Mallow: A Personal AI Assistant That Lives in Telegram — No API Keys, No Mac Mini
Mallow is a personal AI assistant that runs in the cloud on Hermes, lives in Telegram, and connects your apps through Composio. No API keys, no home server, nothing to install.
Dzianis Vashchuk
4 min read
Every personal AI assistant demo ends the same way: with a weekend of setup. Get an API key. Find a spare Mac mini or rent a VPS. Install the runtime, fight the config files, wire up OAuth for every app you use. By Sunday night you have a very capable assistant and a small infrastructure hobby you never asked for.
Mallow is the opposite trade. You sign up, open Telegram, and start talking. The assistant is already running — in our cloud, on the Hermes agent runtime — with models included, a browser built in, and your apps connected. No keys of your own. No hardware on your desk. Nothing to babysit.
Try it: mallow.agentlabs.cc — Mallow is in early preview.
It lives where you already talk
Mallow's home is Telegram. Not another app to check, not a dashboard to log into — a chat thread, next to your friends and group chats. Ask it to plan your day over coffee, forward it a link to summarize, tell it to draft the reply you've been avoiding. It answers in the same thread, in your language.
The same assistant is also in the Mallow web app, and messaging is a choice, not a lock-in: Telegram works out of the box, and Slack and WhatsApp plug into the same Connections screen — you pick where Mallow replies to you.
No API keys, no Mac mini, no YAML
The usual self-hosted assistant stack quietly assumes you are your own DevOps department. Mallow removes that entire layer:
- Models are included. No OpenAI or Anthropic key, no billing account to top up, no quota surprises.
- Compute is included. Your assistant runs in its own Linux container in our cloud, on 24/7 — close your laptop, it keeps working.
- Setup is a conversation. If Mallow needs access to something, it walks you through connecting it. There is no config file in your life anymore.
A real computer, not a chatbot
Under the hood, Mallow runs on Hermes (the Hermes Agent runtime by Nous Research) inside a Linux container with full rights. That one fact changes what "assistant" means:
- It installs what it needs. Missing a tool for the job? Mallow installs software, runs commands, and manages files — like
yt-dlpfor a video you sent, or a script for the spreadsheet you're staring at. - It has a browser. A built-in browser lets it research, compare, and read pages for you. And with cookie sync from your desktop Chrome, it can work inside the sites you're already logged into — the way a human assistant with your laptop would.
- It learns skills. Anything Mallow figures out once becomes a reusable skill. Ask for a stock breakdown in the style of a famous investor, a weekly report in your format, a checklist for your trips — it keeps the recipe, not just the answer.
- It remembers. Persistent memory means context survives between chats: your people, your projects, your preferences, the thing you said last Tuesday.
Your apps, connected in a couple of taps
An assistant that can't touch your inbox or calendar is a very articulate stranger. Mallow connects the accounts you choose through Composio, the integration layer behind the Connections screen: Gmail, Google Calendar, Notion, and 1,000+ other popular services.
You connect only what you want it to use, and you can disconnect just as easily. Inbox triage, draft replies you approve before they send, calendar wrangling, research written back into your notes — that's the daily busywork Mallow exists to absorb.
Who Mallow is for
- People whose to-do list lives across chat, email, and calendar — and who would rather talk to one assistant than administer five apps.
- Builders and founders who want an agent with real tools (shell, browser, files, skills) without running the infrastructure themselves.
- Anyone who tried the self-hosted route, got it working, and then quietly stopped maintaining it.
Early preview, honest edges
Mallow is new, and we'd rather say so. It's in early preview: the core loop — chat, tools, browser, memory, connected apps — works today, and it improves weekly. If you hit an edge, a human (and a very motivated agent) reads the feedback.
Start here: mallow.agentlabs.cc — open Telegram, say hello, and give it the task you've been putting off.
FAQ
Do I need my own API keys? No. Models and compute are included with Mallow. There is nothing to buy from OpenAI, Anthropic, or anyone else, and no key to paste anywhere.
Do I need a Mac mini or a server at home? No. Mallow runs in its own cloud Linux container. Your laptop can be closed, off, or on a beach with you — the assistant keeps running.
Which messengers does Mallow work with? Telegram out of the box, plus a web app. Slack and WhatsApp connect through the same Connections screen, so you can choose where Mallow replies to you.
What can Mallow actually do? Research with sources, inbox and calendar work through your connected apps, drafting and writing, file and data tasks, recurring chores on a schedule, and anything it can do with a shell, a browser, and installable tools — which is a longer list every week, because it learns skills.
How is this different from a chatbot app? A chatbot talks. Mallow acts: it has its own computer, a browser with your sessions, your connected apps, memory, and skills — and it does the task, then reports back.
Is my assistant separate from other people's? Yes. Every Mallow runs in its own isolated container with its own memory, files, and connections.