Custom workflows

Custom AI workflows to take care of the DevOps task for you and your team

What are custom workflows

Custom workflows let you define automated tasks that Cased’s DevOps agent performs on a schedule or trigger. They work just like the default workflows that ship with Cased, but can be tailored to your organization’s specific needs and processes.

How custom workflows work

Cased’s agent excels at devops tasks like monitoring infrastructure, managing deploys, and debugging production issues. Custom workflows enable you to receivesthe agent specific instructions to handle devops tasks automatically.

Each workflow combines three key elements:

Triggers

Workflows can be triggered in three ways:

• Scheduled - Run the workflow on a regular schedule (daily, weekly, monthly)
• Webhooks - Trigger when Cased receives a webhook from external systems
• API - Start on-demand via API call

Context

The agent needs access to your systems to be effective. Connect these integrations to give workflows the context they need:

Essential integrations:

• AWS - Monitor resources, costs, and security
• GitHub - Analyze code, review PRs, track deployments

Recommended additions:

• Slack - Send notifications and alerts
• Sentry - Analyze errors and exceptions
• DataDog - Monitor metrics and logs
• PagerDuty - Respond to incidents

Workflow prompt

The prompt provides clear instructions for what the agent should do. Write prompts as if explaining the task to a senior engineer.

Example prompt for a deployment health check:

After each deployment to production:
1. Check error rates in Sentry for new exceptions
2. Monitor CPU and memory usage in DataDog
3. Verify all health checks are passing
4. Review deployment logs for warnings
5. If issues found, summarize in Slack #deployments channel
6. If critical issues, create PagerDuty incident

Creating your first workflow

1. Navigate to Workflows in your Cased dashboard

2. Click “Create workflow”

3. Configure the basics:

   • Name: Descriptive name like “Daily Infrastructure Audit”
   • Trigger: Choose scheduled, webhook, or API
   • Schedule: Set timing if using scheduled trigger

4. Write your prompt:

   • Be specific about what to check
   • Include success/failure criteria
   • Specify where to report results

5. Test the workflow:

   • Use “Run now” to test immediately
   • Review the agent’s actions and output
   • Refine prompt based on results

6. Enable and monitor:

   • Turn on the workflow when ready
   • Check execution history regularly
   • Iterate based on effectiveness

Best practices

Write clear prompts

✅ Good prompt:

Check all EC2 instances for:
- Instances running longer than 30 days without restart
- Instances without recent backups
- Development instances running outside business hours
Report findings in #aws-alerts with instance IDs and recommendations

❌ Vague prompt:

Check our servers and let me know if anything looks wrong

Start simple, then expand

Begin with basic workflows and add complexity as you validate they work:

1. Start: Check S3 buckets for public access
2. Expand: Also check for missing encryption
3. Enhance: Add automatic remediation for common issues

Use parameters for flexibility

Make workflows reusable across environments:

Check database backup status for {{ environment }}:
- Verify backup completed in last {{ backup_hours }} hours
- Check backup size is within {{ size_variance }}% of average
- Alert {{ slack_channel }} if issues found