Three tools, one shared brain, and no idea which tool saved what. Agent profiles give each one a name, its own key, its own access, and a private memory.

Last time we made a simple point: teams of bots need a shared memory, task list and tool belt that you own. Context isn't the runtime's job. It's yours.

That raises the next question. If every bot reads and writes the same context, how do you tell them apart?

One brain, no names

Picture three tools connected to the same Gryz account: Claude Code for building, Cursor for reviewing, a chat assistant for planning. They share memory and tasks, which is the point. But with a single key between them, the problem shows up fast:

A memory appears and you can't tell which tool saved it.
A task is "assigned to the agent", and nobody knows which one.
One tool gets a bad idea and you can't cut it off without cutting off all three.
Every tool can touch everything, even when it only needs a slice.

Shared context is the goal. Anonymous context is the bug.

What an agent profile is

In Gryz, an agent profile is a team member: a named helper, like QA or Dev, that you connect from an AI tool.

A name and a handle. "QA", @qa. Tasks are assigned to the handle, so "my tasks" means the same thing everywhere.
Its own key. Each connection has its own key. Revoke one without touching the others.
Scoped access. Pick what each member can use: memory (none, read, or read and write), tasks, projects, skills, documents.
A visible trail. Every memory and task shows who saved it and through which tool. The member's page keeps a history of changes.
A profile it can read itself. The get_profile tool returns who a member is, what it has saved, what is open for it, and what it did recently, in one call.

Shared by default, private when it matters

All your team members share the same memory and task list. That is what lets Dev pick up where Plan left off.

Each member also gets its own private memory:

Only that member can read it. Other members can't see it, and can't tell it exists.
A member never sees your personal memory.
You can read, add and delete a member's private memory from the dashboard.
It follows the member across tools. Connect the same member from a second tool and it remembers.

Use private memory for the things only one role needs: how QA likes its bug reports, what Dev already tried on a flaky test. Everything the whole team should know stays in shared memory.

Why this shape

We could have built a "multi-agent runtime". We didn't, for the same reason as before: Gryz owns the context, not the execution. Your bots run wherever they run. A profile is just the identity and access layer on top of your shared graph.

It works with the tools you already use, including Cursor, Claude, ChatGPT and Grok, over MCP.
Nothing is locked to a vendor. The profile, its key and its memory are yours.
Least privilege is the default. A reviewer that only reads doesn't get write access.

What it doesn't do

It doesn't run your agents. It gives them a name, access and memory.
Members share one memory and task list. Access decides what each can use, but there's no separate workspace per member.
Plan limits apply: 1 team member on Free, 3 on Pro, 10 on Pro Plus, with 3, 9 and 30 connections. Tools you connect the usual way still get an automatic member, which doesn't count toward the limit.

FAQ

What is an agent profile? A named team member on your Gryz account, with its own key, scoped access and private memory, that you connect from an AI tool.

Do members share memory? Yes, by default. All members share the account's memory and tasks. Each also has a private memory only it can read.

Can one member see another's private memory? No. It can't see it, and it can't tell it exists. You, the account owner, can see all of it from the dashboard.

Can I limit what a member can do? Yes. Set access per area, like memory read only, tasks on, documents off. You can change it later.

Does it work with Cursor, Claude, ChatGPT and Grok? Yes. Any MCP client can connect with a member's key.

What does it cost? Free accounts get 1 team member, Pro gets 3, Pro Plus gets 10. See pricing.

Try it, if it fits

Agent Team is rolling out gradually. Once it's on for you, open Team in My Docs, create a member, and connect a tool with its key. The Agent Team guide walks through it, and the MCP tool reference covers get_profile.

If you run more than one bot against the same work, give each one a name.

Give every agent a profile: named team members that share your context