OpsPilot — conversational AI copilot for on-call and SRE

The description states that OpsPilot is a conversational incident-response copilot for SREs and on-call engineers. It operates within a Kubernetes cluster and connects to tools like Prometheus, logs, deployment history, and alerts. The system investigates incidents by gathering evidence from these sources, presenting root-cause conclusions with cited evidence, and offering human-approved remediation actions.

It includes:

• Automated investigation triggered by alerts
• Evidence-based root cause analysis using GPT-5.6
• Dry-run previews of proposed changes
• Human approval required for any action
• Audit trail recording all steps
• Postmortem drafting from the audit trail

The system is demonstrated in a local Kubernetes test environment (kind cluster) and uses:

• FastAPI backend
• React/TypeScript frontend
• Prometheus metrics
• SQLite for storage
• OpenAI Responses API with GPT-5.6

Inference The product is not yet deployed in production, but rather a prototype built as part of a hackathon project.

The author positions OpsPilot as an AI-powered assistant that reduces time spent on incident response by consolidating scattered data into actionable insights. It claims to:

• Reduce alert fatigue
• Improve consistency in triage
• Make root cause analysis more accessible through conversational interface
• Ensure safety via human approval and audit trails

The description also notes that the system avoids inventing data or making assumptions — it only answers based on what is present in the environment.

Inference The positioning reflects a desire to solve a common SRE pain point (tool switching, manual investigation) with AI. However, the claim of solving these issues is limited to the demo context and lacks external validation.

The description states that OpsPilot targets:

• On-call engineers
• SREs
• DevOps architects

It aims to reduce time spent switching between tools during production incidents.

Inference The target customer segment appears to be enterprise-level teams managing Kubernetes infrastructure, but there is no evidence of actual customers or market fit beyond the author’s personal experience.

There is no mention of pricing, monetization strategy, or business model in the description. The project is presented as a hackathon submission with no indication of commercial intent or revenue generation.

Not evidenced

The system uses:

• GPT-5.6 via OpenAI Responses API
• Kubernetes (kind) for simulation
• Prometheus for metrics
• React/TypeScript frontend
• FastAPI backend
• SQLite for audit logging
• Codex for development assistance

It includes:

• Evidence gathering from logs, metrics, events
• Structured reasoning with cited evidence
• Dry-run previews of changes
• Allowlisted remediation actions
• Independent verification of recovery
• Audit trail and postmortem generation

Inference The architecture is modular and designed around safety and traceability. However, the system has not been tested in a live production environment.

The project is described as a hackathon submission (OpenAI 2026). It includes:

• A working end-to-end demo
• 59 passing tests
• A claim-verification ledger
• Integration-ready with Alertmanager, PagerDuty, Datadog

There is no evidence of:

• Revenue
• Customers
• Product-market fit
• Live usage
• Production deployment

Not evidenced

The description does not name competitors or reference existing solutions in the SRE/DevOps incident response space. It implies that current tools require engineers to manually switch between platforms and lack AI-driven automation.

Inference The competitive landscape likely includes traditional monitoring and alerting tools (e.g., Prometheus, Grafana, PagerDuty), but no direct comparison or differentiation is made in the description.

• Demo-only scope: The system works only in a controlled Kubernetes test environment.
• No real-world usage: No evidence of adoption or feedback from actual users.
• Limited tooling: Relies heavily on GPT-5.6 and Codex, which may not scale or be production-ready.
• Unproven trustworthiness: While safety mechanisms are described, their effectiveness in practice is unverified.
• No commercialization plan: No mention of monetization, pricing, or go-to-market strategy.

1. What specific problems in your own SRE workflow did you observe that led to building this?
2. How would the system behave if it encountered missing data or incomplete logs?
3. Can you walk us through a scenario where the AI incorrectly identified root cause or proposed an unsafe fix?
4. Have you tested this with any real-world teams or in actual production environments?
5. What are your plans for integrating with existing alerting and incident management systems beyond Alertmanager?
6. How do you plan to scale this beyond a single cluster or demo environment?