EZ OnCall - Voice-First DevOps Incident Agent

EZ OnCall is a voice-first DevOps incidents agent that replaces noisy Slack alerts and dashboards with a short, focused phone call. When our application errored in production, an alert was sent to EZ OnCall n8n workflow to trigger our on-call workflow. The AI agent immediately calls the on-call engineer via ElevenLabs & Twilio over phone, explains the incident using an ElevenLabs AI Agent with full capability of immediate remedy tools like “rollback”, ”fix”, or “escalate” implemented through n8n workflows!

The core flow shows a full end-to-end loop: a real HTTP 500 in NightOwl’s checkout service produces a error alerting event, which triggers a webhook into our backend. The backend creates an incident record, initiates a call, generates a summary, and acts on the engineer’s voice command by triggering n8n workflows (e.g., reboot, rollback, fix, open a GitHub issue).

---

Workflow

1. NightOwl Tickets breaks

• A user triggers the “checkout” flow in the NightOwl Tickets app.
• The /api/checkout endpoint returns HTTP 500 with CheckoutServiceTimeoutError.

2. Error monitoring platform sends the alert

3. EZ OnCall creates an incident and calls the engineer

• The webhook handler parses the sample sentry payload into a clean incident object.
• EZ OnCall stores/updates the incident in Redis.
• EZ OnCall uses ElevenLabs with Twilio to place an outbound call to the on-call engineer.

4. Voice summary + voice command

• When the engineer answers, elevenlabs ai agent generates a short summary from the incident and sends it to ElevenLabs for TTS.
• Twilio plays the ElevenLabs audio: a concise explanation plus options (“Say rollback to roll back the latest deploy, or escalate to page a senior engineer”).
• ElevenLabs agent picks the right tools to execute actions.

5. Agent executes tools via n8n

• The backend parses the speech transcript (simple intent mapping: “rollback”, “reboot”, “fix”).
• For “rollback”, EZ OnCall calls an n8n webhook that simulates a rollback (e.g., hitting a fake deploy/rollback endpoint) and returns a result.
• for “fix”, EZ OnCall creates a GitHub issue with error context and a snippet of the call transcript. The issue is assigned to github copilot to be worked on.

---

Judging Criteria

1. Working Prototype

• End-to-end functionality:
  • Real failing HTTP endpoint (/api/checkout) and crash endpoint in NightOwl Tickets.
  • Sending error events via webhook.
  • EZ OnCall backend creating incidents and initiating outbound Twilio calls.
  • ElevenLabs generating voice summaries for the live incident.
  • Voice commands captured and tools are called through n8n workflows.
• The system runs on a DigitalOcean droplet with a long-lived Node process inside docker.

2. Technical Complexity & Integration

• Multiple systems integrated:
  • Application + API (NightOwl Tickets: React/Tailwind + Node/Next backend).
  • Monitoring (error reporting via webhook).
  • Elevenlabs Agents Platform (Twilio Programmable Voice for outbound calls and webhooks).
  • Voice AI (ElevenLabs).
  • Workflow engine (n8n for rollback/fix/escalate flows).
  • State storage Redis for reports and archives .
  • Source control + issues (GitHub, with CodeRabbit for code review in the broader pipeline).

3. Innovation & Creativity

• Moves incident response from clicking dashboards to speaking on a phone call:
  • The agent behaves like a “voice runbook”: it calls you, explains what’s wrong, and asks what to do.
• Uses a realistic demo app (NightOwl Tickets).
• Connects familiar DevOps building blocks (Error Logging, Twilio, automation, GitHub) through a conversational interface rather than another web UI.
• Extensible design: new tools (e.g., real rollback pipelines, CodeRabbit review) can be added as additional voice commands.

4. Real-World Impact

• Targets on-call engineers and SREs dealing with:
  • Production errors at bad times (overnight, away from a laptop).
  • Alert overload and context switching between Sentry, logs, dashboards, and terminals.
• Potential improvements:
  • Faster initial triage and mitigation for common incidents (“roll back last deploy”, “restart service”).
  • Lower cognitive load: the agent surfaces just the key incident details and options.
  • Keeps human approval in the loop: no remediation runs until the engineer explicitly says “rollback” or “escalate.”
• Built using tools that teams already use (Sentry, GitHub, Twilio, n8n), so the concept can be dropped into real organizations.

5. Theme Alignment – Conversational Agents

• EZ OnCall is a true conversational agent:
  • Listens: receives monitoring events from Sentry and speech from the engineer.
  • Reasons: turns structured error data into a spoken summary and maps human commands to tools.
  • Acts: invokes concrete tools (rollback workflow, issue creation, escalation) and logs the result.
• Multimodal:
• Agentic behavior:
  • Autonomous in detection and initiation (it calls you when something breaks).
  • Tool-using agent: uses n8n, GitHub, SSH/deploy/redeploy endpoints as its actuators.