The setup, as the author describes it
Mishra wanted a local alternative to paid cloud storage. He also didn't want a separate USB drive for every device. His server is an HP t620 thin client, which he says he bought and upgraded for $45. It has 8GB of DDR3 memory and a 128GB M.2 SATA SSD. It runs the x86 edition of DietPi with no desktop environment, plus Kopia in Docker. He reports that the whole OS and software stack uses about 400MB of memory.
His backup targets are modest:
- Windows PC: scripts, documents and a few other folders.
- Raspberry Pi: its Docker files, kept for reinstallation or recovery.
The build went in a few steps:
- He created an encrypted Kopia repository. OpenSSL generated the strong passwords.
- He kept the credentials and the repository password in a private file with restricted access.
- He adjusted the Docker Compose file so the container ran.
- He generated a self-signed certificate with OpenSSL, added TLS to the Compose file and recreated the container. The web UI now asks for credentials on every visit.
- He connected the Windows PC using the GUI version of Kopia. He chose the Kopia server repository option, entered his credentials and supplied the certificate's SHA-256 fingerprint. The server answers over HTTPS on port 51515.
- He had to create a user for each device on the server before it could connect. He made one for the Windows PC and one for the Raspberry Pi.
The author provides no total data volume, backup times, restore test, Kopia version or Compose file. Treat the hardware and memory figures as his reported numbers.
Why deduplication saves space
Kopia's documentation says every snapshot is incremental. Kopia's Getting Started guide isn't in my search results, but the project's feature overview, which I reviewed in the source material, says content is uploaded once based on what the file contains. A file is re-uploaded only if it changes. It also says Kopia splits files with a rolling hash, so a modified large file uploads only its changed parts. Identical files are stored once, even across different computers sharing a repository.
That refines the author's description of snapshots that "reference" earlier ones. Each snapshot is still a full, restorable point-in-time view. The repository simply doesn't store the same content twice. Savings therefore depend on your data:
- Identical files across devices and unchanged content between runs compress well in storage terms.
- Genuinely new or heavily changed data still needs space.
- Skipping a snapshot when nothing changed is a separate policy choice. It isn't the same thing as deduplication.
Kopia's FAQ says deduplication is on automatically for all repositories, so there is nothing to enable. Compression is different. It is controlled by policy and is off by default, so it is a separate lever for saving more space.
Connecting clients: the fingerprint step
The fingerprint prompt Mishra hit matches Kopia's documented workflow. Kopia's repository server documentation shows a server on port 51515 and clients connecting to an HTTPS address with a server certificate fingerprint. The server can generate its own TLS certificate on first start and prints a SHA256 fingerprint. The documented client command passes that fingerprint without colon separators. The documentation also gives an OpenSSL command for getting the fingerprint of an existing certificate file.
A forum thread on the Kopia support site adds a pitfall. A user connecting to a server by IP address got an error. The cause, according to a reply there, was that the auto-generated certificate defaults to the 127.0.0.1 host name. Anyone using a self-signed certificate should make sure it covers the name or address clients will actually use. That's a forum answer, not official documentation.
A 2026 setup tutorial also lists a certificate error as a common connection failure. Its suggested fix is re-copying the fingerprint printed at server start, without colons, in case the certificate was regenerated. Treat that as a practical hint, not official guidance.
Security caveats worth knowing
Kopia's documentation draws some important boundaries.
- Separate users are not total isolation. In server mode, each user sees only their own snapshot and policy manifests. But only those manifests are access-controlled, not the underlying content. Because deduplication gives identical files the same content ID, anyone who knows a valid ID may be able to read that content. The documentation says guessing such IDs should be practically impossible. It's still a limit on how much "one account per device" protects you, especially if you share the server with other people.
- The repository password is a single point of failure. Encryption is repository-wide and on by default, and Kopia says a lost password cannot be recovered. The author keeps his password in a restricted file. If that file lives only on the server, a dead server could take the key with it. Keep a copy elsewhere, such as a password manager.
- Client commands are limited. In server mode, low-level commands such as
kopia repository statusfail on clients. The server is a proxy, not a network share.
Reliability: what the setup doesn't cover
The author suggests a secondary copy of the repository on a USB SSD or HDD in case the internal SSD fails. That helps with drive failure. A copy sitting next to the server won't help with theft, fire or a power surge, which is general backup reasoning rather than a Kopia claim. His inverter outlet keeps the server and router running through power cuts, but it doesn't change that picture.
Kopia's consistency documentation covers testing:
- A full test restore is described as the gold standard for checking a backup.
kopia snapshot verifychecks that the repository structure is consistent. Kopia's documentation says it runs automatically during daily full maintenance. It confirms blobs exist but doesn't download and decrypt them.- Adding
--verify-files-percent=100downloads and decrypts everything, which tests for bit rot. This option must be run from the command line, since KopiaUI can't do it yet. - The same page warns that networked or emulated filesystems may not behave the way Kopia expects. A plain local filesystem on the server is the safer choice.
A tip for Windows users
Kopia's ignore-rule documentation has a Windows-specific warning. Rules in a policy or a .kopiaignore file must use forward slashes. Backslashes copied from Windows paths aren't supported. For example, write Documents/cache/* rather than Documents\cache\*. Test your rules, because a bad pattern can silently leave files out of a snapshot.
Bottom line
Mishra's build shows the pattern works on very modest hardware: one low-power box, a Windows PC and a Pi, with encryption and TLS on. Kopia's documentation supports his core claims about incremental snapshots, deduplication across a shared repository and fingerprint-based connections. The article doesn't prove any particular savings ratio, and it doesn't include a restore test.
If you copy the idea, take the extra steps:
- Store the repository password somewhere independent of the server.
- Don't treat separate users as full content isolation.
- Keep an off-site or disconnected second copy.
- Run a real restore, and a full
kopia snapshot verify --verify-files-percent=100periodically, before you trust it.
References
- I backed up every device in my house to one Kopia server, and deduplication saved more space than I expected XDA · 2026-11-01T12:30:20+00:00
- Error connecting to API server - Support - Kopia Forum kopia.discourse.group
- Improve documentation how to setup and use a kopia server · Issue #1982 · kopia/kopia · GitHub github.com