How to Run Ollama on a CPU-Only VPS as a Private LLM API
Run Ollama on a CPU-only Ubuntu 24.04 VPS, size the model to the RAM, cap concurrency, and publish the API only behind HTTPS and a password.
85 articles
Run Ollama on a CPU-only Ubuntu 24.04 VPS, size the model to the RAM, cap concurrency, and publish the API only behind HTTPS and a password.
Build an encrypted WireGuard tunnel between two Ubuntu 24.04 VPS servers, lock the firewall to one UDP port, enable it at boot, and diagnose a handshake that will not complete.
Set up a locked-down Postfix relay on Ubuntu 24.04 that accepts mail only from your own applications, authenticates to your mail provider over TLS, and logs every message.
Measure the real memory per Apache process and PHP-FPM worker, split the budget between them, keep MaxRequestWorkers and pm.max_children in step, and watch the queue under load.
Back up a WordPress database and its files as encrypted restic snapshots to S3-compatible storage, schedule them with a weekly prune, and prove the backup by restoring it.
Read the memory error, check the limit the web request actually gets, raise it for the right pool, and find the plugin or task that uses the memory.
Read the Nginx and PHP-FPM logs to find the slow request, fix the cause first, and raise timeouts only for the paths that legitimately need them.
Read the client and server logs to find the real cause of a publickey refusal: the wrong key, the wrong account, or file permissions that OpenSSH rejects.
Install Docker from its official repository, understand why published ports can bypass UFW, bind services to localhost, and prove the result with an external scan.