Yes. For most home servers, self-hosting the open source Netdata agent is the cheapest option over three years and the one that leaks the least data. The software is free under GPL-3.0. Over 36 months you pay for a little extra electricity, some disk space for stored metrics, and your own time. Netdata Cloud's free tier is only worth adding if you want to see several machines from one place. Datadog bills per host every month and sells features a home user will rarely touch, so you pay more and also send your server metrics to a third-party company.
TL;DR: which option fits you
- The single-box tinkerer (one mini PC running Jellyfin and a few containers): run the open source Netdata agent alone with no Cloud account, because the local dashboard covers everything one machine needs and nothing leaves your network.
- The multi-device household (a NAS, a Raspberry Pi and a desktop): use the open source agent with Netdata Cloud's free tier or a Netdata parent node, because a single dashboard across several machines is the first thing you will miss.
- The privacy-first user (no third-party accounts, ever): run the agent with anonymous telemetry turned off and never claim it to Cloud, because then your metrics stay on hardware you own.
- The side-project builder (a homelab plus one public VPS): start with the open source agent and only consider a paid Netdata Cloud plan if alert routing and longer team retention become real needs, because subscription costs add up over 36 months.
- The professional rehearsing for work (learning tools used in their day job): a Datadog trial is worth a few weeks, because the skills transfer to work, but a home setup cannot justify per-host billing for three years.
The central tradeoff is simple: Netdata Cloud and Datadog give you convenience in exchange for a monthly bill and your metrics on someone else's servers, while the self-hosted Netdata agent gives you control and costs you some setup time.
Table of contents
- What does Netdata actually monitor on a home server?
- Netdata Open Source vs Netdata Cloud vs Datadog: what you actually get
- How much CPU, RAM and disk does the Netdata agent use?
- Is the free Netdata Cloud plan enough for a home server?
- When is a paid Netdata Cloud plan worth the money?
- What would Datadog cost for the same home setup?
- The hidden cost: hours spent installing, updating and tuning alerts
- What data does Netdata send off your network, and can you turn it off?
- Data sovereignty: where your monitoring data lives with each option
- Can you see all your servers in one dashboard without Netdata Cloud?
- Where should you run Netdata: home server, NAS, VPS or managed personal cloud?
- Matching the right option to your setup
- Key points
What does Netdata actually monitor on a home server?
Netdata is an open source monitoring agent. You install it on each machine, and it starts collecting metrics within seconds without any configuration. By default it records a new value every second and serves a live dashboard on port 19999. You open it in a browser at http://your-server:19999. A home user doesn't have to write any collection rules. Netdata finds most things by itself.
On a typical home server, that covers:
- System resources: CPU per core, RAM, swap, system load and temperature sensors where the hardware exposes them, so you can tell whether a slowdown comes from your hardware or your apps.
-
Disks and filesystems: free space, read and write throughput, I/O wait, and drive health through the
smartctlcollector, which can warn you about a failing drive before you lose data. - Docker containers: CPU, memory and network use for each container, read from cgroups, so you can see which of your apps is using up the machine.
- Network traffic: bandwidth per interface, packet drops and connection counts, which helps when your media streaming stutters or a backup saturates the uplink.
- Applications: dedicated collectors for services such as Nginx, PostgreSQL, Redis and Pi-hole, picked up automatically when they run on standard ports.
- Alerts and anomalies: preconfigured health alerts for full disks or high load, plus machine learning anomaly detection running locally on the agent.
All of this runs on whatever machine you choose: a home server, a NAS, a rented VPS, or a managed platform such as Yundera. Yundera is a managed Personal Cloud Server, built on CasaOS, that runs self-hosted apps as Docker containers on a server dedicated to the user. Wherever you run it, the agent does the same job.
Netdata Open Source vs Netdata Cloud vs Datadog: what you actually get
The three options are less different than the names suggest. Netdata Cloud is not separate software. It is a hosted control panel that connects to the same open source agent you would run anyway. Datadog is a fully separate commercial platform with its own proprietary agent, and it ships your metrics to Datadog's servers for storage and analysis.
The practical differences come down to where your data ends up and how you get billed:
| Option | Where your metrics live | How you pay |
|---|---|---|
| Netdata Open Source agent alone | On each machine's own disk, in the agent's dbengine database | Nothing; GPL-3.0 licence, you supply hardware and power |
| Netdata agent + Netdata Cloud free tier | Still on your machines; Cloud queries agents live and keeps metadata such as node names and alert events | Nothing, within the free plan's limits on nodes and features |
| Netdata agent + Netdata Cloud paid plan | Same as the free tier, with extra Cloud features such as role-based access and longer alert history | Subscription, billed by number of nodes |
| Datadog | Datadog's SaaS platform; standard metric retention is 15 months on their side | Per host per month on paid tiers; the free tier caps at 5 hosts with 1 day of retention |
This table matters more than any feature checklist. With both Netdata options, your raw time series data never leaves your hardware unless you set up streaming yourself. With Datadog, sending your data to them is how the product works. You can't turn it off.
That doesn't make Datadog a bad product. It is built for companies running hundreds of hosts, with log management, APM and security tooling sold as add-ons. A home server uses very little of that, but per-host billing charges you for it anyway. The later sections put numbers on each option.
How much CPU, RAM and disk does the Netdata agent use?
Netdata collects thousands of metrics every second, which makes people expect a heavy process. In practice the agent is built to be light, but how much it uses depends on how much it monitors. A Raspberry Pi with five containers puts much less load on it than a server with 40 containers and a dozen application collectors. Instead of trusting a single headline figure, measure it on your own machine. The agent charts its own CPU and memory use under the Netdata Monitoring section of its dashboard, and docker stats netdata shows the same thing from the outside.
The main things that affect its footprint, and what you can change, are:
- Number of metrics collected: every container, disk and network interface adds charts, so load grows with the size of your setup, not with how long it has been running.
-
Collection frequency: the default is one value per second. Setting
update every = 2innetdata.confroughly halves the number of samples stored, at the cost of finer detail. -
Machine learning: anomaly detection trains models on the agent itself, which uses extra CPU. On a low-power board you can turn it off with
enabled = nounder the[ml]section. -
Disk retention: metrics go into the dbengine database in tiers, from per-second detail up to coarser long-term rollups. The
[db]section ofnetdata.confsets how much disk each tier may use, so disk growth has a cap you choose instead of growing forever. -
Enabled collectors: auto-discovery may enable collectors you don't need. Turning off unused ones in
go.d.confcuts both CPU and the number of charts.
For the cost analysis, the key point is that both RAM and disk use have a ceiling you can set. That keeps the three-year storage cost predictable.
Is the free Netdata Cloud plan enough for a home server?
For one machine, you usually don't need Netdata Cloud at all. The local dashboard on port 19999 already shows every chart and alert. Cloud becomes useful once you have two or more machines and want to see them in one browser tab instead of bookmarking a separate dashboard for each. Connecting a node is called claiming. In Docker you do it by setting the NETDATA_CLAIM_TOKEN and NETDATA_CLAIM_ROOMS environment variables, and on a bare-metal install you add the token to claim.conf.
The free plan gives a home user:
- A combined view: charts from every connected node in one place, grouped into Rooms. You can put your NAS, a Raspberry Pi and a VPS side by side.
- Alert notifications: email and the Netdata mobile app, so you hear about a full disk without leaving a dashboard open.
- Remote access: you reach your dashboards through Netdata's website without opening port 19999 to the internet.
- No metric storage on their side: Cloud fetches charts live from your agents. If a machine is offline, its history is not available through Cloud.
The limits matter more over 36 months than on day one. The free plan caps the number of nodes and keeps some features, such as some notification integrations and role-based access, for paid tiers. The cap and the feature split have changed before, so check the current Netdata pricing page before you plan for three years on the free plan.
It also doesn't matter where the agents run. A home server, a NAS, a self-managed VPS or a Yundera server can all be claimed into the same space. If you count your machines and the total sits comfortably below the free cap, the free plan is enough.
When is a paid Netdata Cloud plan worth the money?
A paid plan is billed per node, so the three-year cost is simply the per-node price × the number of nodes × 36 months. Annual billing usually lowers the per-node price. Plug your own node count into the current published price before deciding. The real question is whether the extra features replace something you would otherwise do yourself. For most home setups they don't, because the open source agent already covers a lot.
| Need | Free route with the open source agent | What a paid plan adds |
|---|---|---|
| Alerts in Discord, Telegram, Slack or ntfy | The agent sends them directly through health_alarm_notify.conf, no Cloud involved |
The same integrations configured once in a web interface for every node |
| More nodes than the free cap | A parent node collects every child's metrics through streaming and shows them on one local dashboard | Unlimited nodes in Cloud with no parent to maintain |
| Sharing dashboards with family or friends | Shared login behind a reverse proxy such as Caddy or Nginx | Separate user accounts with role-based access per Room |
| Long alert history and audit trail | Alert logs stay in the agent's local database within your retention settings | Cloud-side event history kept for you |
| Help when something breaks | Community forum and GitHub issues | Vendor support with response commitments |
Paying makes sense once your free workarounds start costing more time than they save. A typical case: someone with 10 or more machines spread over several locations who would otherwise run a parent node, a reverse proxy and per-agent notification configs. Another is a small group of people who each need their own login.
If neither describes you, the paid plan mostly buys convenience. Over 36 months that convenience costs real money, and your setup doesn't become any more private.
What would Datadog cost for the same home setup?
Datadog publishes its prices, but home users tend to underestimate the bill. That's because the headline per-host figure is only the starting point. To compare honestly with Netdata, put your setup through each of Datadog's billing dimensions and multiply by 36 months.
- The free tier: up to 5 hosts at no cost, but metrics are kept for only 1 day. A homelab can run on this indefinitely if it only wants live dashboards. You lose any history for answering "when did this disk start filling up?".
- Infrastructure per host: paid tiers charge each host every month. Annual commitments cost less per host than on-demand billing, so the three-year total depends on whether you are willing to commit up front.
- Containers: each paid host includes a fixed number of containers, and extra containers are billed separately. A single home server running 20 Docker apps can go over the included number quickly.
- Custom metrics: metrics from integrations beyond the standard set count against a per-host allotment, and anything above it is billed. Application collectors that Netdata includes for free can show up here as extra charges.
- Logs and APM: both are separate products with their own usage-based pricing, so shipping container logs adds a second bill that grows with how much your apps log.
Where your servers run doesn't change the structure. A home server, a NAS, a self-managed VPS and a Yundera box would each count as a billed host once the Datadog agent is installed. The result is a bill that grows with your setup: every new machine or extra container raises the monthly cost, while Netdata's software cost stays at zero.
The hidden cost: hours spent installing, updating and tuning alerts
Software that costs nothing still costs time. If you have no sysadmin background, that time is the biggest variable in a three-year comparison. It is spread unevenly: a burst at the start, a little each month, and occasional problems that take longer to fix.
-
Initial install: on a regular Linux machine, Netdata's official
kickstart.shscript installs the agent with a single command. The Docker route takes more setup, because thenetdata/netdatacontainer needs read access to/proc,/sysand the Docker socket, plus theSYS_PTRACEcapability, before it can see the host properly. - Updates: native installs enable an automatic updater by default and let you choose between the stable and nightly release channels. Docker installs only update when you pull a new image, by hand or with a tool such as Watchtower. Choosing stable over nightly means fewer surprises.
-
Alert tuning: the default alerts are written for typical servers, so a Raspberry Pi or a NAS that spins down its drives can trigger warnings that don't matter. You quiet them by editing files under
health.d/with theedit-confighelper in/etc/netdata. This takes trial and error. -
Retention adjustments: after a few months you will probably revisit the
[db]settings once you see how fast storage fills on your actual workload. - Troubleshooting: a collector that stops after an app update, or a container that loses host visibility after a Docker upgrade. These problems are rare, but one evening of debugging can outweigh months of routine upkeep.
How you deploy decides how much of this lands on you. On a self-managed VPS or a bare home server, you handle all of it. On Yundera, apps are installed from an app store in one click rather than assembled from compose files by hand, which removes the Docker setup step but not the alert tuning.
What data does Netdata send off your network, and can you turn it off?
A fresh Netdata install isn't completely silent. Most outbound connections are small and documented, and you can turn each one off. A privacy-conscious user should check them once at install time rather than assume the defaults suit them.
| Outbound connection | What it sends | How to turn it off |
|---|---|---|
| Anonymous telemetry | Usage and crash events: agent version, OS type, which features are enabled. No metric values | Create an empty .opt-out-from-anonymous-statistics file in /etc/netdata, pass --disable-telemetry to kickstart.sh, or set DISABLE_TELEMETRY=1 in Docker |
| Netdata Cloud connection | Node metadata, alert events, and live chart data when you open a Cloud dashboard | Never claim the node; leave NETDATA_CLAIM_TOKEN unset |
| Dashboard registry | Your browser tells a registry which Netdata dashboards you have visited, so it can list them | Point registry to announce in the [registry] section of netdata.conf to your own agent |
| Automatic updater | A regular check against Netdata's release servers for new versions | Disable auto-updates at install time, or update Docker images manually |
| Streaming to a parent | Full metrics, but only to a parent node you set up yourself | Don't configure stream.conf
|
With all of these switched off, Netdata collects, stores, analyses and alerts entirely on your own hardware. The machine learning models are trained locally, so anomaly detection doesn't need a connection either.
That's the real privacy difference from Datadog. With Netdata, every external connection can be turned off. With Datadog, sending data out is required for the product to work at all.
Make the opt-out the first thing you do after installing, not something you plan to get to later. Telemetry can start sending as soon as the agent first runs.
Data sovereignty: where your monitoring data lives with each option
Monitoring data feels harmless until you look at what it reveals. Per-second CPU and network charts show when someone is home, when the media server is streaming, when backups run, and which apps you have installed. Container names alone can list your whole digital life. Where this data is stored matters as much as where your files are.
With the open source agent alone, everything lives in the dbengine files on each machine's disk. A home server or NAS keeps it in your house. On a rented VPS it sits in the provider's data centre, under the provider's jurisdiction and terms of service, even though nobody else holds a copy. Netdata Cloud adds node metadata and alert events on Netdata's infrastructure. Datadog stores everything on its own platform. You can choose a regional site such as its EU region, but the data still sits on their servers.
Advantages of keeping metrics on your own hardware:
- No third-party profile: no company builds a picture of your daily patterns from your charts.
- No retention you don't control: deleting a dbengine directory actually deletes the data.
- No exposure to vendor breaches: a leak at a monitoring vendor can't include your metrics.
- No lock-in: switching tools leaves no history stuck on someone else's servers.
Sovereignty checklist before you commit:
-
Telemetry opted out: the opt-out file is in
/etc/netdatabefore the agent first runs. - Nodes unclaimed: unless you have chosen to use Cloud on purpose.
- Dashboard not public: port 19999 is reachable only on your LAN or through an authenticated proxy.
- Host location known: you know which country your VPS, if you use one, physically runs in.
-
Retention capped: the
[db]limits match how much history you actually want to keep.
Can you see all your servers in one dashboard without Netdata Cloud?
Yes. The built-in method is called streaming. One machine acts as the parent and the others, the children, send it their metrics in real time. The parent stores every child's data and shows all of them from its own port 19999. Nothing leaves your network.
Setting it up takes a few config edits rather than any new software:
-
Generate an API key: run
uuidgenon any machine. The resulting UUID is the shared secret between the parent and its children. -
Configure the parent: in the parent's
stream.conf, add a section named after that UUID withenabled = yes. This tells the parent to accept children that present the key. -
Configure each child: in the child's
stream.conf, setenabled = yesunder[stream], setdestinationto the parent's address and port, and setapi keyto the same UUID. -
Lighten the children: since the parent keeps the history, you can switch a child's
[db]mode toram. A Raspberry Pi then stops writing metric files to its SD card, which protects the card and saves disk. -
Size the parent: the parent's disk and RAM now carry every node's retention, so size its
[db]limits for the whole fleet, not for one machine. -
Consider an alternative: Prometheus and Grafana can scrape each agent's
/api/v1/allmetrics?format=prometheusendpoint if you would rather build your own dashboards.
There's one caveat to check on your installed version. Recent agent dashboards have tied some multi-node views to signing in to Netdata Cloud, so the parent's local interface may not show everything that Cloud's combined view does. Your metrics still stream and are stored locally either way. Try the parent's dashboard with all your nodes before you rule out Cloud completely.
Where should you run Netdata: home server, NAS, VPS or managed personal cloud?
Netdata has to run on every machine you want to monitor, so the real decision is where your parent node and your history live. That choice sets most of the three-year cost: power, storage, and how much setup you take on.
| Location | Three-year cost shape | Main tradeoff |
|---|---|---|
| Existing home server | Only the extra power from the agent's added load, since the box runs anyway, plus disk space you already own | Your data stays in your house; if that box dies, its history dies with it unless you back up the dbengine directory |
| NAS (for example via Synology Container Manager) | Same as the home server, but storage is usually plentiful and already protected by RAID | NAS operating systems can limit what a container sees of the host, so some hardware sensors may be missing |
| Dedicated Raspberry Pi as parent | One-off board and storage cost, plus a few watts running all the time | Cheap and isolated, but SD cards wear under constant writes; use an SSD for the parent's metrics |
| Rented VPS | A monthly fee for 36 months, which adds up even on the smallest plans | Monitors services that are public anyway; for home machines to stream into it you need a tunnel such as WireGuard, and your metrics leave the house |
| Managed personal cloud server | A recurring service fee instead of buying hardware | Less setup work, but read what the provider can access before you assume your data is sovereign |
For most readers the answer is whatever is already running all day. The cheapest parent is a machine you already pay to power, and the most private one sits on your own network.
A VPS only earns its place if you already rent one for other reasons. Paying 36 months of fees just to host monitoring data costs more than any cost from running it at home.
Matching the right option to your setup
The earlier sections boil down to a few profiles. Find the row closest to your setup and follow its branch below.
| Profile | Recommendation | Main reason |
|---|---|---|
| One mini PC running a handful of containers | Open source agent only | The local dashboard on port 19999 covers a single machine completely |
| NAS plus one or two Raspberry Pis | Open source agent with a parent node | Streaming gives you one view with nothing leaving the house |
| Strict no-accounts privacy user | Agent only, telemetry opted out, never claimed | Every outbound connection is switched off |
| Several machines, little time to configure | Agent with Netdata Cloud free tier | Claiming takes one token, with no stream.conf work |
| Ten or more nodes across several locations | Agent with a paid Netdata Cloud plan | Per-node fees replace a lot of parent and notification upkeep |
| A household sharing dashboards with separate logins | Agent with a paid Netdata Cloud plan | Role-based access is a paid feature |
| Learning tools used at work | Datadog free tier for a few weeks | 5 hosts with 1 day of retention is enough to practise, not to keep long term |
Next steps:
Agent only:
- Install with
kickstart.sh --disable-telemetry. - Cap retention in the
[db]section ofnetdata.conf. - Route alerts through
health_alarm_notify.conf.
Parent node:
- Generate a key with
uuidgen. - Set up
stream.confon the parent and each child. - Switch the children's
[db]mode toram.
Netdata Cloud:
- Count your nodes against the current free cap.
- Claim them with
NETDATA_CLAIM_TOKEN. - Check the pricing page again before your node count grows.
Datadog:
- Install the agent on no more than 5 hosts.
- Uninstall it before your trial or free usage turns into a subscription.
Key points
Over three years, the self-hosted Netdata agent costs the least in money and data exposure, and the main thing you pay with is your own time.
| Point | Details |
|---|---|
| The software itself is free | The Netdata agent is open source under GPL-3.0, so your three-year cost is the extra electricity from its load, the disk you set aside in the [db] section of netdata.conf, and hardware you most likely already own. There is no per-node or per-host fee for running it yourself. |
| Netdata Cloud is optional, not required | The free tier adds a combined view and alerts through the mobile app. Paid plans are billed per node, so the cost grows with every machine you add. Check the current free cap and prices before planning a full 36 months on either tier. |
| Datadog's bill grows with your setup | The free tier stops at 5 hosts with 1 day of retention. Beyond that, paid hosts, extra containers, custom metrics and logs are each billed separately every month, which suits companies far better than home servers. |
| Privacy is configurable, not automatic | Create .opt-out-from-anonymous-statistics, leave nodes unclaimed, point the registry to your own agent, and keep port 19999 off the public internet. With those set, your metrics never leave hardware you control. |
| One dashboard without a subscription is possible | Parent-child streaming through stream.conf collects every machine onto one local dashboard. Some recent multi-node views still ask for a Cloud sign-in, so test your version before you decide. |
| Time is the real variable | Installing takes one command, while Docker installs need host mounts. The ongoing effort is mostly tuning alerts under health.d/ and the occasional evening of troubleshooting after an upgrade. |
Top comments (0)