Skip to content
Merged
Binary file added public/blogs/2026-09-15/02-computer.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added public/blogs/2026-09-15/04-files.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added public/blogs/2026-09-15/05-schedule.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
File renamed without changes.
165 changes: 165 additions & 0 deletions src/content/blogs/en/2026-09-15.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,165 @@
---
title: Introducing Memoh: A Cloud Computer for Every Agent
author: Team Memoh
---

# Introducing Memoh: A Cloud Computer for Every Agent

Introducing Memoh — a cloud computer for your agent. Every agent gets its own desktop, files, browser, network, and long-term memory, running 24/7. Your laptop closes. It doesn't.

Models are already smart enough. What they lack is a place to work.

An agent in a chat window can think, then it stops. Close the tab and the context is gone. Close the laptop and the bot goes to sleep with it. If you wanted it on duty, to remember last week's files, to open a real page and click through it, the usual path was renting a server, configuring an OS, managing keys, and writing scripts. That work stayed locked behind people who already knew how to run infrastructure.

Memoh gives the computer to the agent. It is open source. You can self-host it, or use the hosted service.

## Why an agent needs a cloud computer

A cloud computer gives you two things a laptop can't.

**It never turns off.** Laptops sleep, drop Wi-Fi, and shut their lids. A cloud computer doesn't. While you're on a plane, asleep, or away for the weekend, the bot on Telegram, Lark, or Discord keeps answering. Scheduled jobs still run. When something needs a decision, it reaches out first.

**Files, tools, and memory stay put.** Ordinary chat starts from a blank slate every time. It can't see what you did yesterday. A Memoh workspace behaves more like your own machine: files you wrote are still there, runtimes you installed stay installed, and long-term memory carries across sessions and channels. What you set up on Monday is still there on Friday.

Together, those two things change the kind of work you can give an agent. You stop running one-off tasks and start keeping an always-on coworker.

It also differs from a throwaway sandbox: the workspace has a graphical desktop, so you can watch the agent work and take over when it stalls; and you can put ACP agents you already use — Claude Code, Codex, Hermes — on the same cloud computer.

![Each bot has an isolated container workspace: filesystem, processes, network, and resources live on its own computer](/blogs/2026-09-15/02-computer.png)

_The workspace is the agent's computer. Files, commands, desktop, and network all follow that bot._

## When to use this computer

Memoh is not a single shape. Desktop, self-hosting, and Memoh Cloud all do the same thing: give the agent a real computer. The difference is where that computer runs.

| Scenario | The right fit |
| --- | --- |
| Trying it out, keeping data on your own machine | [Memoh Desktop](https://memoh.ai/download) |
| Stay online with the laptop closed, share with a team, keep channels on 24/7 | Self-hosted server, or [Memoh Cloud](https://app.memoh.net) |
| One-off Q&A, no environment needed | Regular chat is enough |
| Run Claude Code / Codex, write code, preview the app | The agent's cloud computer |
| Morning briefings, release watches, on-duty bots | The agent's cloud computer |
| Open a real page, sign into an admin console, take over if needed | Workspace desktop |

Desktop is not replaced by Cloud. Use Desktop when the work lives in your local files and apps. Use self-hosting or Memoh Cloud when the work needs a machine that never closes.

:::tip
If you don't want to operate infrastructure, open [app.memoh.net](https://app.memoh.net) and sign in with GitHub or Google. If you want the runtime fully on your side, [download Desktop](https://memoh.ai/download) or follow the [self-hosted docs](https://docs.memoh.ai/self-hosted/).
:::

## What you can run on it

Once the setup wall is gone, the question is which job you want it to own for a long time. A few that actually work today:

### 1. Host a bot that stays on 24/7

A bot is useful when it doesn't drop. Support, on-call, lead triage, community ops — the hard part has always been a machine that never sleeps, reliable channel connections, and sessions that don't vanish when you close your laptop.

Memoh attaches Telegram, Discord, Lark, WeChat, Slack, and email to the same agent. In a group it does not have to answer every line — [Discuss mode](/blogs/2026-05-02) already lets it decide whether to speak. What you briefed in the web UI is still there in Telegram.

Sample prompt:

> Connect Telegram and make this our on-call assistant. Every morning at 8, summarize unread messages and repo activity into the group. Keep it online after I close my laptop.

### 2. Bring your own coding agent

You don't have to throw away tools you already like. Through ACP, put Claude Code, Codex, or Hermes inside the bot's container workspace, using your own subscription or API key — or models provided by Memoh.

Coding, installing dependencies, running commands, and landing changes all happen on this cloud computer. You dispatch the work from the web UI, Desktop, or any connected channel.

Sample prompt:

> Put my Claude Code on this bot's cloud computer and sign in with my own subscription. After that I should be able to ask it to change code from the web UI and Telegram, and the workspace files should still be there.

### 3. When the work is on a screen, open the workspace desktop

Login pages, admin consoles, OAuth, file pickers — that work is not in an API. A Memoh workspace can light up a screen: the agent clicks on it, you can watch, and if it stalls you take over.

How that desktop reaches a headed browser, when to use Browser Use, and when to fall back to pixel-level GUI control is in [the May workspace-desktop post](/blogs/2026-05-15). After this launch, that is not an add-on. It ships with the cloud computer. Close the laptop; the screen stays in the cloud.

Sample prompt:

> Open our admin console and walk through this feature. If the login page asks for verification, stop and wait for me to take over the desktop. I'll finish that, then you continue.

### 4. Keep files and knowledge on the same disk

Chat is no longer a disposable sandbox. Reports, scripts, screenshots, spreadsheets, last week's drafts — they stay under `/data`. The agent can search them, read them, and keep editing. Long-term memory follows the workspace too: searchable across sessions and channels by default, with Mem0 or OpenViking if you want them.

![Workspace files persist: spreadsheets, screenshots, scripts, and memory files live on the same computer](/blogs/2026-09-15/04-files.png)

_Files written today are still there next week. The agent does not start from zero every time._

Sample prompt:

> Create a competitor pricing sheet in the workspace. Check it every morning at 4, write the changes into the sheet, and message me first if anything drops. Don't wait for me to ask.

### 5. Run on a schedule, and build things in the cloud

Scheduled tasks fire on cron. They don't need you to be in the chat. Morning briefs, release watches, inbox cleanup — they belong on a computer that stays on. When something matters, it finds you first.

Dependencies, connectors, and skills live on that same computer. Install Node.js, Python, or a GitHub connector from Supermarket, and the agent uses them next time. Write code, run a dev server, and open a preview in the cloud.

![Scheduled tasks live on the bot and fire on time, without anyone clicking](/blogs/2026-09-15/05-schedule.png)

_The job hangs on the agent's computer, not on your laptop._

Sample prompt:

> Every weekday at 9am, summarize GitHub releases and issues from the last 24 hours and send them to the Lark group. Also build a reading-list web page on the cloud computer, run it, and give me the preview link.

## How to start

Pick the path that matches what you need right now. All three run the same Memoh, not three different products.

**1. Use Memoh Cloud**

Open [app.memoh.net](https://app.memoh.net), sign in with GitHub or Google, and give your agent a hosted cloud computer.

**2. Download Desktop**

Get the macOS, Windows, or Linux installer at [memoh.ai/download](https://memoh.ai/download). Desktop starts Memoh on your machine. It's the fastest way to try it while keeping files local.

**3. Deploy to your own server**

Use this when you need always-on uptime, multiple users, or channels that stay connected after your laptop goes offline:

```bash
curl -fsSL https://memoh.sh | sh
```

Source is at [github.com/felinics/Memoh](https://github.com/felinics/Memoh), licensed AGPLv3. Docs are at [docs.memoh.ai](https://docs.memoh.ai).

## Frequently asked questions

**How is Memoh's cloud computer different from regular chat?**

Regular chat is one session. When the session ends, the environment often goes with it. Memoh gives each agent a persistent workspace: files, desktop, browser, network, memory, and scheduled jobs stay there. That's how it can take a shift, instead of answering a single message.

**How should I choose Desktop, self-hosting, or Memoh Cloud?**

If the work is in your local files and apps, use Desktop. If you need 24/7 uptime, multiple users, or channels that don't drop, use self-hosting or Memoh Cloud. Cloud hosts the computer for you. Self-hosting runs the same software on your own infrastructure.

**Is my local data safe?**

On Cloud and self-hosted deploys, the agent's workspace runs in an isolated container. It does not get your laptop desktop by default. Desktop and "connect your own computer" are different: you are explicitly giving a bot files and a shell on that machine, and writes/executes ask for approval by default. Don't treat an unisolated local runtime as a normal cloud workspace.

**Can I put Claude Code or Codex on it?**

Yes. Enable an ACP agent on the bot's Agents page and finish setup with an API key or OAuth. They run inside that bot's container workspace and share the computer with Memoh's own agent. Codex, Claude Code, and Hermes ship as built-in setup paths.

**Can I run more than one agent?**

Yes. Run one for yourself, give each teammate one, or spin up a fleet on the same deployment. Each bot has its own workspace, tools, channels, and permission boundary.

**What OS does it run?**

The agent's cloud computer is an isolated Linux workspace. The official Docker deploy uses a container runtime. Desktop runs Memoh on your macOS, Windows, or Linux machine.

## After you close the laptop

What an agent lacks is not a longer next reply. It lacks a computer that stays on: files still there, jobs still running, and a screen you can sit down in front of when you need to.

Memoh turns that into a product. Open source, self-hosted, or Cloud. Give it a computer. Then close your laptop.

[Start with Memoh Cloud](https://app.memoh.net) · [Download Desktop](https://memoh.ai/download) · [Read the docs](https://docs.memoh.ai) · [GitHub](https://github.com/felinics/Memoh)
File renamed without changes.
File renamed without changes.
Loading