Getting Started

Rounds is an AI SRE that lives next to your existing telemetry and operations tools. Get it running in under five minutes — no schema changes, no second copy of your data.

1. Install

Pick the install path that matches your environment:

npm (single machine)

npx @syntropize/rounds

Helm (Kubernetes)

helm upgrade --install rounds \
  oci://ghcr.io/syntropize/charts/rounds \
  --namespace observability \
  --create-namespace \
  --set secretEnv.LLM_API_KEY='replace-with-your-provider-key'

For more detail, see Install with npm or Install with Helm.

2. Open the web UI

The web UI runs on http://localhost:5173 (npm install) or whatever Ingress / port-forward you configured for the Kubernetes deployment.

The setup wizard walks you through the minimum needed to start asking questions:

1. Create your administrator account (name, email, password — minimum 12 characters)
2. Configure an LLM provider — paste an Anthropic, OpenAI, or Gemini API key, point at a local Ollama server, or supply an apiKeyHelper script for short-lived / vault-issued credentials
3. Add a metrics datasource — Prometheus, VictoriaMetrics, Mimir, Thanos, Cortex, or any Prometheus-API-compatible backend
4. Optionally add Loki for log search and a Kubernetes ops connector so investigations can inspect pods, events, rollouts, and prepare approval-gated remediations

You don't have to do step 3 or 4 in the wizard — once Rounds is running, you can also add datasources, ops connectors, and low-risk org settings by chatting with the agent ("connect my prod Prometheus at http://..."). The agent collects what it needs, previews the change, and applies it under your RBAC and the GuardedAction risk model.

3. Try a prompt

Once setup is complete, click the chat button and ask:

Create a dashboard for HTTP latency

Rounds will discover your metrics, build queries, validate them, and create a dashboard with overview stats, trend charts, and per-handler breakdowns — all grounded in your actual data.

Then try an investigation prompt:

Why is checkout latency high right now?

If metrics, logs, and a Kubernetes connector are configured, Rounds will query telemetry, inspect cluster state, write a report with citations on every claim, and recommend next actions. Mutating cluster actions are never executed silently:

• When you ask the agent to do something risky, it surfaces a Run / Confirm / Apply prompt inline in chat.
• When the agent is running unattended (auto-investigation triggered by a firing alert), the proposed fix is delivered as a RemediationPlan with formal Approve / Reject / Modify controls; the owning team / on-call is notified.

Common first prompts

Goal	Prompt
Build a dashboard	Create a dashboard for checkout latency and errors
Edit a dashboard	Add p99 latency by route and remove panels with no data
Understand a dashboard	Explain what this dashboard tells me and what looks abnormal
Create an alert	Alert me when p95 latency is above 500ms for 10 minutes
Investigate an alert	Why did the high latency alert fire?
Investigate the cluster	Check whether Kubernetes is causing the latency spike

What's next

• Configuration — environment variables for production tuning
• Chat & agents — dashboard, alert, investigation, and remediation workflows
• Authentication — adding users, OAuth providers, role-based access control
• API Reference — automate via REST and service account tokens