aigent-OS · open-source · live now
The best agent OS. Free.
Stock Claude makes you carry the bag: tools, prompts, agents, context. aigent-OS gives Claude a caddy that hands it the right tool at the right moment, a crew of named specialists — each one an AI agent with its own job — and a memory it keeps. The paid platforms charge for less. This one is open-source, and it is live now.
01Get the repoClone aigent-OS from GitHub — it is live now.
02Drop it into ClaudeOne command installs the vault, skills, and crew.
03Say "set this up"The setup interview wires Claude into your operator.
Free, and staying free. Starring is how the next builder finds it. Star the repo
Free and open-source. Clone it, read every file, run it — the same system that ships The AIgent every morning.
Every repeated mistake becomes a guardrail.
A pantheon of named specialists you own and carry.
The enhancement
Same Claude. Enhanced output.
Think of it as a legal, packaged enhancement layer that turns a stock session into a high-performance operator. You keep the model you already pay for and add the system around it.
A caddy that surfaces the right one of about sixty skills at the exact moment it matters, so you are not the one remembering which tool to reach for. A learning loop that logs every repeated mistake into a durable guardrail, so the same wrong turn is far less likely to bite twice and the system gets sharper the more you run it. Body-state awareness that warns you when a session is running low on working memory before it costs you. A vault of notes and rules you commit to git like code, and a nine-specialist crew you own and carry across machines. Same Claude. Enhanced output. One install.
Is this for you
Built for builders who live in the terminal.
For you if
You run Claude Code — whether today is day one or day five hundred — and you want it operating like a system that remembers, not a blank assistant you re-brief every session.
Not for you if
You use Claude occasionally, or you want one single-purpose bot. This is the full operator system, not a one-trick agent.
Proof it runs
The proof is that it runs a real business.
aigent-OS is the system that produces The AIgent newsletter every single day. Not a demo, not a template, a live operation. Here is one run.
›Task in. Draft and ship today's issue to the list.
›Caddy surfaces /aigent-daily the moment the task lands.
›The crew runs: research pulls the drops, a writer drafts, a critic stress-tests the claims.
›A guardrail catches an unverified stat before it ships and sends it back for a source.
✓Shipped to 9,000+ inboxes. The miss is written back as a rule so it does not happen twice.
9,000+ readers every morning · a new issue shipped every single day · read the latest →
The three moats
What the run is actually showing.
Auto-Refresh. Context management is automatic. Every turn the session checkpoints itself to a capsule, and when the window clears or compacts it resumes exactly where it was, memory intact. A closed terminal, a compaction, a /clear: it comes back knowing your project and your place in it. The only thing you ever type is /clear itself.
The caddy. You do not have to know which of about sixty skills fits the moment. It reads your task and surfaces the right one, right when it matters. You stop being the one who remembers the toolbox.
A filed Claude Code issue asked for exactly this: "the memory should function as a pre-action checklist, not just passive context." (anthropics/claude-code #37847)
The learning loop. Every repeated mistake becomes a guardrail. The system diagnoses the failure, writes the fix into a memory, a rule, a check, or a caddy trigger, and surfaces it before you hit the same wall again. The same wrong turn is far less likely to cost you twice, and it gets sharper the more you run it. It also keeps score on itself: honesty and trust ledgers measure how well its own calls hold up over time.
What you get
The operator system you'd otherwise wire by hand.
The caddy
the headline enhancement
Surfaces the right one of about 60 skills the moment your task needs it. You stop being the one who remembers the toolbox.
The learning loop
compounds as you run it
Every repeated mistake becomes a durable guardrail, so the same wrong turn is far less likely to cost you twice and the system gets sharper the more you run it.
16-document kernel
days of architecture
Sixteen documents that teach your agent to reason, delegate, and manage its own memory — like an operator, not a one-off chat.
9 specialist sub-agents
a named pantheon
Hand real work to a ready-made crew — one builds, one designs, one researches, one stress-tests — instead of babysitting a single generalist thread. Standard Claude sub-agents start blank; these arrive configured.
About 60 slash-command skills
weeks of curation
Your toolbelt: sixty ready-made abilities, one command each, surfaced by the caddy the moment a task needs them.
Committable vault + body-state awareness
rarely gets built
A human-readable vault you own and carry across machines, and a warning when a session is running low on working memory — before it costs you.
One-command install
the setup you skip
Live in minutes. Markdown and shell, no build step, no server, no database.
Branded desktop launcher
the terminal is optional
Install once, then double-click the AIgent icon: it opens warm-resumed, right where you left off. No cold start, no directory hunting.
The core is genuinely ours: the caddy, the memory architecture, the agent roster. Bundled third-party skills are curated open-source, each credited with its license in the included credits file. Read every file, then run it.
Open source
Read every file. Then run it.
The full aigent-OS is live and public: the caddy that surfaces the right skill in real time, the learning loop that writes every miss back as a rule, the committable vault you own across machines, and the routed crew. Markdown + shell — no build step, no database, no server. Clone it and run one command:
$ git clone https://github.com/wrg32786/aigent-os && cd aigent-os && bash install.sh
Prefer chat? Open the cloned folder in Claude Code and say "install this": Claude runs the installer for you, then "set this up" starts the operator interview. It installs into whatever directory you already work in — your project, your home folder — and the installer wires Claude Code settings with your real paths. Prefer the lighter rung first? The starter operator-kit is also free: a five-agent crew and an auto memory-loader so you stop re-explaining your project every session.
The difference
Stock Claude codes. Enhanced Claude operates.
Most kits ship skills and agents. The enhancement is what surrounds them: a caddy that surfaces the right rule and skill the moment it matters, and a learning loop that turns your mistakes into guardrails.
That is the difference between a folder of tools and an operating system. The paid platforms guard it behind a login and a bill. This one is open, and it is yours.
Questions
The honest answers.
Is it really free?
Yes. aigent-OS is open-source and live on GitHub. Clone it, read every file, run it. No license fee, no seat, no subscription.
Where do I get it?
github.com/wrg32786/aigent-os — the full OS is public now. Clone the repo, run bash install.sh, and it installs into whatever directory you already work in.
Why is it better than the paid platforms?
Most paid platforms give you a chat with tools bolted on. aigent-OS gives Claude a caddy that surfaces the right skill at the right moment, a learning loop that turns mistakes into guardrails, a ready-made crew of specialist agents, and a memory vault you own across machines. Same idea, open, and yours.
What does the install actually run?
Markdown and shell, nothing hidden. You can read the install script before you run anything. The whole OS is plain files: read every file, then run it.
Do I need Claude Code?
Yes. aigent-OS installs into Claude Code; it is the enhancement layer on top, not a replacement. Never used Claude Code? Installing it takes a few minutes — day one is a fine day to start.
Can it change itself?
It proposes with /dream and can implement on a branch with /meta-improve. Merges are always yours: the self-improvement loop is human-gated by design.
Does it know when it is wrong?
It keeps score. Honesty and trust ledgers track how the system's own calls hold up over time, and repeated misses become guardrails. Most setups assume their AI is calibrated; this one measures it.
Will it work with my stack?
Yes. aigent-OS is markdown and shell that installs into Claude Code, so it is language and framework agnostic. The vault learns your project and stack as you use it; it does not assume one.
The best agent OS. Free.
Open-source, one command to install, and it runs a real business every morning. The repo is live — clone it now.
Free forever. MIT-licensed. Every file readable before you run it.
Built by Will Gwyn. I ship The AIgent to 9,000+ readers every morning with this exact system.