AI does rounds on your production.
Self-hosted AI SRE — investigate incidents, build dashboards, manage alerts, and approve remediations from natural language.
Quick Start · What it does · Deploy · Docs
▶ Watch the 1-minute demo on YouTube
You'll need a Kubernetes cluster and an API key for an LLM provider (Anthropic, OpenAI, Gemini, DeepSeek, Azure OpenAI, or a local Ollama endpoint). Rounds cannot investigate anything without a model, so have the key ready before you start — the setup wizard asks for it and won't let you past that step.
helm install rounds oci://ghcr.io/syntropize-ai/charts/rounds \
--namespace observability --create-namespaceThe chart creates a private ClusterIP service. To open it locally:
kubectl -n observability port-forward svc/rounds 3000:80Then open http://127.0.0.1:3000. The setup wizard walks through six steps: create an admin account, paste your LLM key, and connect a datasource. The connector and notification steps can be skipped and done later in Settings.
Once you're in, try:
Create a dashboard for HTTP latencyAlert me when p95 latency is above 500ms for 10 minutesWhy is checkout latency high right now?
Questions about your own systems need a datasource connected — point Rounds at your Prometheus in the wizard's Connectors step, or later under Settings → Connectors.
For shared access, expose Rounds with Ingress or service.type=LoadBalancer — see docs/install/kubernetes.md.
To see the product before committing to an install, the npm package runs the
same binary on a laptop with SQLite under ~/.rounds/:
npx @syntropize/roundsThen open http://localhost:3000. Single-node and unauthenticated by default — fine for evaluation, not for production.
- Observe — create, edit, clone, explain, and delete dashboards from natural language.
- Detect — create and tune alert rules through chat, with preview and backtest before save.
- Investigate — correlate metrics, logs, recent changes, and (when connected) Kubernetes state, with citations on every claim.
- Remediate safely — propose fixes; user-driven actions confirm in chat (Run / Confirm / Apply), background-agent actions go through formal approval (Approve / Reject / Modify) with owner / on-call notification.
- Configure by chat — add datasources, ops connectors, and low-risk org settings through the agent (gated by RBAC and the GuardedAction risk model).
Kubernetes is the first deep production workflow. Planned integrations include Prometheus alerting rules, Loki log routing, GitHub deploys, Jira / PagerDuty incident sync, CI/CD systems, and database read connectors — these are clearly marked as PLANNED in the docs and not promised by the current release.
helm install rounds oci://ghcr.io/syntropize-ai/charts/rounds \
--namespace observability --create-namespaceBy default the chart runs SQLite on a PVC at /var/lib/rounds/rounds.db. That's fine for evaluation; for production set secretEnv.DATABASE_URL before first start so every Rounds repository uses Postgres:
helm install rounds oci://ghcr.io/syntropize-ai/charts/rounds \
--namespace observability --create-namespace \
--set-string secretEnv.DATABASE_URL='postgres://user:pass@host:5432/rounds'Treat the database backend as an install-time choice — changing it later does not migrate existing data.
The default service is ClusterIP. Production deployments should front it with Ingress (ingress.enabled=true) or switch to service.type=LoadBalancer. Full options in docs/install/kubernetes.md.
helm template rounds ./helm/rounds # render manifests locallySee Try it without a cluster above, or the npm install guide for persistent installs, upgrades, and where data lives.
git clone https://github.com/syntropize-ai/rounds.git && cd rounds
npm install
npm run build
npm run startAGPL-3.0-or-later — Copyright (c) Syntropize.