MoltBot

Moltbot (formerly Clawdbot) is the AI assistant that shows up in your messages — and does the work

Open clawd.bot and it doesn’t waste your time. The pitch is blunt: “The AI that actually does things.” Clear your inbox. Send emails. Manage your calendar. Check you in for flights. And do it “from WhatsApp, Telegram, or any chat app you already use.” 

That’s the hook. Not another chat window. Not another tab. Instead, it’s an assistant that lives inside the places you already talk to humans.

And yes, the branding is a little confusing right now. The project that many people first heard about as Clawdbot is now called Moltbot. More on that in a minute. 

The simple concept: stop visiting your AI, make it come to you

Most AI tools still work like this: you open an app, type a request, copy-paste the result somewhere else, then repeat.

Moltbot flips that. You message your assistant the way you’d message a coworker. It replies in the same thread. If you’ve hooked up tools and permissions, it can take action instead of just suggesting actions. 

The project calls itself a personal AI assistant you run on your own devices, and it’s designed to answer you on a long list of channels: WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, Microsoft Teams, and more. 

That “run it yourself” part matters. A lot.

What it can do

The front page leads with the everyday stuff: inbox, email, calendar, travel check-ins. 

But the community examples show where it gets interesting: daily briefs, weekly reviews, auto-resolving calendar conflicts, prepping meeting briefs, breaking down projects into tasks, even spinning up “sub-agents” to research ideas. 

Think of it less like “chat with a bot” and more like:

• A command line you can text
• A personal assistant that can remember context
• A glue layer between your messages and the software you already use

The project’s own docs and showcase lean hard into real workflows: automation, memory, voice, deployment setups, and lots of “i asked it to do X, then it did X.” 

Under the hood: the “Gateway,” channels, and skills

Moltbot is open-source and built around a “local-first” control plane it calls the Gateway. The GitHub README describes the Gateway as the thing that manages sessions, channels, tools, and events — while the “product is the assistant.” 

In practice, you can think of the system as three layers:

1. Channels: where messages come from (WhatsApp, Telegram, Slack, iMessage, etc.). 
2. The assistant: the agent that reads messages, plans actions, and replies. 
3. Skills/tools: integrations that let it do things (work with files, browse, automate, talk to services, and so on). The community has even started curating lists of skills. 

It also supports multiple model providers. The README explicitly mentions subscriptions via OAuth for Anthropic and OpenAI, plus model configuration and failover. 

So you’re not locked into one “house AI.” You’re wiring a system where the model is a component, not the whole product.

Why it caught fire

The Verge summed up the vibe well: Moltbot is an open-source agent that people like because it feels like an assistant that’s “always-on” and programmable, and it interacts through the messaging apps people already live in. 

This is the part many AI demos skip: the boring daily friction.

If you want an AI to be useful, you can’t make the user do a bunch of ceremony:

• open app
• paste context
• re-explain preferences
• copy output
• manually perform actions

Moltbot tries to remove that ceremony. It’s right there in your chat list. And if you set it up well, it can go from “answer” to “done.”

The catch: giving an AI “hands” can bite you

An assistant that can send messages, touch files, and run automations is powerful. It’s also a bigger security story than a chatbot that only talks.

The Verge highlighted concerns like prompt-injection and the risks that come from giving an agent high levels of access, especially if it can be manipulated through messages or connected tools. 

If you’re running something like this:

• keep it single-user unless you really know what you’re doing,
• be careful with admin permissions,
• treat unknown inbound messages like you’d treat unknown links.

In other words: don’t give a bot your house keys and then invite the whole internet into your living room.

Who this is for (and who should skip it)

Good fit:

• People who already tinker with automations
• Developers and power users
• Anyone who wants a local-ish assistant they can shape

Maybe not:

• Folks who want “install app → done”
• Anyone who doesn’t want to manage keys, permissions, or deployment details
• People who prefer cloud tools precisely because they don’t want to run infrastructure

Moltbot is trying to make setup smoother (the README points to an onboarding wizard and daemon install flow), but it’s still a real system, not a toy. 

Where this is headed

Moltbot is part of a broader shift: AI products moving from “chat” to agency — models wired into channels, tools, and workflows.

That trend is already spilling into the business world. One example: MarketWatch noted investor attention around the infrastructure and security implications of viral agent-style tools, even when the connection is indirect. 

The big question isn’t “can an AI write my email.” It’s “can an AI run my day without making things worse.”

Moltbot is an early, very real attempt at that. It’s scrappy. It’s a bit chaotic. But it’s also one of the clearest examples of what “personal AI” looks like when it stops being a website and starts being a coworker you can text