Agents

The Agents view is the registry for every agent attached to the current Studio process. Use it to confirm that Studio loaded the expected agent and to review its model, runtime composition, and enabled capabilities before starting a Playground run.

Open http://localhost:4021/ui/agents when Studio is running on port 4021.

Register agents with Studio

Pass one or more built agents to Studio:

import { AgentBuilder } from '@anvia/core/agent'
import { OpenAIClient } from '@anvia/openai'
import { Studio } from '@anvia/studio'

const client = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
})

const supportAgent = new AgentBuilder(
  'support-operations',
  client.completionModel('gpt-5.6-luna'),
)
  .name('Support operations')
  .description('Looks up tickets and prepares support actions.')
  .defaultMaxTurns(4)
  .build()

new Studio([supportAgent]).start({ port: 4021 })

Each row uses the agent's stable id and displays its name and description when provided. Registering several agents adds them to the same Studio workspace; the selected badge identifies the agent currently shared by agent-aware views such as the Playground and Tools.

Read the runtime summary

Studio asks the running server for a summary of each registered agent. The summary describes configuration, not the outcome of a particular run.

Area	What Studio reports
Model	A readable provider/model identity when the model exposes one.
Tools	Registered static and dynamic tools.
MCP	The subset of registered tools carrying MCP server provenance.
Context	Static context plus dynamic context registrations.
Observers	Observers attached to the agent.
Prompts	Quick prompts configured for this agent in Studio.
Memory	Whether the agent has a memory implementation.
Hook	Whether the agent has a prompt hook.
Schema	Whether the agent requires structured output.
Turns	The configured default maximum number of turns, when set.

The counts are intentionally compact. For example, context 3 means the agent has three static and dynamic context registrations in total; it does not mean that all three were retrieved for the latest prompt.

Likewise, an MCP count is provenance, not an additional tool set. MCP tools participate in the agent's registered tools and are identified separately so you can see which capabilities come from connected servers.

What the view does not show

The Agents view is an inventory surface. It does not expose full instructions, context documents, memory records, tool arguments, or previous responses.

Use the more focused Studio views for those questions:

• Tools shows tool definitions, origins, schemas, and approval metadata.
• Playground runs the agent and streams its tool activity.
• Pipelines inspects registered workflows rather than agents.

If a row says Runtime unavailable or Waiting for summary, the browser could not load that agent's runtime endpoint. Confirm that the Studio process is still running and that the agent remains registered under the displayed ID.

Interpret capabilities carefully

A capability badge tells you that the agent is configured to use that facility. It is not a health check for an external provider, database, MCP server, or vector store. Run a representative prompt—or invoke a safe tool directly—to verify behavior against real dependencies.

Continue to Tools to inspect the capabilities behind the tool count.