Support Automation

Build an L1/L2 support system where agents handle routine questions across Slack, email, and Teams — escalating to senior agents or humans only when they can't resolve the issue. Most teams see 60–80% of tickets resolved without human intervention.

What you'll have

Agents answering support questions in real-time across 3+ channels
Automatic escalation when confidence is low or timeout is reached
A growing knowledge base that improves resolution rates over time
Daily standup reports showing ticket volume, resolution rates, and blockers

Architecture

Component	Role
`support-agent`	L1 — handles routine questions using knowledge base
`senior-support`	L2 — resolves escalated issues with deeper tooling
Escalation chain	Routes unresolved tickets through L1 → L2 → human
Team KB	Shared knowledge base that agents contribute to and search
Channels	Slack, email, Teams — customers reach agents where they already are

Step 1: Configure agents and channels

yaml

# astra.yml — support automation setup
agents:
  - id: support-agent
    systemPromptTemplate: |
      You are a technical support agent for {{company_name}}.
      Answer questions using the knowledge base. If you cannot resolve
      the issue with confidence > 0.7, escalate immediately.
    providers: [claude, groq]
    tools:
      allow: [knowledge_retrieve, web_search, jira_integration]
    budget:
      maxCostCents: 50
      maxToolCalls: 15

  - id: senior-support
    systemPromptTemplate: |
      You are a senior support engineer. You receive escalated tickets
      that L1 could not resolve. You have access to internal docs and
      can execute diagnostic tools.
    providers: [claude]
    tools:
      allow: [knowledge_retrieve, code_execute, database_query, shell_execute]
    budget:
      maxCostCents: 200
      maxToolCalls: 30

channels:
  slack:
    enabled: true
    defaultAgent: support-agent
  email:
    enabled: true
    defaultAgent: support-agent
  teams:
    enabled: true
    defaultAgent: support-agent

Step 2: Set up escalation

The escalation chain gives support-agent 5 minutes to resolve. If it can't (or confidence drops below 0.7), the ticket escalates to senior-support with 15 minutes. After that, it goes to a human.

bash

# Create the escalation chain
curl -X POST http://localhost:3000/escalation-chains \
  -H "Authorization: Bearer ${JWT_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "support-escalation",
    "levels": [
      { "agentId": "support-agent", "timeoutSeconds": 300, "confidenceThreshold": 0.7 },
      { "agentId": "senior-support", "timeoutSeconds": 900 },
      { "agentId": "human-escalation", "timeoutSeconds": 3600 }
    ]
  }'

Step 3: Load your knowledge base

Pre-load common support procedures. Agents will also contribute new entries as they resolve tickets — the KB grows automatically.

bash

# Load your support knowledge base
curl -X POST http://localhost:3000/team-kb/entries \
  -H "Authorization: Bearer ${JWT_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "agentId": "support-agent",
    "title": "Password Reset Procedure",
    "content": "1. Verify identity via email. 2. Generate reset link...",
    "tags": ["auth", "password", "common"]
  }'

Step 4: Monitor with standups

Generate a daily standup report to track ticket volume, resolution rates, tool usage, and blockers.

bash

# Generate daily support standup
curl -X POST http://localhost:3000/standups/generate \
  -H "Authorization: Bearer ${JWT_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{ "agentIds": ["support-agent", "senior-support"] }'

Tuning

Lower the confidenceThreshold to reduce escalations (at the cost of some inaccurate responses)
Add more entries to the Team KB to improve L1 resolution rates
Use peer learning to automatically extract strategies from senior-support and apply them to support-agent
Monitor costs on the cost dashboard — Groq as primary provider keeps L1 costs very low