PagerZero Pro

The autonomous AI Site Reliability Engineer (SRE) pager that fixes routine incidents while you sleep, and calls your phone for one-word voice approvals when critical fixes are needed.

Comment

Inspiration

PagerZero is an end-to-end autonomous incident remediation platform and pager replacement built on CALL-E:

  • Autonomous Diagnostics & Root Cause Analysis: Ingests telemetry spikes and alert webhooks, evaluates system metrics, traces logs, and scores risk.
  • Tier 1 (Safe Policy Auto-Remediation): Low-risk, idempotent failures (e.g., runaway log disks, stale cache buildup) are remediated automatically by executing verified runbooks without waking anyone up.
  • Tier 2 (CALL-E Voice Approval): High-impact actions (e.g., recycling production database connection pools, restarting critical microservices, tripping circuit breakers) trigger an outbound phone call via CALL-E. The on-call engineer hears a concise, 15-second voice briefing and simply replies "Approved" or "Rejected". PagerZero executes the fix, verifies health telemetry, and lets the engineer go back to sleep.
  • Tier 3 (Escalation & Safety Fallback): If a call times out, the engineer rejects the plan, or health checks fail to recover, PagerZero escalates to secondary alerts.
  • Interactive Mission Control: A real-time web dashboard streaming cluster telemetry over WebSockets, with built-in Chaos Outage simulators, live audio waveform visualizers, and instant post-mortem markdown generators.

How We Built It

  • Telephony & Agent Voice Pipeline: Integrated CALL-E's @call-e/calle SDK to dial real phone lines, run structured conversational tasks, and extract strict JSON schema outputs (approved: boolean, reason: string). We also built an in-browser Web Speech simulator for testing and zero-credit demonstrations.
  • SRE Reasoning Engine: Developed in TypeScript on Node.js/Express, featuring an autonomous diagnostic pipeline that matches telemetry signatures to remediation runbooks and calculates risk coefficients.
  • Frontend Dashboard: Built with React 18, Vite, and Tailwind CSS. Features live WebSocket telemetry broadcasting (CPU, memory, latency, connection pools), one-click chaos injection, and interactive post-mortem review modals.
  • Production Deployment & Edge Proxy: Packaged into a lightweight Alpine multi-stage Docker container managed via Docker Compose and fronted by Caddy for automated Let's Encrypt SSL at https://pagerzero.pro.

Challenges We Ran Into

  • Telephony Latency vs. Briefing Clarity: Waking an engineer requires an ultra-concise voice prompt. We had to tune the agent's briefing so it delivers the service name, problem, and proposed action within 10 to 15 seconds without cutting off the user's immediate voice response.
  • Strict Schema Extraction over PSTN: Human responses over phone lines can be noisy ("Yeah, go ahead", "Do it", "Approve"). Structuring CALL-E's task prompt to deterministically parse conversational affirmations into a boolean was essential for safe automation.
  • Safe State Machines & Verifications: Ensuring an agent doesn't enter an infinite restart loop required strict pre-remediation baselines and post-remediation health polling before marking an incident resolved.

Accomplishments That We're Proud Of

  • Built a zero-friction voice experience where an outage that normally takes 15–20 minutes of manual laptop triage is resolved over an 18-second phone call.
  • Seamless multi-tier escalation policy that balances safe autonomous fixes with human-in-the-loop governance.
  • Production-ready packaging with automated SSL and domain deployment on pagerzero.pro.

What We Learned

Voice interfaces for developer operations (DevOps/SRE) shouldn't be open-ended chatbots; they thrive on goal-oriented voice contracts. When telephony models return structured, validated schemas instead of freeform text, the phone becomes an API for physical authorization.

What's Next for PagerZero

  • Direct Monitoring Integrations: Pre-built webhook adapters for Datadog, Prometheus Alertmanager, New Relic, and AWS CloudWatch alarms.
  • Dynamic Bi-directional Voice Q&A: Allowing the engineer on the call to ask follow-up questions (e.g., "What was the P99 latency right before it spiked?") before granting voice approval.
  • Multi-Cluster Kubernetes Operator: A native Helm chart and CRD for deploying PagerZero alongside Kubernetes clusters to execute rolling pod restarts and resource scaling natively.

Built With

Share this project:

Updates

Submission history