OpenClaw alternative: a local AI team with guardrails

OpenClaw won the argument. People want AI that lives on their own computer, works through the apps they already use, and belongs to them rather than to a cloud dashboard. The project's GitHub page shows about 393,000 stars as of October 21, 2026. That makes it one of the biggest open-source projects anywhere.

So this isn't a "why OpenClaw is bad" article. It's about what comes next. If you love the local idea but the 2026 security headlines made you nervous, or you're tired of running your assistant out of config files and chat threads, here's what to look for in an alternative, and where Fluf fits.

Why local won

Three reasons, and none of them are hype.

Ownership. Your files, your memory, your machine. When the assistant runs on your computer, your context isn't sitting in someone else's database waiting for a policy change.

Reach. A local assistant can touch what a cloud tool can't: your folders, your terminal, your desktop apps, your signed-in browser. That's where real work happens.

Cost control. Local runners let you choose your own model and pay your provider directly. No mystery credit meter sitting between you and the model.

OpenClaw packaged all three into something people could actually install, and the market answered.

What OpenClaw does well

Credit where it's due. OpenClaw is free, MIT-licensed open source, maintained by the OpenClaw Foundation, a non-profit. It runs on your own machine and talks to you through the chat apps you already have: WhatsApp, Telegram, iMessage, Slack, Discord, Signal, Teams and more than 20 others, according to its README. Native apps cover macOS, Windows, Linux, iOS and Android.

Its personality and memory live in plain files you can read and edit. If you're a developer who likes owning every moving part, that's a feature, not a bug. And the community around it is huge, which means plugins, guides and answers for almost everything.

If you want a free, hackable assistant you reach from your phone's chat apps, OpenClaw is a strong pick. We'll come back to who should stay on it.

The 2026 reality check

Being the most popular local assistant also made OpenClaw the most attacked one. These are the incidents that made headlines, with sources.

One-click takeover. CVE-2026-25253 was a vulnerability where a single malicious link could give an attacker remote control of the machine running OpenClaw. It was fixed in release 2026.1.29, reported by The Hacker News. If you updated, you're covered for that one. The lesson stands: an assistant with access to your computer is a high-value target.

Poisoned skills. In February 2026, The Hacker News reported that researchers found 341 malicious skills on ClawHub, OpenClaw's community skill marketplace, with the large majority installing a password-stealing malware known as AMOS. Open marketplaces are great for speed and terrible for trust when anyone can publish code that runs with your assistant's permissions.

The subscription rug pull. On April 4, 2026, Anthropic's Boris Cherny posted: "Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw." Anthropic cited "outsized strain" on its infrastructure and offered affected users a one-time credit or a refund, per Cybersecurity News. OpenClaw users on Claude plans had to switch to pay-as-you-go extra usage or an API key.

None of this means the people building OpenClaw did a bad job. Fast-moving open source at that scale gets hit hard, and the team shipped fixes. But it does show which questions to ask before you hand any local runner the keys to your computer.

What to demand from any local runner

Use this checklist whether you pick Fluf, OpenClaw or anything else.

1. Approvals before risky actions. Spending money, sending messages, deleting files, posting publicly: the assistant should stop and ask. By default, not as an opt-in buried in settings.
2. Scoped access. You decide which folders, apps and accounts a bot can touch. "Full access" should be a deliberate choice, never the starting point.
3. No skill-marketplace roulette. Skills that run code on your machine need provenance. Prefer skills the bot learns from your own work, or ones from a vetted source, over a public bazaar.
4. Official sign-in flows. Anthropic's published terms say third-party apps may not offer Claude.ai login or route requests through your plan's credentials, and may not collect or store your Claude session tokens. What the terms do allow is you signing in to the official, unmodified Claude Code app with your own subscription. A tool that does it the allowed way won't break the next time a provider tightens enforcement.
5. Visibility. You should be able to see what every bot is doing right now, what it did overnight, and why. If bots talk to each other, you should be able to read the conversation.
6. Updates that don't break your setup. Your assistant is infrastructure now. Treat stability as a feature.

How Fluf approaches local

Fluf is a desktop app for Mac and Windows (Linux is in release testing) that runs a team of AI bots on your own computer. Every bot is a Fluf: a small, fluffy 3D character with its own name, personality and habits, so you can see at a glance who's working on what.

Here's how it maps to the checklist.

Local by design. In v1, bots run only on your computer. Fluf's account service handles your account, licence, billing and downloads, nothing else. Your bots, their memory and your files stay on your machine. The only thing that leaves is the context a task needs, sent to the model you picked.

Asks before risky actions. When a bot is about to spend, send, delete or publish, it pauses and lands in your Inbox under "Needs you". You approve, change course or say no.

A real team, not a pile of chats. Bots sit in an org chart: workspace, business, department, team, bot. Manager bots delegate and review. A main bot runs each business. One chief-of-staff bot oversees everything. You can read every bot-to-bot conversation.

Your AI apps, the allowed way. Fluf runs your installed Codex or Anthropic's Claude Code, each signed in with your own login through the provider's own sign-in. Fluf never sees or stores those credentials. You can also add API keys for Claude, OpenAI, Gemini, Grok, OpenRouter and more. Connect several accounts per provider, rename them, and see remaining usage and tokens per bot. Fluf isn't affiliated with Anthropic, OpenAI or any model provider.

Memory that learns. Each bot remembers what matters, learns skills from the work it does, and knows its role and where it sits in the team. You can open its memory in the Library and edit it.

Computer use, no GPU needed. On Mac and Windows, bots can click through your apps when there's no API for the job, inside the access you grant.

Always visible. Home shows your team alive, with each Fluf reacting to what it's doing. When a task finishes, you get an in-app and system notification.

The honest trade-offs: in v1, your bots work only while your computer is on and awake. Cloud is coming very soon: your bots will run in the cloud with their own computer. Fluf isn't open source, and it's not a gateway for every chat app: each bot can have its own Telegram contact, but there's no WhatsApp or Slack. And OpenClaw is free, while Fluf Pro costs money (more on that below).

Side by side

Who should stay on OpenClaw

Stay if you want something free and open source you can audit and fork. Stay if your main way of working is messaging your assistant from your phone through WhatsApp or Telegram. Stay if you enjoy owning the config and you keep up with security releases. That's a real and good fit, and OpenClaw serves it better than we do.

Choose Fluf if you want a polished desktop app instead of files and terminals, a team of bots with a real structure instead of one assistant, approvals by default, and your AI apps and keys connected the way providers allow. And if you like seeing your team actually working, that's the part we obsess over.

Try it

Fluf is free for 2 businesses and 6 bots. Pro is $19 a month or $139 a year, with a 14-day trial and no card required. Download it for Mac or Windows, connect an API key or your installed Codex or Claude Code, and meet Pom, who will introduce you to the rest of the team.

Download Fluf

Read next: How to build an AI org chart · Run AI employees on your Claude subscription