Governance and permissions
As your workspace grows, not everyone should see and edit everything. Governance is how you control who accesses each layer and table, and what the AI sees.
- Understand the Viewer, Editor and Manager levels
- Know what permissions apply to (layers, tables, MCP)
- Control exactly what an AI agent can access
At first, a workspace has few people and everyone trusts everyone. As it grows, the question comes up: should the media intern really be able to delete the Gold billing layer? Should client A be able to see client B's data? Governance is the answer: controlling who sees, who edits and who manages each piece of data.
Data-level permissions are a Growth feature
On the Starter plan, every member has Manager access to all data, which is great for getting started quickly, without barriers. When you need fine-grained control (isolating clients, protecting sensitive data, limiting what the AI can access), granular permissions come in, and they are part of the Growth and Custom plans.
Learn about the Growth plan Starter: everyone is Manager on everything · Growth and Custom: control per member and per resourceThe three permission levels
Every permission in Nekt falls into one of three levels. They are cumulative: each one can do everything the previous one can, and a bit more.
| Level | Can view and use | Can edit and create | Can delete and grant access |
|---|---|---|---|
| Viewer | Yes, read-only | No | No |
| Editor | Yes | Yes. Changes descriptions, creates dependent tables in the layer, adds Sources and Queries | No |
| Manager | Yes | Yes | Yes. Deletes the resource and grants Viewer or Editor to other members |
Anyone without any permission simply cannot access the resource. Workspace owners and admins override any restriction.
What permissions apply to
Permissions apply directly to data resources: layers, tables and volumes. Access to Sources, integrations and other features derives from the permissions on the tables and layers underneath. In other words, you think in terms of data, and the rest follows.
You can grant permission to a person directly, or to a group. When there is more than one, the rule below applies.
Nekt applies the highest permission a person has, whether it comes from a group or a direct assignment. Removing access from one group does not restrict the person if another group still grants it. When revoking access, check every path.
The link with AI: what the agent sees
This is where governance becomes powerful. The MCP server does not have its own permission system: the AI agent sees exactly what the token's user sees on the platform. In other words, the same permissions that protect people protect what the AI can read. This gives you two ways to control an agent.
| Token type | What it accesses | Updates on its own? |
|---|---|---|
| Full Access | Inherits everything the token's user sees. If the user gains or loses access on the platform, the token reflects it immediately | Yes, dynamic |
| Scoped | Only the tables chosen at creation. The list is fixed | No. To change it, edit or recreate the token |
Tools like execute_sql and
get_table_preview return an error when they try to access a
resource without permission, instead of returning a partial result. So
an agent with a Scoped token on your three
Gold tables will never stumble onto anything
sensitive.
When you set up an agent for a team or client, use a Scoped token on the Gold tables it needs. It is the safest way to open up AI without opening the whole workspace. It pairs directly with the guide Building an AI agent on top of your data.
When governance becomes essential
A multi-tenant agency needs to guarantee that client A's team never sees client B's data. With permissions per layer and per table, each team gets Viewer only on its own client's tables, and each client's agent uses a Scoped token. Isolation guaranteed, not by convention, but by rule.
An accounting firm handles sensitive financial data. Analysts get Editor on the working layers, but only partners have Manager to delete or grant access. The untouched Bronze layer, with restricted access, also serves as an audit trail.
Fine-grained access control, from data to AI
If your situation calls for isolating clients, protecting sensitive data or opening up AI safely, Growth's granular permissions solve it. Talk to the team to upgrade.
Learn about the Growth planUse cases that apply what this guide explains.