All comparisons
OpenCharts · Always-on workers
Theo Claw vs Grok Bot
xAI's always-on AI teammates with a shared cloud computer, distributed with Cursor
Grok Bot gives every account one cloud computer and a roster of bots that sign into your apps. Theo Claw gives every worker a computer of its own and a governed runtime around it: pick the engine per job, watch the browser and take the keyboard for a login, approve a whole plan at once, let a worker earn the right to stop asking, read the receipt afterwards, and reach it by text, email, or a phone call.
By OpenCharts TeamPublished Updated
Theo Claw is OpenCharts' always-on autonomous worker, comparable to xAI's Grok Bot: both give every worker a persistent computer with a browser, files and a terminal, run multi-step jobs across your connected apps without supervision, let you watch the live browser and take over for a login, coordinate several workers in one room, learn a job by being shown it once, and run named routines on a schedule. Theo Claw differs in governance, isolation and reach. Each worker gets its own computer instead of one shared per account, routes across engines from several labs instead of one model, earns autonomy one kind of action at a time, records a receipt for every run, can rehearse a configuration change against its own past runs before it goes live, treats running out of credits as a pause rather than an open meter, can run on your own provider key, and holds its own phone number, email address and calendar. Grok Bot leads on distribution: it ships bundled with Cursor and SuperGrok plans, with native desktop apps and a large plugin catalogue.
Theo Claw vs Grok Bot✳The verdict✳When to choose✳Feature matrix✳Pricing snapshot✳Migration✳FAQ✳Theo Claw vs Grok Bot✳The verdict✳When to choose✳Feature matrix✳Pricing snapshot✳Migration✳FAQ✳
The verdict
Where each one wins
Theo Claw wins at
- A computer per worker, not one per account: a worker's files, browser sign-ins and running jobs never leak into another worker's
- A choice of engines from several labs per worker, with cost tiers, or your own provider key, instead of one model with no picker
- Receipts for every run: the engine used, the sources read, the messages sent, the credits spent, and which actions ran on earned trust
- Autonomy that is earned: approve the same kind of action a few times in a row and the worker stops asking for that kind, and only that kind, with deleting never earnable
- Rehearse a change before it ships: replay a worker's own past runs against a new engine, persona or tool set and see what would change, without executing anything
- Out of credits is a pause that resumes on a top-up, not an uncapped weekly meter
- A worker with its own phone number, email address, and calendar that answers calls and books meetings
- One approval for a multi-step plan, with each step marked reversible or not, and org-wide approval rules
- Deliverables land as editable decks, documents, sheets, and apps in the same workspace
Grok Bot still wins at
- Bundled distribution: included with Cursor and SuperGrok plans rather than sold as part of a separate workspace
- Native desktop apps on macOS and Windows with the same thread on every surface
- A large plugin catalogue (around 220 reported at launch) for apps the bots can sign into
How they compare
Theo Claw and Grok Bot solve overlapping problems differently. The matrix below makes the trade-offs concrete, and the migration note covers what it takes to switch when you're ready.
When to choose
The right pick for your team
Choose Theo Claw when…
- You want to choose, or let Theo choose, which engine a worker thinks with, and see what each run cost.
- Your team needs approvals, receipts, a trust ladder and org-level rules before it hands real work to an agent.
- You want each worker on its own computer so one worker's sign-ins and files never reach another's.
- You want a worker people can text, email, or call, not only a chat window.
- The output should be a deck, a document, a sheet, or an app you keep editing, not a file in a folder.
- You already run OpenCharts and want the worker inside the same plan and workspace.
Choose Grok Bot when…
- Your team is already on Cursor or SuperGrok plans and wants agents bundled with them.
- You want a native desktop app rather than a browser workspace.
- You prefer one shared computer that every bot on the account signs into with your sessions.
Side by side
Feature comparison
Feature
Theo Claw
Grok Bot
Which model does the work
A Brain policy per worker: let Theo orchestrate across engines from several labs, pick economy, or pin one engine, with per-lane routing for writing, reasoning, vision, and code
One model, chosen automatically; no model picker for members or admins
Where the worker runs
A persistent computer per worker (browser, files, terminal) that sleeps when idle and wakes with its sign-ins intact; a code worker's computer also holds a repository clone. Deliverables it leaves on its desk are saved to your Drive
One cloud computer per account (browser, files, terminal) shared by every bot on the account
Watching and taking over
Watch the live page beside the chat, act on its real fields and buttons, and hand it back; the worker can ask you to step in for a login, a code or a payment, and what you type is never recorded or shown to it
Live browser preview in a side panel, full-screen takeover, hand control back when done
Earning autonomy
Opt a kind of action in and a few consecutive approvals earn it; one rejection resets the streak; deleting is never earnable; every change is announced in the worker's own chat and recorded on its receipts
Always Allow per action, or approvals before send, publish, buy, or delete
Testing a change before it ships
Replay a worker's own past runs against a new configuration and diff what it would do against what it did; nothing is executed, and only lost reach or gained risk counts as a regression
Not offered
Approvals
Full autonomy, ask before risky, or ask before everything; one approval covers a whole multi-step plan with reversibility marked; org rules override any worker
Pauses before send, publish, buy, or delete; a model reviews tool calls; Always Allow available
Audit trail
A receipt per run naming the engine, the web pages and company knowledge read, the messages sent, the teammates used, and the credits spent
Audit view described as coming; enterprise release added access, network, and audit controls
Spend control
A per-turn credit gate warns early and pauses the run when the balance cannot cover the next step; a top-up resumes exactly where it left off
Weekly usage allowance with on-demand overage billed at token cost; no spend cap at launch
How you reach it
In-app chat, Slack channels, Telegram, SMS from its own phone number, email from its own address, live phone calls, a REST API, and a native phone tab where the roster shows who needs you and you approve or decline in one tap
Desktop and mobile apps; a group chat with other bots; plugins for Slack and mail
Long-running work
Missions that outlive a session: wait until a time or a page changes, engine-free monitoring, resume across hours or days; plus several named routines per worker, each on its own schedule, that switch themselves off after repeated failures
Always-on computer keeps working with the app closed; routines run on a schedule or an event
Teaching it a job
Take over its browser, do the job once, press Done: it saves the procedure as a skill it can edit and reuse, keeping the shape of what you typed but never the value; or describe the steps in chat; plus opt-in learning of how you write
Record your screen once and it saves the workflow as a routine that takes corrections
Working with other agents
Rooms where you and up to eight workers talk in one conversation, address a worker by name and let them message each other, with the roster showing what each one is doing; plus delegation with a checked output contract, live steering, stop-and-keep-partial, and a cap on how many run at once
Bots message each other, share threads, hand off ownership, and coordinate in group chats
Sharing a worker
Duplicate a tuned worker, or share its setup with your team or as a link; the template carries the goal, persona, tools and policies and never an account, a credential or a memory
Template exchange and a team marketplace
Posting to your own systems
A worker can post a signed message to an endpoint you registered, by name, over the same signed delivery your platform webhooks use; there is no free-form address for it to send to
Through plugins for the apps the bot signed into
What it produces
Editable decks, documents, sheets, apps, images, video, and notes in the same workspace
Work lands in the third-party tool it signed into; files in its cloud computer
Company knowledge
Reads a reviewed company wiki and cites the article, its space, and its review status
Connects to knowledge tools through plugins; cites the web
Scroll to compare
Pricing and migration
What it costs, what it takes
Pricing
Pricing snapshot
Theo Claw is included from OpenCharts Starter at $9.99/mo, with Pro at $24.99/mo and Teams at $14.99 per user per month; runs spend the same AI credits every other Theo surface uses, and a dedicated SMS number is a monthly add-on. Grok Bot is not sold on its own: it is included with SuperGrok, SuperGrok Plus, and SuperGrok Heavy, and with Cursor Pro, Pro+, Ultra, and Teams plans, with its own weekly usage allowance and on-demand overage billed from model and token cost. List pricing at time of writing (September 2026).
Migration
Migrating from Grok Bot
There is no import between the two. Recreate a bot as a Theo Claw worker from the wizard: paste its instructions as the goal and persona, tick the tools it may use, connect the same accounts, and choose an approval posture. Standing schedules become the worker's trigger or a mission; anything it should learn about you goes into its memory.
Questions
Frequently asked questions
Is Theo Claw the same kind of product as Grok Bot?
Yes. Both are always-on autonomous workers you message like a colleague: you give them a goal and the apps they may use, they run multi-step jobs without supervision, and they stop for your approval before anything consequential. The differences are in how the work is governed, which engine does it, and how you reach the worker.
Does Grok Bot let me pick the model?
Not today. Grok Bot selects its model automatically and exposes no picker for members or admins. Theo Claw's Brain policy lets each worker orchestrate across engines from several labs, run in economy mode, or pin one engine, with separate lanes for writing, reasoning, vision, and code.
Does Theo Claw have a computer of its own like Grok Bot?
Yes, and one per worker rather than one per account. Every worker can open its own desk: a persistent computer with files, a terminal and a browser whose sign-ins survive a restart. You can watch the live page beside the chat, act on its real fields, and hand it back; when the worker hits a login, a code or a payment step it asks you to step in rather than asking for a password. A code worker's computer also holds a repository clone. Grok Bot gives each account one shared computer that every bot signs into with your sessions.
Can I teach a Theo Claw worker a job by doing it in front of it?
Yes. Press Show me, take over the worker's browser, do the job once, name it and press Done. The worker turns what you did into a procedure it can reuse, edit and, if the job repeats, run as a routine. It keeps the shape of what you typed (an email address, eight characters) but never the value, and no screenshot of the page is sent anywhere. Grok Bot records your screen for the same purpose.
How does Theo Claw decide when to stop asking for approval?
You opt a kind of action in, such as sending messages or changing a calendar, and a few consecutive approvals earn it; one rejection resets the streak. Deleting can never be earned, an earned class never reaches past your organization's rules or a draft-held reply, and every change is announced in the worker's own chat and recorded on its receipts. Grok Bot offers Always Allow per action.
How do the two handle approvals and audit?
Grok Bot pauses before sending, publishing, buying, or deleting, and a model reviews its tool calls; a full audit view was described as coming at launch. Theo Claw pauses per its posture, lets one approval cover a whole plan with each step marked reversible or not, applies org-wide rules on top of any worker, and writes a receipt per run naming the engine, sources, messages, and credits.
What happens when a worker runs out of budget?
Grok Bot ships with a weekly allowance and on-demand overage billed from token cost, with no spend cap at launch. Theo Claw checks the balance every turn, warns before it runs short, and pauses the run rather than failing it; a top-up resumes the same work from where it stopped.
Can a Theo Claw worker be reached outside the app?
Yes. A worker can hold its own phone number for texts and live calls, its own email address and calendar, a Slack channel it listens to, a Telegram chat, and a REST API. Grok Bot is reached through its desktop and mobile apps and through plugins for tools like Slack and mail.
Which one runs several agents together?
Both. Grok Bot puts several bots in a group chat where they message each other and pass work along. Theo Claw has rooms: you and up to eight workers in one conversation, where you address a worker by name, workers can message each other within set limits, and the roster shows what each one is doing. Underneath, a manager can also delegate a sub-goal to a teammate with an expected output shape, steer or stop it mid-flight, and cap how many run at once.
Hire a worker that keeps working
Give a Theo Claw worker a goal, the tools it may use, and how much rope it gets. It runs while you sleep and asks before anything risky. Included from the Starter plan, with the whole workspace alongside it.
Included from Starter · Approvals and receipts on every run