Incident-to-release autopilot for SaaS teams.
ReleaseOps Pilot embeds the GitHub Copilot SDK inside a web application so support, SRE, and engineering teams can turn fragmented customer escalations, Azure telemetry, and GitHub delivery context into a validated release response. The app runs an agent loop that plans the investigation, collects Microsoft 365 signals through Work IQ, queries Azure Monitor, inspects GitHub changes, runs validation checks, and produces ready-to-use artifacts: an incident summary, root-cause hypotheses, a GitHub issue, a patch plan, a canary checklist, and a customer update.
ReleaseOps Pilot is a Copilot SDK powered incident-to-release workspace for SaaS teams. A product or support lead describes a customer-impacting regression, and the embedded agent plans the response, gathers Microsoft 365 context with Work IQ, pulls Azure Monitor telemetry, inspects recent GitHub changes, runs validation gates, and produces the artifacts needed to act: severity, likely root cause, patch plan, GitHub issue draft, customer update, canary rollout checklist, and executive summary. The local demo ships with realistic sample signals so the product works immediately; authenticated teams can switch to live Copilot SDK, Azure, GitHub, and Work IQ integrations with environment variables and MCP configuration.
Modern incidents are not only telemetry problems. The signal lives across customer emails, Teams threads, dashboards, pull requests, deployments, and tribal knowledge. ReleaseOps Pilot creates a single guided workspace for the person coordinating the response.
| User | Pain | Outcome |
|---|---|---|
| Support lead | Customer escalations arrive faster than engineering can triage | Customer-ready update with evidence and next steps |
| SRE | Telemetry points to symptoms, not ownership | Risk-ranked blast radius and canary guardrails |
| Engineering manager | Root cause, fix, and release plan live in different tools | GitHub issue plus implementation and validation plan |
| Founder / product owner | Customer trust depends on crisp response | Clear executive summary and measurable recovery plan |
- Plan - Break the request into investigation, validation, release, and communication tasks.
- Collect - Pull customer signal, Azure telemetry, GitHub changes, and runbook context.
- Diagnose - Rank hypotheses with supporting evidence and confidence.
- Validate - Generate tests, rollback checks, and canary thresholds.
- Package - Produce an incident report, issue draft, customer note, and release checklist.
Requires Node.js 20 or newer.
git clone <your-repo-url>
cd releaseops-pilot
npm install --omit=optional
npm run demo
npm startOpen http://localhost:8080, choose Run sample investigation, and review the generated artifacts.
The default runtime is deterministic demo mode, which makes the product easy to run without external accounts. To use the live GitHub Copilot SDK runtime:
npm install
export COPILOT_RUNTIME=sdk
export COPILOT_GITHUB_TOKEN=<token-with-copilot-access>
export COPILOT_MODEL=gpt-5
npm startOptional live integrations:
export GITHUB_TOKEN=<repo-token>
export GITHUB_REPOSITORY=owner/repo
export AZURE_MONITOR_WORKSPACE_ID=<workspace-id>
export AZURE_MONITOR_BEARER_TOKEN=<bearer-token>
export WORKIQ_MCP_ENABLED=trueWork IQ is configured in mcp.json; APM setup is declared in apm.yml.
npm start # Start the web app on PORT or 8080
npm run demo # Run the sample investigation and write artifacts to artifacts/demo-run
npm test # Run unit tests using node:test
npm run check # Syntax-check JavaScript files and run testsreleaseops-pilot/
app/ Web server and browser client
src/agent/ Copilot SDK runtime, demo runtime, tools, prompts
src/services/ Work IQ, Azure Monitor, GitHub, validation services
src/workflows/ End-to-end incident workflow orchestration
data/ Realistic sample signals for instant demos
docs/ Setup, architecture, deployment, RAI, demo guide
presentations/ Two-slide pitch deck
website/ Static GitHub Pages site
infra/ Azure Container Apps deployment starter
.github/workflows/ CI and Pages workflows
The SDK runtime lives in src/agent/copilotSdkRuntime.js. It creates a Copilot session with:
- custom system instructions for incident response,
- tool definitions for Work IQ signal collection, Azure telemetry, GitHub context, validation, and artifact packaging,
- a scoped permission handler that denies shell access while allowing read-only and registered tool calls,
- an agent prompt that asks for structured JSON output aligned to the app schema.
src/agent/deterministicRuntime.js mirrors the same tool loop for offline demos and CI.
- Azure Monitor / Application Insights -
src/services/azureMonitor.jsincludes production KQL queries and a live REST path when a workspace token is provided. - Azure Container Apps -
infra/main.bicepandinfra/azure.yamlprovide a deployable hosting baseline. - Microsoft 365 via Work IQ -
mcp.jsonwires the Work IQ MCP server for emails, meetings, documents, Teams messages, and people context. - APM -
apm.ymldeclares portable agent context, MCP dependencies, and project instructions.
- Demo video:
demo-video/releaseops-pilot-demo.mp4 - Demo script:
docs/DEMO_VIDEO_SCRIPT.md - Pitch deck:
presentations/ReleaseOpsPilot.pptx - Generated sample run:
artifacts/demo-run/incident-report.md - GitHub Pages site:
website/index.html
ReleaseOps Pilot keeps generated recommendations decision-support oriented. It preserves evidence links, labels assumptions, requires human approval before customer-facing communication, blocks destructive shell tools in the SDK runtime, and records every artifact produced by the workflow. Full notes are in docs/RESPONSIBLE_AI.md.
See docs/DEPLOYMENT.md for local, Docker, Azure Container Apps, and GitHub Pages deployment steps.
MIT. See LICENSE.