01 / COMPUTER
A cloud work environment
xAI says Bots can sign in to apps and websites and continue working when the user's own computer is closed. This is positioned as useful where a clean API or MCP is unavailable.
Product definition
What is Grok Bot?
A source-led definition of xAI's beta product: the job model, cloud-computer workflow, multi-Bot coordination, routines, and public evidence boundary.
Beta product
The key word is delegation. Instead of keeping a chat open and moving each answer into another tool yourself, the product is designed around a Bot taking ownership of a job, working across the relevant software, and returning when the result or a judgment call is ready.
That is xAI's launch position, not an independent guarantee of completion quality. The public launch pages provide examples and testimonials but no reproducible task-success, error-rate, or escalation benchmark.
The product model
02 / THREAD
A persistent conversation and job record
Users can message a Bot from phone or desktop and return to the same thread. xAI says context and working preferences persist over time.
03 / ROUTINE
A demonstrated workflow becomes a routine
A user can show the Bot how a repeated job is done. The product page says the workflow is saved, rerun, and adjusted when the user supplies corrections.
04 / TEAM
Several Bots can coordinate
xAI says Bots can share context, message one another, pass ownership, and coordinate inside group conversations without the user routing every handoff.
What Grok Bot is not
It is not presented as a new public model ID. It is not the same thing as using Grok chat, and the reviewed launch material does not document a dedicated public Grok Bot API.
- Grok chat: conversation and content creation are the primary interface.
- Grok Bot: delegated work and completion inside other tools are the primary interface.
- xAI API: developers integrate documented model endpoints into their own software.
The evidence boundary
“Can operate,” “keeps context,” and “gets smarter” are vendor claims in the launch record. They are useful for understanding product intent, but they do not tell you how often a Bot succeeds on your workflow, how it handles stale context, or whether it asks for approval at the right moment.
Practical implication: start with a reversible job, a narrowly scoped account, a written acceptance test, and a human review step.