptime"
   • Found: "No active alerts or incidents"

Synthesizer Agent's Work:

Without Synthesizer (disconnected fragments):

Fragment 1: "There are 2 open Jira tickets..."
Fragment 2: "The documentation says..."
Fragment 3: "GitHub shows 3 commits..."
Fragment 4: "Monitoring indicates..."

With Synthesizer (coherent answer):

Authentication Service Deployment Status

Current State:
The authentication service is running stable with 99.9% uptime 
and no active incidents.

Recent Activity:
• 3 commits merged today, all CI/CD checks passed
• Deployment procedure updated yesterday with new security requirements

Pending Work:
• 2 open tickets related to deployment
• 1 ticket is marked as "Ready for Production"

Recommendation:
The service is stable and code is ready. Review the pending 
tickets before proceeding with the next deployment.

Why You Can't Delete It

What happens without the Synthesizer Agent:

• ❌ Answers are fragmented and disconnected
• ❌ Users receive separate responses from each agent
• ❌ No context or connections between information pieces
• ❌ Users must manually combine and interpret data
• ❌ Queries involving multiple sources become confusing

How They Work Together

The Router and Synthesizer agents form a complete system:

User Question
     ↓
[Router Agent] → Analyzes question
     ↓           Selects agent(s)
     ↓
[Specialist Agents] → Gather information
     ↓                Use connections
     ↓                Retrieve data
     ↓
[Synthesizer Agent] → Combines information
     ↓                 Creates coherent answer
     ↓
Complete Answer to User

Example: Multi-Step Query

User asks: "Should we deploy the new feature today?"

Router Agent decides: This needs multiple sources

• DevOps Agent (deployment readiness)
• Jira Agent (open issues)
• Monitoring Agent (system health)
• GitHub Agent (code status)

Specialist Agents work:

• Each agent gathers their specific information
• All agents return their findings

Synthesizer Agent combines:

• Code is ready ✅
• 2 critical bugs still open ⚠️
• System health is good ✅
• Last deployment had issues ⚠️

Final Answer: "Not recommended. While the code and system are ready, there are 2 critical bugs still open. Previous deployment issues suggest waiting until those are resolved. Estimated timeline: 2 days."

Protected Status in the UI

Both system agents have special protections in the Junet interface:

Visual Indicators

• 🔒 Lock Icon: Displayed next to agent names
• "System Agent" Badge: Clear label indicating special status
• Disabled Delete Button: Cannot be removed accidentally
• Tooltip on Hover: Explains why the agent is protected

What You CAN'T Do

❌ Delete system agents ❌ Disable system agents ❌ Modify their core behavior ❌ Change their routing/synthesis logic

What You CAN Do

✅ View system agent activity in logs ✅ See which agents were used for each query ✅ Monitor system agent performance ✅ Create and manage your own specialist agents

Best Practices

Trust the System Agents

Optimize for the Router

Since the Router Agent uses names and descriptions to make decisions:

Good Agent Configuration:

• Name: "Confluence DevOps Documentation Agent"
• Description: "Expert in CI/CD deployment procedures, infrastructure documentation, and DevOps best practices from the DevOps Confluence space"

Bad Agent Configuration:

• Name: "Agent 1"
• Description: "Helps with stuff"

The Router Agent needs clear, descriptive information to make good routing decisions!

Design for Synthesis

When creating specialist agents, consider how their responses will be combined:

Good Response Structure:

• Clear, concise information
• Proper context included
• Specific data points
• Actionable insights

Poor Response Structure:

• Vague statements
• Missing context
• Unclear references
• Ambiguous data

The Synthesizer Agent works best when specialist agents provide well-structured, contextual information!

Monitoring System Agents

While you can't modify system agents, you can monitor their performance:

Available Metrics

• Routing Decisions: See which agents were selected for queries
• Routing Accuracy: Track if users refine queries (indicates good/bad routing)
• Synthesis Quality: Monitor user feedback on combined answers
• Response Times: Track how long system agents take to process

Troubleshooting

If you notice routing issues:

1. Check specialist agent names and descriptions

   • Are they clear and specific?
   • Do they accurately describe capabilities?

2. Review query patterns

   • Are users asking ambiguous questions?
   • Are there gaps in specialist agent coverage?

3. Analyze routing logs

   • Which agents are being selected?
   • Are there patterns in routing decisions?

Summary