Claude · Setup

OmniRoute, claude-mem, Headroom, Claude Code Setup: the 4 plugins to install in Claude Code

No more stalling mid-task, a memory that carries over between sessions, fewer tokens burned and a cleaned-up setup. Commands checked on November 9, 2026, the order to follow and a bonus.

QQuentin Megevand
November 9, 2026 · 7 min read

Four free, open source plugins for four problems you run into every week in Claude Code: the usage limit that kills your session, the project you re-explain every morning, tokens burning away, and a setup that keeps getting heavier. Installed in the right order, they turn Claude Code into a tool that lasts the whole day.

For each one, this guide gives you the install command, the setting that matters, and how to check it actually works. Everything was checked on November 9, 2026 against the official repositories and Claude Code 2.1.319.

Claude AI Lab

The Claude AI Lab is my Skool community where I share my Claude systems and the more advanced modules. Access is $67/month.

Join the Lab →
4
free, open source plugins, plus a bonus
115,000+
GitHub stars for claude-mem on November 9
2.1.319
Claude Code version checked
✅What you need
1
Claude Code. Check your version with claude --version.
2
Node.js. Version 22.22 or newer for OmniRoute (or 24 and up), at least 20.19 for claude-mem. Check with node --version.
3
Python 3.11 or newer. For Headroom. Check with python3 --version.

How plugins install

OmniRoute and Headroom install from a terminal and sit between Claude Code and the model. claude-mem, Claude Code Setup and the bonus are real Claude Code plugins: in a session, type /plugin, the plugin manager opens, you add a marketplace (a GitHub repository) and install from the list. In a terminal, claude plugin list shows what is actually installed. One simple rule: install one plugin, use it for a day, then move to the next. Install four at once and you'll never know which one broke your setup.

1

OmniRoute: never run out of usage again

🔗 github.com/cheaper-inference/OmniRoute
⚙️Setup
4
Install it globally. npm install -g omniroute.
5
Wire Claude Code to it. omniroute setup-claude. If you'd rather be guided step by step, omniroute setup runs the first-time wizard.
6
Connect at least one provider. Your subscription, an API key, or a free account with a provider. The dashboard lives at localhost:20128.
7
Pick the auto model. When you hit your limit, OmniRoute switches on its own: your subscription first, then your keys, then cheap providers, then free ones.
What to know
OmniRoute is a local gateway (MIT license, over 85,000 stars) that connects Claude Code to more than 370 providers, over 150 of them with a free tier. Free tokens come through your own accounts with those providers: OmniRoute doesn't hand you any. If answers suddenly get bad, it has fallen back to a weak free model: limit your fallback chain to a few models you trust.
2

claude-mem: memory between sessions

🔗 github.com/thedotmack/claude-mem
⚙️Setup
8
In a terminal. npx claude-mem install. Or inside Claude Code: /plugin marketplace add thedotmack/claude-mem, then /plugin install claude-mem.
9
Keep secrets out. Anything you wrap in <private> tags is never stored. Your memory lives on your machine, in ~/.claude-mem.
10
Turn off the analytics if you want. Anonymous usage analytics are on by default: npx claude-mem telemetry disable turns them off.
✍️The first thing to do after installing: seed it on purpose
Write a memory entry for this project: what it is in one sentence, the tools it uses, the three files to read first to be useful here, the decisions already made that we are not revisiting, and the two things that have broken before. Under 200 words.
Why
Without this, the memory fills up with noise. With it, Claude starts every session knowing what you're building and what's already decided. Skip npm install -g claude-mem: it only installs the library, without the hooks that make the memory work.
3

Headroom: fewer tokens per request

🔗 github.com/headroomlabs-ai/headroom
⚙️Setup
11
Install the CLI. pip install "headroom-ai[all]".
12
Launch Claude Code through it. headroom wrap claude, every session: the proxy stops when Claude Code does. For a proxy that stays on, use headroom deploy.
13
Check the routing works. headroom doctor.
14
See what you're saving. headroom perf, or headroom dashboard to watch savings live while the proxy runs.
Know this first
Headroom (Apache 2.0 license) compresses tool outputs, logs and files before they reach the model. Its makers report about 20% fewer tokens for a coding agent. headroom wrap also installs Serena, a code navigation tool: add --code-memory none if you don't want it. If headroom perf shows no savings, compression isn't applying on your setup: an open report on the repository describes exactly this case with Claude Code.
4

Claude Code Setup: clean up your setup

🧹 Anthropic's official plugin
⚙️Setup
15
Install it. /plugin install claude-code-setup@claude-plugins-official. It's Anthropic's own plugin, and it only reads: it never edits a file.
16
Ask it for a recommendation. "Recommend automations for this project": it scans your code and suggests the best hooks, skills, MCP servers and subagents for that project.
17
Then actually clean up, with the prompt below.
✍️The cleanup prompt
Scan this project and give me two lists:
1. the skills, hooks, agents and MCP servers I have loaded that this project never needs, ranked by how much context they cost me each session, with the exact path of every file to delete;
2. what I'm missing that this project would clearly benefit from.
The move that counts
Actually delete the files in list 1. Every tool that's loaded and unused takes up your context in every session.
5

Bonus: TaskObserver, skills that improve themselves

🎁 github.com/rebelytics/one-skill-to-rule-them-all
⚙️Setup
18
Add the marketplace. /plugin marketplace add rebelytics/one-skill-to-rule-them-all.
19
Install the skill. /plugin install task-observer@one-skill-to-rule-them-all.
20
Activate it. Copy the activation instruction from the repository's references/environments.md file into your CLAUDE.md. Without it, the skill triggers too rarely.
What it does
It watches how you work, notes your corrections and preferences, then suggests improvements to your other skills. It changes nothing without your OK. Install it last, on a setup you've already cleaned, and give it a week before you judge it.
6

The order and quick fixes

🛠️ keep this handy
🔢The order, and why
21
OmniRoute first, so you don't run dry while you install and test the rest.
22
claude-mem next, so context stops evaporating.
23
Headroom, so what's left costs less.
24
Claude Code Setup, to delete what has piled up.
25
TaskObserver last, so it learns from a clean setup.
🚫
Claude Code won't start
Disable the last plugin you added with claude plugin disable <name>, restart, then turn the others back on one at a time.
🔁
The plugin is installed but nothing happens
Most need a real restart. Fully quit Claude Code and open it again.
🤖
Answers get bad
OmniRoute has fallen back to a weak free model. Narrow your fallback chain to models you trust.
🧠
Memory brings back old decisions
Once a month, ask it to list everything it holds on the project, and delete what's no longer true.
📉
Headroom saves nothing
Run headroom doctor to check the routing, and make sure you started Claude Code with headroom wrap claude.

To go further on claude-mem and Claude Code Setup, with three more free tools, read the 5 free plugins and skills for Claude Code.

Do this in the next 10 minutes

Install OmniRoute and open its dashboard at localhost:20128 to check it's running. Then install Claude Code Setup, run the cleanup prompt, and actually delete the files in list 1. The rest can wait until tomorrow, one plugin a day.

Want to go further?

And day-to-day, I post one reel a day on Instagram: @quentin_iamarketing