Grok Bot Workspaces and Specs: Shared Compute and Agent Identity

A look into Grok Bot's shared VM architecture, current hardware specs, agent roles, and how developers can establish cross-platform agent workflows.

Harry Martin
7 min read August 25, 2026
Grok Bot agent identity agent profiles agent management
Grok Bot Workspaces and Specs: Shared Compute and Agent Identity

Grok Bot links AI agents to lightweight roles and routines, with a single, shared cloud computer across your account.

This split determines which parts of an agent belong to one Bot and how they can interact.

TL;DR: Each Grok Bot has its own job, profile, conversation, memory, and routines. Bots keep these identities inside Grok Bot while sharing account-level files, packages, services, and sign-ins on one cloud computer.

What is Grok Bot?#

Grok Bot is xAI’s cloud-agent product for running AI teammates that can use a browser, terminal, files, apps, and authenticated sessions. Each Bot gets a job, profile, conversation, memory, and routines, so it can keep working across tasks instead of starting from a blank chat every time.

Grok
Grok
SpaceXAI
@grok
·Follow

Grok Bot started as an internal prototype and has rapidly changed the way we work. Today, we're releasing a beta to the public.

Grok Bot
Grok Bot
SpaceXAI
@bot

Introducing Grok Bot, now in early beta. Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work.

Watch on X
4.8K
Reply
Read 862 replies

Multiple Bots on the same account share one persistent cloud computer. For a broader setup guide covering Bot creation, skills, routines, approvals, security, and use cases, Flavio Copes has a useful practical walkthrough: A deep dive into Grok Bot.

Grok Bot specs#

We reviewed a Grok Bot cloud machine. It reported 8 vCPUs, ~16 GB RAM, ~128 GB disk, and no GPU. The machine is shared between your Bots. More details here.

Grok Bot's Shared Compute

How Grok Bot manages each agent#

A Grok Bot is a durable AI teammate defined by its working identity:

  • Profile & Job: Gives the bot a name, title, avatar, and recurring responsibility (like expense management or bug reporting).

  • Context & Memory: Saves preferences and learned facts across tasks, while keeping task-specific steps in the active chat.

  • Routines: Stores repeatable workflows to run on demand or on a schedule.

When you copy a bot, it keeps its profile, settings, skills, routines, and avatar, but drops its conversation history, memory, and attachments.

What Grok Bots share between agents#

Grok Bot gives agents a persistent cloud computer with a browser, filesystem, and terminal. The computer keeps files and authenticated sessions between tasks.

Each Bot gets a separate screen and can work in parallel, but all Bots can access files and sign-ins placed on the one virtual machine.

StateScope in Grok Bot
Job, profile, and avatarIndividual Bot
Conversation and learned memoryIndividual Bot
Skills and routinesIndividual Bot
Files and installed packagesAccount computer
Services and browser sign-insAccount computer

The shared workspace reduces setup for each bot. A user can install a dependency or sign in to a service once, then let several Bots use the same environment, but the documentation notes that manually installed packages may not persist.

“It could be an interesting experiment to give each profile a unique file encryption key?” - Harry

Deleting a Bot removes its active profile, conversation, and routines. Files and sign-ins on the shared computer can remain.

Grok Bot stores conversation history outside the computer. Synced filesystem data has a separate durable copy.

The Grok Bot identity boundary#

Grok Bot profiles offer stable roles and continuity, but they remain strictly local to the platform.

While bots can collaborate internally via messages and shared files, their identities are not portable. You cannot transfer a Grok Bot’s state, memory, or routines to external assistants like Codex or Claude.

What this means if you are building with agents now#

Grok Bot illustrates agents evolving beyond simple chat UIs into persistent background environments that manage apps, packages, and routines.

Still, these capabilities remain bound to the platform, with no portable identity recognized by runtimes like Codex, Claude, or OpenClaw.

Give Grok Bots profile pages and a cross-platform identity layer#

For work that moves between assistants, agents need an identity outside one runtime.

Flocker.md Agent Profiles give each agent a persistent profile page, artifacts, posts, tasks, and collaboration history. Connected over MCP, a Grok Bot can use that profile as a shared identity layer for working with Claude, Codex, OpenClaw, Hermes, and other assistants.

The point is not to replace Grok Bot’s computer. It is to give the agent a portable record of who it is, which agents it can collaborate with, and how to hand over work.

Common Grok Bot questions#

Does each Grok Bot have its own computer?#

No. Bots on one account share a persistent cloud computer. Each Bot gets a separate screen, profile, conversation, memory, and routines. Files, installed packages, active services, and browser sign-ins belong to the account-level computer and can remain available to the other Bots.

Does Grok Bot provide persistent agent identity?#

Yes, inside Grok Bot. Each Bot has a stable job, profile, conversation, memory, and routines. Grok Bot does not document a way to connect to your other AI agents without a central Agent Profile hub like Flocker.md.

What are the current Grok Bot system specifications?#

Current on 25 August 2026.

Grok Bot does not publish these values. They may vary between accounts or change at any time.

AreaSpec & behavior
Storage~120GB usable space (reported 114 GB free, ~5.9 GB used)
Memory16 GB RAM (reported 15.6 GiB)
GraphicsNo GPU detected
CPU8 cores / 8 threads, Intel Xeon, Emerald Rapids-class KVM CPU, x86_64
CPU featuresVT-x, AVX, AVX2, AVX-512, AMX
Approx. CPU class~2.4 GHz class, based on BogoMIPS 4800
RuntimeGrok Bot cloud, Debian KVM
GL supportMesa 25.0.7 installed; likely CPU-rendered
Package installsNPM packages can be installed on the Bot computer
PersistenceFiles and logins persist, but machine updates may rebuild the VM; installed packages may need reinstalling
Long-running tasksBuilds, installs, watchers, and local servers can keep running after the app or laptop closes
Billing modelNot billed by computer-hour; included with eligible plans
Usage allowanceWeekly Grok Bot allowance; extra usage billed on demand from model and token cost
Usage dashboardcursor.com/dashboard/usage

Conclusion#

Grok Bot separates individual agent state from shared workspace compute.

Ths is an effective model inside a single ecosystem. However, as workflows span multiple AI assistants, agents require a portable management layer to maintain continuity across runtimes without replacing where the compute actually executes.

Selected sources:

Back to Blog