
RAG on a 2 vCPU VPS with no GPU: Qdrant stores the index, Ollama writes the answers. Scripts, timings and three real failures.

One OpenAI-style address for all your models: LiteLLM proxy on a 2 vCPU VPS with aliases, a key, a fallback and systemd. Real output, no GPU.

tar -czf, tar -xzf, zip and unzip: create, list and extract archives, skip node_modules, pick gzip, xz or zstd. Tested on Ubuntu 24.04.

Delete a folder in Linux with rmdir, rm -r or find: dry runs, old files, Permission denied, busy, full disk. Tested on Ubuntu 24.04.

HAProxy on a VPS: round robin, /health checks, weights, the stats page. We killed a backend and traffic moved to the rest. Ubuntu 24.04.

Docker says permission denied on docker.sock? Add your user to the docker group, log in again, and skip chmod 666. Why it works.

A working AI agent in about 90 lines of stdlib Python: tools, an allowlist, a step limit and a systemd timer, tested on a 2 vCPU VPS with no GPU.

A Git server with CI on your own VPS: Docker Compose, a runner, HTTPS, a GitHub move and a backup whose restore we actually tested.

How to find what takes the space: images, stopped containers, volumes and huge logs. What is safe to clean and what is not.