Agentic Data Plane

Tag Resources

Tags let access policies and spend reports act on a whole team, environment, or cost center instead of on a list of individual resource names. A tag is a key/value label, such as team=payments or env=prod, that you attach to an Agentic Data Plane resource. The payments team’s policy covers a new agent tagged team=payments, and the agent appears in the payments spend report as soon as you create it.

After reading this page, you will be able to:

  • Identify which Agentic Data Plane resources carry tags and where to set them

  • Distinguish how Agentic Data Plane uses tags in access policies, cost allocation, and list filters

  • Choose tag keys that serve both access control and cost reporting

Resources that carry tags

Set tags when you create a resource or edit it later. Each resource’s page describes its Tags field.

Resource In the UI With rpk ai

Agents

Yes

--tag on rpk ai agent create and rpk ai agent update

MCP servers

Yes

--tags on rpk ai mcp-server create and rpk ai mcp-server update

LLM providers

Yes

--tags on rpk ai llm-provider create and rpk ai llm-provider update

Guardrails

Yes

No

Budgets

Yes

No

Secrets

Yes

--tag on rpk ai secret create and rpk ai secret update, and --remove-tag on rpk ai secret update

OAuth clients, OAuth providers, and access policies also carry tags that policy conditions can read. You can’t set them in the UI or with rpk ai. See Attribute reference for the full list.

Limits and rules

  • A resource holds at most 50 tags, and each value can be up to 256 characters.

  • In the UI, every tag needs both a key and a value, and adding a key the resource already has replaces that key’s value.

  • An update replaces the whole set of tags. When you edit tags, keep the pairs you want.

  • Don’t put secrets or personal data in tags. Tags appear in lists, in policy evaluation, and in spend data.

For the limits on every resource, see Limits and Defaults.

How Agentic Data Plane uses tags

Agentic Data Plane reads tags in three places, and one set of tags can serve all three. A tag that none of them reads only organizes. A budget’s own tags, for example, don’t change which agents the budget covers.

In access policies

An access policy can permit or forbid an action based on a resource’s tags, with conditions such as resource.hasTag("team") && resource.getTag("team") == "payments". Lists of agents, MCP servers, and LLM providers return only the resources the caller’s policies allow, so a team granted access by tag sees its own resources and nothing else. See Give a team access to its own resources for a worked example, and Write conditions for the rules every tag condition follows.

In cost allocation

Every LLM call an agent makes records the agent’s tags with its spending event, as they were when the call ran. You can then group and filter spend by tag. Agentic Data Plane records only the calling agent’s tags, not the tags on its LLM provider or MCP servers, and a call that doesn’t come from an agent records none. See Track Spend by Tag.

In lists and filters

The lists of agents, MCP servers, LLM providers, guardrails, and secrets show each resource’s tags. The agents and secrets lists also have a Tag filter that matches a key=value pair, such as team=payments.

Choose tag keys

Pick a small set of keys and use them the same way on every resource type:

  • Use the same key, such as team, on agents, MCP servers, and LLM providers. One access policy and one spend report can then cover a team’s whole footprint.

  • Keep values lowercase and stable. A condition compares values exactly, so Payments and payments are different teams.

  • Tag at creation. A resource without the tag falls outside every policy and report that relies on it, and a team doesn’t see it in its lists.

Next steps