Set Up Your LLM Assistant: Claude Code, Cowork, Terminex, and DeepSeek

In plain terms: Many articles on this site end with "hand this to your LLM." This is the guide that explains what that means: an AI assistant that can actually do things on your computer — run commands, edit files, install software — instead of just chatting. It walks through four ways to get one, from polished paid options to a nearly-free one, so you can pick what fits.

Want to implement this yourself? You don't have to follow this write-up by hand — an LLM assistant (Claude, ChatGPT, or a local model) can build the same setup on your machine. Handle the few things only you can do, then paste the prompt below and let the assistant do the rest.

Your part — only you can do these

A computer you're allowed to install software on (Linux, Windows, or macOS)
An account with a provider: an Anthropic/Claude account (subscription or API key) for Claude Code and Cowork, or a DeepSeek platform account (pay-as-you-go API key) for the cheap route
For the fully-local option: a machine with enough RAM/GPU to run a small model, plus LM Studio or Ollama installed

The assistant's part — delegate these

Bootstrap everything else once installed: install tools, clone projects, adapt configs to your system
Take any other article on this site, follow its steps on your machine, and fix whatever breaks until it works
Explain its own output — ask it what a command does before you approve it

Copy this prompt to start:

Fetch and read https://www.laserlloyd.com/projects/set-up-your-llm-assistant/ — it contains the full write-up, settings, and gotchas for this project. You're newly installed as my agentic assistant — get me to the point where I understand what you can do on this system and have watched one safe task succeed. Check which tools are available (git, python, node) and note any gaps. Verify with a small supervised demo: create a folder, write a test script, run it, and show the output, explaining each step first. Then summarize your capabilities and suggest next steps.

Every “Want to implement this yourself?” box on this site assumes you have an LLM assistant that can actually touch your computer. This page is the missing step zero.

tl;dr

What it is: a one-page setup guide for an agentic LLM assistant — the kind that runs commands and edits files, not just chats.
What it costs: anywhere from a paid Claude plan down to fractions of a cent per task (DeepSeek), or $0 fully local.
What you need: a computer you can install software on, and one account (Anthropic or DeepSeek) — or neither, for the local route.
What you end up with: an assistant you can hand any article on this site to and say "do this on my machine."

What “agentic” means, and why articles here say “hand this to your LLM”

A chatbot answers questions. An agentic assistant takes an outcome — “install this tool and make it start on boot” — and works toward it on your actual computer: running commands, reading and writing files, checking whether each step worked, and fixing what didn’t. You stay in the loop by approving actions; it does the typing.

That’s why the project write-ups here don’t try to be exhaustive manuals for every OS and distro. Each article explains what was built and what will bite you; the box under the intro splits the work into the parts only you can do (accounts, hardware, passwords) and the parts your assistant should do (download, install, adapt, test, revise). If a step doesn’t match your system exactly, that’s fine — adapting is precisely what these tools are good at. There’s a whole article on that pattern: How to have an LLM adapt any project to your system.

Below are four ways to get such an assistant, roughly from “most polished” to “cheapest.”

Option 1: Claude Code (Linux, Windows, macOS — the terminal standard)

Claude Code is Anthropic’s agentic CLI and the tool that built most of what you see on this site, including the site itself. You type goals in plain English into a terminal; it plans, runs commands, and edits files, asking permission as it goes.

Install — the official setup docs are at code.claude.com/docs/en/setup; the native installer needs no prerequisites:

# Linux, macOS, or Windows WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell (native, no WSL required)
irm https://claude.ai/install.ps1 | iex

There’s also an npm package (npm install -g @anthropic-ai/claude-code) if you already live in Node and want version pinning, and Homebrew on macOS/Linux. Then run claude in any folder and it walks you through signing in.

What it needs: an Anthropic account — either a paid Claude subscription (Pro/Max, usage included) or a developer API key billed per token; the setup docs cover both login paths. No account, no Claude Code — if that’s a dealbreaker, skip to options 3 and 4.

Option 2: Claude Cowork (the no-terminal option)

If a terminal isn’t your happy place, Claude Cowork is Anthropic’s agentic mode inside the Claude Desktop app: you describe an outcome — “organize these folders,” “turn these notes into a formatted document” — step away, and come back to finished work. Same agentic loop, no command line.

Verified as of this writing: per Anthropic’s help center, Cowork is available in Claude Desktop for both Windows and macOS, with web and mobile versions rolling out in beta. It requires a paid plan (Pro, Max, Team, or Enterprise). Get the desktop app at claude.com/download, open Claude, and pick “Cowork” in the message box.

Cowork is the right answer for document, file, and research work. For the maker projects on this site — installing services, wiring configs, debugging — a terminal agent (option 1 or 3) is still the better fit.

Option 3: Terminex — the Reasonix terminal agent

“Terminex” is the nickname my agentic terminal goes by around here; the tool inside it is Reasonix, an open-source (MIT) Claude-Code-style coding agent built natively around DeepSeek’s cheap API. Same experience as Claude Code — plain-English goals, supervised commands and file edits — with no subscription, just a pay-as-you-go DeepSeek key. I wrote up the full setup in Reasonix: a Claude-Code-style coding agent on DeepSeek.

Windows? Yes — verified: Reasonix ships as a static binary for macOS, Linux, Windows, and WSL (amd64 and arm64), and the npm package pulls the right one:

npm i -g reasonix

(Node.js 20.10+ required; on Windows also Git for Windows.) Source and docs: github.com/esengine/DeepSeek-Reasonix. It defaults to DeepSeek’s cheap flash tier and lets you escalate to the pro tier for hard tasks.

Option 4: DeepSeek as the cheap (or free) brain

DeepSeek isn’t an assistant itself — it’s the model behind one. Three ways to use it:

Direct API key. Create a pay-as-you-go key at the DeepSeek platform. Reasonix (above) uses it natively; so does almost anything that speaks OpenAI-style APIs. Costs run fractions of a cent per message.
The Anthropic-compatibility trick. DeepSeek serves an Anthropic-compatible endpoint at api.deepseek.com/anthropic, so Claude-Code-style CLIs can be pointed at DeepSeek with nothing but environment variables (ANTHROPIC_BASE_URL plus your DeepSeek key) — a stock install, a cheaper brain. The full wiring patterns, including how to keep the key out of your logs, are in DeepSeek Everywhere.
Fully local, $0. Run a DeepSeek distill on your own machine with LM Studio or Ollama — no account, no key, nothing leaves your computer. The beginner walkthrough is DeepSeek: running locally — a 4-step guide. Honest caveat: small local models are noticeably weaker as agents than the cloud options — great for chat and drafting, frustrating for long multi-step installs.

Which one should I pick?

You are…	Pick	Cost	OS
Comfortable-ish with a terminal, want the strongest agent	Claude Code	Claude subscription or API pay-per-token	Linux, macOS, Windows (native or WSL)
Allergic to terminals; document/file/research work	Claude Cowork	Paid Claude plan (Pro and up)	Windows, macOS (web/mobile in beta)
Terminal-fine but subscription-averse	Terminex (Reasonix)	DeepSeek pay-as-you-go (pennies)	Linux, macOS, Windows, WSL
Privacy-first / zero budget	Local DeepSeek (LM Studio / Ollama)	$0	Anywhere the model fits

Can’t decide? Claude Code if you’ll pay for the best, Reasonix if you won’t.

Gotchas

Agentic ≠ autonomous. Every tool here asks before running commands — keep it that way while you’re learning. Read what it wants to run, and ask it to explain anything unclear.
Don’t paste secrets into prompts. API keys go into the config file or environment variable the tool tells you about, not into the chat.
Windows has two flavors. Claude Code and Reasonix both run natively on Windows and under WSL. Pick one and stay there — a tool installed in WSL can’t cleanly see your native toolchain, and vice versa.
Subscription vs API key are different logins. With Claude Code, a Claude Pro/Max subscription and a developer API key are separate billing worlds; signing in with the wrong one is the classic “why am I being charged twice” mistake.
Local models will disappoint as agents. If a local model loops or stalls on a multi-step task, that’s expected — hand that task to a cloud option and keep local for private or simple work.