🎯 What This Collection Is
This section documents eight command-line tools that, taken together, give a Unix/Linux
SysAdmin a complete AI-augmented workflow — from asking questions at the prompt, to
editing code with AI assistance, to shipping that code to GitHub, to running models entirely
on your own hardware.
These are not eight random tools. They form a toolchain. Each one does a specific job,
and several of them are designed to work together. Understanding how they connect is more
useful than reading each page in isolation.
Every page in this section follows the same philosophy: complete installation steps
(no hunting for the EPEL step), real SysAdmin use cases, honest assessments of where
each tool helps and where it doesn’t, and copy-paste ready examples. No assumptions.
New to all of this? Start with sgpt — it has
the fastest time to value of any tool here. You can be asking AI questions from your
shell prompt in under ten minutes.
🗺️ The Toolchain Map
Here is how the eight tools relate to each other:
DAILY DRIVER
────────────
sgpt Natural language at the shell prompt. Ask questions, get commands,
pipe log output for analysis. Your first AI tool.
AUTOMATION LAYER
────────────────
fabric Reusable prompt patterns that behave like Unix filters. Pipe text
through patterns: summarize logs, generate incident reports, extract
action items. Build your own patterns for institutional workflows.
CODING WORKFLOW
───────────────
aider ──────► gh
AI edits your Ships aider's work to GitHub. PRs, issues, releases from
files directly. the terminal. When aider finishes a fix, gh delivers it.
Every change
is a git commit.
LOCAL ALTERNATIVE
─────────────────
ollama Run LLMs on your own hardware. No API key, no cloud, no data
leaving your network. Powers sgpt, aider, or any tool that
accepts an OpenAI-compatible endpoint.
SUPPORT TOOLS
─────────────
fzf Interactive fuzzy selection glue. Pairs with every tool above:
search logs, pick git commits, select systemd units, browse
AI output interactively. Ctrl+R history search that works.
jq Parse JSON output from gh, AI APIs, and modern CLI tools.
Extract fields, filter results, reformat data for scripts.
Essential once any tool starts producing JSON.
github-copilot AI pair programming suggestions at the CLI. Explains pipelines,
suggests shell commands, generates bash scripts inline.
SESSION LAYER
─────────────
tmux Terminal multiplexer. Sessions survive SSH disconnects. Split panes
let you run sgpt, aider, or ollama alongside your shell. Not an AI
tool — the infrastructure that makes all the others more reliable.
SEARCH LAYER
────────────
ripgrep Fast recursive search. grep that gets out of your way. The natural
extraction layer — pull the relevant 40 lines from a 50,000 line log
before handing content to sgpt or fabric for AI analysis.
📋 Suggested Learning Order
If you are starting from zero, work through these in order. Each step builds on
the previous one and adds a distinct capability to your workflow.
-
1
sgpt — Shell GPT
Fastest time to value. One pip install, one API key, and
you are asking AI questions from the prompt. No configuration complexity.
Establishes the core habit of reaching for AI at the command line.
-
2
fzf — Fuzzy Finder
Install it early because everything else gets better with
it. Shell integration takes five minutes. Ctrl+R history search alone is worth
the install. Once it is in, you will use it constantly without thinking about it.
-
3
jq — JSON Processor
Learn this before you need it. Modern tools output JSON.
gh outputs JSON. AI APIs return JSON. Once you can slice JSON from the command
line, scripting against any of these tools becomes straightforward.
-
4
fabric — AI Workflow Automation
Once you are comfortable with sgpt, fabric gives you
reusable patterns. Stop typing the same prompt every day. Build patterns for
your specific institutional workflows: log analysis, incident summaries, change
request drafts.
-
5
ollama — Local LLM Runner
When you need AI that never leaves your network. Air-gapped
servers, sensitive log analysis, cost control. Also useful for experimenting with
different models without API fees. Point sgpt or aider at it as a drop-in backend.
-
6
aider — AI Pair Programming
AI that edits your actual files. Works with shell scripts,
Python, anything in a git repo. Every change is a commit so you can see exactly
what the AI did and revert if needed. The step up from asking for code snippets
to having AI work directly in your codebase.
-
7
gh — GitHub CLI
The delivery mechanism for aider’s work. Manage PRs,
issues, Actions workflows, and releases from the terminal. Once aider is in your
workflow, gh closes the loop from AI edit to merged PR without touching a browser.
-
8
GitHub Copilot CLI
Useful if your shop already has a GitHub Enterprise or
Copilot subscription. Explains commands, suggests shell one-liners, integrates
with your existing GitHub authentication. Good complement to the others if the
licensing is already in place.
🤔 Which Tool For Which Job
Scenario-based guide — what to reach for when you have a specific task:
| I want to… |
Reach for… |
| “Explain what this command does” |
sgpt or github-copilot |
| “Search my command history for that pipeline I wrote” |
fzf (Ctrl+R) |
| “Summarize 500 lines of log output” |
sgpt or fabric |
| “Generate an incident report from these logs every day” |
fabric (custom pattern + cron) |
| “Fix a bug in my bash script” |
aider |
| “Parse JSON output from an API call” |
jq |
| “Open a PR for the fix aider just made” |
gh |
| “Run AI on a server with no internet access” |
ollama |
| “Keep sensitive logs off external API servers” |
ollama + sgpt or fabric |
| “Interactively pick which log file to analyze” |
fzf piped to sgpt |
| “Extract all PR titles from a repo as a list” |
gh + jq |
| “Write and commit new monitoring scripts” |
aider + gh |
| “Hit an AI API directly from a shell script” |
curl + jq |
| “Test whether an endpoint is responding and how fast” |
curl (-w %{http_code} %{time_total}) |