Permissions and plans
Who can connect a chatbot, what it can do on each plan, and what it never does.
The chatbot acts as you#
Each connection belongs to one person and one workspace. Everything the chatbot does goes through the same rules as the console:
- your plan's limits (agents, HTTP tools, MCP servers, knowledge storage, quality suites);
- validation of the draft before saving a version;
- the quality gate, if the agent has one;
- the audit log, where every change appears under your name.
Who can connect#
| Role in the workspace | Can connect a chatbot? |
|---|---|
| Owner | Yes |
| Admin | Yes |
| Manager | No |
| Viewer | No |
If you stop being an admin or owner, the chatbots you connected lose access in less than a minute.
What each plan allows#
| Plan | Access from the chatbot |
|---|---|
| Piloto | Read-only: look up agents, versions, instructions, tests, and components |
| Automatiza, Orquesta, Opera, A medida | Full: create, configure, test, and publish |
With read-only access the chatbot doesn't even see the writing tools. If you change plans, access updates by itself within a minute, without reconnecting.
Note
Quality evaluations spend your account's credits, just as when you run them from the console.
What the chatbot never does#
For security, these actions only happen in the console. When they're needed, the chatbot gives you a direct link:
- Entering keys: provider API keys (Anthropic, OpenAI, Google, AWS) and an MCP connection's token.
- Authorizing services with OAuth: connecting a third-party MCP that asks you to sign in.
- Uploading files: the PDFs, Word and Excel files, and images of your knowledge bases. The chatbot can add Markdown text.
That way no secret goes through the chatbot's history or its provider's servers.
Not included yet#
This first version also doesn't let the chatbot:
- chat with the agent to test it;
- see your customers' conversations or usage;
- manage members, billing, or the workspace itself.
All of that is still available in the console.