Anvia

Run a team ​

Register an existing AgentTeam with Studio, pick it in the Playground, and submit a task to the coordinator. Studio uses the team as configured; you do not need a separate development-only definition.

ts
import { Agent, AgentTeam } from '@anvia/core/agent'
import { Studio } from '@anvia/studio'

const researcher = new Agent({ id: 'researcher', model })
const team = new AgentTeam({ id: 'research-team', model, members: [researcher] })

new Studio([researcher, team]).start({ port: 4021 })

1. Select the team ​

The Playground target selector lists registered agents and teams. Select a team to open its team view at /playground/teams/:teamId. A Studio configured with only teams opens its first team automatically.

2. Submit a task and follow up ​

Submit a task to the coordinator and keep working while it runs: send follow-up prompts, watch member activity stream in, or stop the whole team. Team tasks are local to the current page, and stopping or leaving a task requests server cancellation independently of closing its stream.

The Messages panel shows sender, recipient, and delivery status for inter-agent messages (latest 500). The Members panel shows queued and active instances, including recursive children; select an instance to inspect its output, tool activity, run messages, and usage.

3. Approve as the requesting member asks ​

Approvals and questions surface as independent cards labeled with the requesting member, so several members can wait on input at once. Resolve them as the member requests; the team runtime enforces that only an application response counts as authorization.

4. Read the terminal state ​

Completed tasks stay visible until you start a New task — a new task can begin immediately after the result arrives. Completed tasks do not appear in saved agent sessions; team runs use ephemeral conversations, matching the HTTP team run lifecycle.

Built for Anvia.