Dots vs Grok Bot vs Muse, and what teams of bots still don't share

Within one month, three big companies launched the same idea. Meta launched Muse on September 8. xAI had already put out Grok Bot in August and opened it to enterprises on September 3. OpenAI announced Dots on September 29.

In short, each one is an always-on agent with a computer of its own in the cloud. It signs into your apps, keeps working after you close the laptop, and checks with you before anything risky.

All three point the same way: from one bot to a team of bots. OpenAI says it wants "teams of dots working together on your behalf." Grok Bot is already sold as "your team of always-on agents." Muse launches subagents of its own.

That's real progress. But a team of bots has the same problem as a team of people. Hiring more workers doesn't help much if each one keeps its own notebook — and its own key ring.

What they're actually selling

Leave out the branding and all three products sell a runtime: a place where agents live and work.

• A computer. Each agent gets a cloud machine with a browser. Muse runs on its own secure VM. Dots and Grok Bot each run on hosted cloud computers.
• Access. You connect your apps, and the agent works in them for you.
• Always on. Work keeps going while you're away, and the agent comes back when it needs a decision.
• Guardrails. Each one has an approval layer for sensitive actions.

That's hard engineering, and all three are working on it seriously. Running agents is a real job. It's just not the only job.

Where teams of bots still break

Think about a kitchen. You can hire the best cooks in town. If each one keeps their recipes in a private notebook, writes orders on their own sticky notes, forgets who's allergic to what, and only has a key to their pantry, dinner still comes out wrong.

Bots fail the same way. Four things decide whether they work as a team:

• Memory. What's true about you, your projects, and your decisions, and what you've already ruled out.
• Skills. The how-to: your routines, your conventions, the right way to do a task here.
• Tasks. What's in flight, who owns it, and what's blocked.
• Tools. The same GitHub, Jira, Slack, and other systems every bot should reach — without reconnecting each runtime from scratch.

Today each product keeps all four inside its own walls. Grok Bot says its bots can message each other and share context, but inside Grok Bot. A dot learns your preferences, but inside ChatGPT. Muse gets sharper as you use it, but inside Muse. And each one wants you to wire up your apps again, inside that product.

That's fine if you only ever use one product. Most people and teams don't:

• Your coding agent is in one tool, your always-on assistant is in another, and a teammate likes a third.
• Next quarter's best model may come from a different company than this quarter's.
• Some of the "team" is people, and people need to see what the bots know too.

When context and connections stay private to each runtime, you end up passing them along by hand. You repeat the same preference to three assistants, paste the same checklist into three setups, reconnect the same apps three times, and no single place shows what's done.

A runtime vs. a shared graph you own

The question isn't which bot is smartest. It's where the context lives — and how tools get reached.

These aren't either/or. A runtime is where a bot works. A shared graph is what it knows when it shows up — and how it reaches the systems you already use. You can use both.

What Gryz is, and isn't

Gryz is shared context for people and agents. Memory, Skills, and Tasks sit in one store that you own. An MCP Broker sits beside them so tools do too. All of it connects over MCP to the clients you already use.

• Memory. Small, attributable facts. Personal facts follow you everywhere, and project facts stay tied to a repo or workspace. They're private to your account and never used for training.
• Skills. Write the how-to once, and any connected agent can pull it.
• Tasks. One shared list of what's in flight and what's blocked, readable by bots and people alike.
• MCP Broker. Register GitHub, Jira, Slack, or any MCP server once in the Gryz dashboard. Any agent connected to Gryz can search those tools and call the ones you've enabled. Gryz permission-checks and proxies the call; a disabled tool never appears.

What Gryz isn't: a cloud computer. We don't run your agents or host your bots, and we're not trying to replace Dots, Grok Bot, or Muse. Let those products run the work. Gryz is the layer underneath that keeps the work consistent when there's more than one of them — same memory, same skills, same tasks, same tool belt.

Longer term, we think this layer becomes something like an operating system for teams of bots. We aren't there yet. Today it's simpler: one place where your bots and your people read from the same page, and reach the same tools.

FAQ

Is Gryz a competitor to Dots, Grok Bot, or Muse?
Not really. They're runtimes, places where agents run. Gryz is shared context — and a tool broker — those agents can plug into. You can use Gryz alongside any of them.

Don't these products already have memory?
Yes, each has its own. The catch is that it lives inside that product. Gryz is memory that isn't private to any one product. You own it, and any MCP client can read and write it.

Don't these products already connect my apps?
Yes, inside each product. The MCP Broker is for tools every bot should share: connect once in Gryz, enable what you want, and any MCP client talking to Gryz can discover and call them.

Grok Bot says its bots share context with each other. Isn't that the same thing?
It covers bots inside one product. Gryz is for context — and tools — that have to cross products: your coding agent, your assistant, a teammate's tool, and the people looking over all of it.

Does Gryz run my agents?
No. Gryz doesn't run models or host computers. Your agents run where they already run, and Gryz gives them a shared memory, skill set, task list, and tool broker.

What if I only use one assistant?
Then the gap shows up the day you switch. Owning your context and your tool connections means a new model or product doesn't start from zero.

Who can see what's stored?
You. Facts are private to your account, visible and editable in your dashboard, and never used for training. Brokered tools only include the ones you've explicitly enabled.

Try it, if it fits

You don't need to move anything. Connect Gryz as an MCP server in Claude, Cursor, or any MCP client (setup guide). Then ask your assistant to remember the next decision worth keeping. The Agent Memory overview covers scopes and the dashboard.

There will be more bots, and that's good. What should stay put is what they know about you — and how they reach the tools you already use.