What Grok Bot Is and How It Differs From Other Grok Bots

Most chat AI tools wait for you to type, answer once, and stop. Grok Bot (at x.ai/bot) gives you Bots instead. A Bot is a named AI agent that stays around between conversations, works through multi-step tasks, and keeps going in the background.

Grok Bot is made by SpaceXSI (formerly xAI) with Cursor, a sister company whose infrastructure it runs on. It launched in beta on 11 August 2026. Its screens have changed between versions since then, and this course tells you where yours may look different.

An agent is an AI that takes actions as well as writing text. A Bot can open a browser, read files in connected apps, draft documents and report its progress. Before a risky action, such as sending an email, it stops and shows an approval card, which lets you allow or deny that one action.

Example. In a chat tool, you paste in a new deposition notice and ask for a summary. With a Bot, you tell it once to check the matter folder every morning, draft an inventory entry for anything new, and draft a note to the paralegal without sending it. Tomorrow it does that without being asked again.

Each Bot needs somewhere to work. In Grok Bot, all of your Bots share one cloud computer. It is a remote machine, located in the US, with a browser, a shared file folder called /workspace, and a terminal (a text window for running commands). It keeps running when your laptop is closed or your phone is off. That is why work can happen overnight.

Each Bot gets its own screen on that computer, but files, browser sessions and logins are shared. If your Chronology Bot signs in to the firm's document system, your Research Bot can use the same login. The official docs say plainly that separate Bots are not a security boundary. Your computer is isolated from other users' computers, but your own Bots are not isolated from each other.

This matters early in legal work. Giving a sensitive matter its own Bot does not protect it. What protects it is controlling what the whole account can reach. Later lessons cover how: firm accounts only, the least access that works, and nothing connected until it has vendor approval.

Three features turn a Bot from a helper into a standing worker.

Memory. Each Bot keeps preferences, facts, summaries and a dated work log (a record of what it did and when). A few basics, such as your name and time zone, are shared across your personal Bots. Memory is not a source of truth. Suppose a Bot remembers last month's trial date, and the court has since moved it. The Bot will repeat the old date unless you tell it to reopen the live scheduling order.

Skill. A saved, reusable procedure. You get the chronology update right once in chat, then save it as a skill called "update chronology" with its sources, rules and output format.

Routine. A routine runs a workflow on a schedule ("every weekday at 8:00 AM") or when an event happens, such as a Slack message, an email to the Bot's own address, or a webhook (an automatic signal from another system). The pattern is: a skill describes how the work is done, and a routine decides when it runs.

The Grok chat app (grok.com and the Grok mobile app) has a feature that is often called Custom Agents. Grok's official release notes confirm "Agent customizations" (17 February 2026) and an "Agent library" (3 March 2026). A Custom Agent is a saved persona: a name, a personality and instructions you choose inside Grok chat.

It is a much thinner feature than Grok Bot. A Custom Agent has no cloud computer of its own, no Bot roster and no routines of its own. No source we read says it keeps working while you are away. The Grok app does have separate features called Connectors, Skills and Automations, but those belong to the Grok app, not to Grok Bot. Don't mix them up.

The menu path, the character limit for instructions, and the number of agents you can create are reported only by third-party guides, and those guides contradict each other. For that reason this course only compares Custom Agents with Grok Bot and does not teach how to set one up.

Example. A Custom Agent named "Clause Explainer" can answer questions about a clause you paste into chat. It can't watch a matter folder overnight.

The third kind is an API-built bot. An API (application programming interface) lets a developer's code call Grok models directly. Someone writes the code, gets a key, hosts the bot on a server, and connects it to Slack, Discord or Telegram by hand. It can only do what that code allows. It is billed per token (per chunk of text processed), separately from any Grok plan. A firm might build one for its client portal, but that is a development project. This course does not cover it. No code is needed here.

Two names on X cause confusion:

• @grok is the chatbot that replies to posts on X.
• @bot is the Grok Bot account that forwards a tagged post to your own Bot.

Neither of them is your matter Bot, and tagging on X is out of scope for this course. For legal work, the rule is simple: client matters never go near a public post.

Flip the cards to see the main differences.

Andrew Perlman, Dean of Suffolk University Law School, describes a new kind of issue spotting. Lawyers learned to spot which doctrines a client's problem raises. Now they also need to spot how the legal service could be delivered better, faster and cheaper. Ask both questions about every workflow: what is the legal issue, and what is the delivery issue?

A Bot suits work that is recurring, document-heavy and tied to fixed points in the matter. Tyson Mutrux, a personal-injury firm owner, describes a red-flag review at about week three of every case. AI analyzes the file and flags concerns, then creates a task for the attorney, who reviews both the flags and the file. Other good fits are the day before a matter meeting, a new batch of documents arriving, and a chronology that has to stay current.

A Bot does not suit privilege determinations, filing, sending discovery requests or signing. Those stay with the attorney. A one-off question about a pasted clause needs only a chat tool. A product the firm builds for its own clients is a developer's API project.

Before the next lessons cover plans and safe setup, practise the decision you'll make for each workflow you think of handing to a Bot.

Start with the legal issue and who decides it. That doesn't change. Next, name the delivery issue: which preparation steps (reading, extracting, comparing, drafting) a Bot could do. Then check the fit: does the work recur, run on documents, and have a fixed trigger? Next, remove anything that stays with the attorney: privilege calls, filing, discovery requests, signing and sending. Only then check whether your plan includes Grok Bot and whether your firm and client have approved it as a vendor. Do both before any matter system is connected.

Example. Take a commercial-lease dispute with a weekly client call. The legal issue is the scope of the repair covenant, and the partner decides it. The delivery issue is that the call agenda and chronology updates take a paralegal two hours each week. Both recur and run on documents. Nothing in them gets sent or signed. Next comes the eligibility and approval check, which the following lesson starts on.