By Billy G.R. · 2 October 2026
Checked 2026-10-02. Author: Billy G.R. Retail prices move; the hardware catalog stores the Amazon snapshot, not a promise of stock.
Claim. The product name is Grok Bot, two words. The work runs on a cloud computer the bots share. The desktop and iOS apps are control surfaces.
Method. xAI news post at x.ai/news/introducing-grok-bot, read 2 October 2026. Product page x.ai/bot.
As of.
Claim. The same post lists access for SuperGrok, SuperGrok Plus, SuperGrok Heavy, and for Cursor Pro, Pro+, Ultra, plus Cursor Teams Standard and Premium. Enterprise is a waitlist there. Usage is separate from Grok and Cursor plan usage.
Method. Direct read of the introducing post. Not inferred from grok.com chat.
As of.
Claim. A local teammate at 27B Q4 needs about 18.2 GB. A 7B always-on helper needs about 6.2 GB.
Method. Qwen3.8-27B: 27 × 0.5 × 1.2 + 2. K2-Horizon-7B: 7 × 0.5 × 1.2 + 2.
As of.
Methodology — parameter math, quantization bytes, and the source list.
Grok Bot is xAI’s name for always-on AI teammates. The introducing post says each bot has a computer, signs into tools, and keeps working when you step away. That computer is in their cloud. The desktop app and the iOS app are how you message the bot. They are not a local weight file. Typing into grok.com chat is a different surface. Overview docs are at docs.x.ai/grok-bot/overview, and Cursor’s note is at cursor.com/docs/grok-bot.
On 2 October 2026 the news post listed Grok Bot in beta for SuperGrok, SuperGrok Plus, and SuperGrok Heavy, and for Cursor Pro, Pro+, Ultra, and Cursor Teams Standard and Premium, on desktop and iOS. It said bot usage is separate from your Grok and Cursor allowances, and that enterprise access was a waitlist. Bots can message each other, sit in a group, and keep a routine after you demonstrate a workflow once. I am not repeating the customer quotes. I also am not claiming you can download Grok weights and run them beside Ollama. Nothing on those pages says that.
If the appeal is “a named coworker with a browser and a filesystem,” the local version is a self-hosted agent plus a model that fits. If the appeal is Grok’s cloud computer and their login to your SaaS, staying on Grok Bot is the honest choice. The comparison hub lines this up against Meta Muse and ChatGPT dots.
CopilotKit OpenBot is the closest shape: agents as named channels, a computer per agent, your model, your Postgres. It is not OpenBots.ai, the healthcare RPA company. If you want a coding teammate instead of a company chat, use OpenHands with a Docker sandbox and an Ollama or vLLM backend. CrewAI is how you give those roles a script when you want more than one specialist sharing one model server. browser-use covers GUI apps that have no API. None of these products are “Grok, local.” They are a different stack that happens to answer the same chore.
Grok Bot’s post describes several bots in parallel on one shared cloud computer. Locally, parallel bots do not get free computers. They share VRAM. Prefer one 27–32B server that every role calls, unless you have a 128 GB box and a reason to load a second model.
Grok Bot itself needs zero local VRAM. Quality of the replacement tracks the model you can hold. Formula: params × bytes × 1.2 + 2, Q4 bytes 0.5. The calculator is the interactive check.
| Tier | Model | Role |
|---|---|---|
| 6–12 GB | K2-Horizon-7B Q4, 6.2 GB | One always-on helper. Not a fleet. |
| 16–24 GB | Qwen3.8-27B Q4, 18.2 GB. K2-32B Q4, 21.2 GB. Muse Glimmer Q4 formula 20 GB. | The primary teammate. 32B is tight on 24 GB once the KV cache grows. |
| 96–128 GB | A large MoE quant, for example DeepSeek-V4-Flash class (284B total). | Room for a bigger brain, or one strong model plus embeddings. Still not N full 70B copies. |
Active parameters on a MoE change speed. They do not change the fact that every expert has to reside. A “13B active” Flash model with 284B total weights is a 128 GB conversation, not a 16 GB one.
An always-on single teammate fits the Mac mini M5 Pro 24 GB (B0HGGHNQY6, $1,669.99 on 1 October 2026) if you stay at 7B or accept a tight 27B. Several roles plus a larger model want the EVO-X2 128 GB (B0F53MLYQ6, $3,649.99 the same day). I would not buy a second mini and call it a “team.” The weights do not split cleanly across two 24 GB machines without you building that serving layer yourself.
Mac mini M5 Pro 24 GB on Amazon
ASIN B0HGGHNQY6. Amazon Associates link. The live price is on that page, not locked in here.
GMKtec EVO-X2 128 GB on Amazon
ASIN B0F53MLYQ6. Amazon Associates link. The live price is on that page, not locked in here.
A correction or a hardware question goes to the contact page.