OpenMausBot: Claude, Codex as Persistent AI Bots (Grok Bot Alternative)

OpenMausBot is a free, open-source desktop app that turns AI coding agents into a small team of persistent bots. You can connect Claude, Codex, or Grok, create bots for different jobs, and give each one its own instructions, model, conversation history, and tools.

A coding bot can stay with one repository across multiple sessions, while a research bot keeps its own context and another handles work in Gmail, GitHub, Slack, Notion, or a remote Linux desktop. You can jump between them like ongoing conversations instead of rebuilding the same setup every time you start a new task.

The OpenMausBot harness runs locally on 127.0.0.1, with transcripts, keys, and events stored under ~/.openmausbot. Model providers, Composio connections, and Box cloud computers operate outside that local storage boundary when you use them.

OpenMausBot vs. Grok Bot

OpenMausBot is an independent open-source project inspired by Grok Bot. Both products organize AI agents as persistent contacts, but they run those agents in different environments.

Grok Bot is a managed cloud service with persistent cloud execution, mobile access, and scheduled routines. OpenMausBot runs its harness on the desktop, uses agent CLIs already available on your computer, and includes optional Box computers for remote desktop work.

How the Bot Works

When you create a bot, you choose its role and model. The bot then appears in the left sidebar with its own conversation history. Open it again tomorrow or next week and you return to the same thread, along with the instructions and context already attached to that bot.

That works well for jobs that keep coming back. A repository bot can stay with issue investigation, code changes, and release checks over time. A research bot can hold a long source-gathering thread. An operations bot can handle connected services without mixing those conversations into the coding work.

The sidebar includes controls for pinning, renaming, duplicating, hiding, marking unread, deleting, and copying a conversation ID.

For work that needs several bots at once, you can create a Room. Each Room has a bulletin for shared instructions, and @ mentions call a specific bot into the thread. Bot-to-bot communication is also included in the current desktop beta.

Agents and Model Selection

OpenMausBot works with agent CLIs that are already installed and authenticated on your computer.

Claude, Codex, and Grok are the main supported agents for everyday use. The built-in driver registry also contains Gemini Agent, Antigravity, Grok Agent, and Box Agent. OpenMausBot detects the providers available on your machine and shows their models in the model picker.

Model choice belongs to the bot. You can keep Claude on one bot, Codex on another, and change a bot’s selected model without starting a new conversation.

OpenMausBot does not include model access in the desktop app. Each provider keeps its own account, subscription, quota, and authentication requirements.

Example Bot Roles

You do not need a large agent team to make the app useful. A small roster can divide recurring work into a few recognizable conversations:

• Repository bot: code investigation, issue triage, patches, and release checks.
• Research bot: source collection, research threads, and summaries.
• Operations bot: Gmail, Slack, Notion, Linear, and other connected services.
• Browser bot: web and desktop tasks on an isolated Box computer.
• Review bot: second-pass inspection with another model.

Connected Apps

If a bot needs to work outside its own chat, Composio Connect links it to services such as Gmail, GitHub, Slack, Notion, and Linear. The available catalog contains more than 500 apps.

Basic agent chat works without Composio. External app access requires Composio credentials and authentication for the services you connect.

Computer Access

For browser work, desktop software, or jobs that need an isolated machine, a Box account provides a remote Linux computer with a desktop and terminal. OpenMausBot shows the remote screen in its Computer panel, and you can open that desktop in a browser when you want to take over manually.

macOS also supports local screen capture and local computer control. Ubuntu disables both in the current desktop beta. Windows supports the packaged app and Box computers.

Approval Cards

Long-running agents often reach points where they need permission before continuing. OpenMausBot places those requests directly in the conversation. Shell commands and file edits can wait for an Allow or Deny decision, while questions appear with a response field.

You do not have to hunt through another window to see what triggered the request. The proposed action and its approval stay together in the same chat.

Getting Started

macOS

The macOS release is distributed as a signed and notarized .dmg.

1. Download OpenMausBot.dmg.
2. Drag OpenMausBot to Applications.
3. Install and authenticate at least one compatible agent CLI.
4. Open OpenMausBot.
5. Create a bot.
6. Choose an available model.
7. Send the first task.

Windows

The Windows release uses a 64-bit per-user installer. Note that the installer is not code-signed, so Windows SmartScreen displays an unknown-publisher warning.

1. Download OpenMausBot-setup.exe.
2. Run the installer.
3. Authenticate a compatible agent CLI.
4. Open OpenMausBot.
5. Create a bot and choose a model.

Ubuntu 24.04 Beta

Ubuntu 24.04 x86_64 has a desktop beta for GNOME Xorg and Wayland. Chat, streaming responses, approvals, local data storage, connected apps, Box computers, and bot-to-bot communication are included. Local screen preview, local computer control, native dictation, and ARM64 packages are not part of this beta.

Build the Linux packages from source:

git clone https://github.com/milind-soni/OpenMausBot.git
cd OpenMausBot
corepack enable
pnpm install --frozen-lockfile
pnpm package:linux

Run From Source

git clone https://github.com/milind-soni/OpenMausBot.git
cd OpenMausBot
pnpm install

pnpm dev:server
pnpm dev
pnpm dev:desktop

How OpenMausBot Works Behind the App

OpenMausBot has a React front end and a local harness server. The harness starts agent processes, receives their events, and converts provider-specific protocols into one runtime event format. Server-Sent Events carry messages, tool activity, approval requests, and computer events back to the relevant chat.

Local Storage and Privacy

OpenMausBot stores transcripts, keys, and events under ~/.openmausbot. Its harness binds to the loopback address, and the desktop app sees whether a credential is configured without displaying the stored secret value.

The moment a bot calls an external service, that part of the task leaves the local machine. Claude sends model traffic through Anthropic. A Gmail task also involves Google and Composio. Work performed on a Box desktop runs on Box infrastructure.

If you plan to use OpenMausBot with private repositories, customer data, or sensitive documents, check the policies of the model provider and every connected service involved in that task.

"Local-first" describes the OpenMausBot harness and its application storage. It does not mean every bot task stays on the local machine.

Pros

• Persistent bot histories keep recurring responsibilities organized.
• Multiple agent providers are available in one desktop app.
• Existing agent subscriptions and logins remain usable.
• Approval requests stay attached to the relevant conversation.
• Composio connects bots to hundreds of external apps.
• Box provides remote Linux desktops for computer tasks.
• Rooms support shared conversations with multiple bots.

Cons

• Native desktop features vary across operating systems.
• Windows installation triggers an unsigned-publisher warning.
• Hosted model traffic falls outside the local runtime.
• Box and connected services can create extra costs.

Alternatives & Related Resources

• Wayland: A local-first command center for Claude Code, Codex, Gemini, and other agent CLIs, with shared memory and coordinated agent teams.
• AionUi: A desktop Cowork app for multiple CLI agents with project folders, MCP, file previews, office tools, WebUI access, and scheduled tasks.
• Manaflow: A better match when the main goal is running several coding agents in parallel across isolated workspaces and comparing their code changes.
• Aperant: A Claude Code-focused desktop harness for planning, implementation, QA, isolated worktrees, and controlled merges.
• 7 Best CLI AI Coding Agents: Compare the terminal agents you can run directly or connect to a higher-level desktop manager.
• Claude Code Resource List: Browse Claude Code agents, GUIs, skills, plugins, SDKs, MCP tools, and related developer resources.
• Grok Build: Explore the Grok coding CLI that can also run as a standalone terminal agent.

FAQs

Do OpenMausBot bots share the same context?

Not by default. Each bot keeps its own conversation and role. When several bots need the same discussion, you can bring them into a Room, where they see the shared thread and follow the Room bulletin.

Can OpenMausBot keep working after my computer goes to sleep?

The free desktop app depends on the local OpenMausBot harness, so it is not an always-on hosted agent service. Always-on bots, scheduled work, and remote access are part of the planned OpenMausBot Pro direction.

Can I change the model without creating a new bot?

Yes. Model selection can be changed during an existing conversation. The bot remains in the same OpenMausBot thread, while the available tools and behavior depend on the provider and model you select.

Does every bot need Composio or a Box computer?

No. A bot can work through an authenticated agent CLI without either service. Composio is only needed for connected apps, and Box is only needed when the bot requires a remote Linux computer.