---
title: Realtime Tasks for Humans and AI Agents | Lynta
description: Lynta Tasks represent live work people, agents, and services can follow and coordinate around.
canonical: https://lyntahq.com/tasks
last-updated: 2026-10-06
---

# Realtime Tasks

A Task is a live unit of work. Agent work can continue after the request that started it has returned; a Task gives an application a unit to follow while that work progresses, pauses for input, or completes.

A Task can expose status, progress, steps, questions, responses, completion, failure, or cancellation to an application experience. Lynta coordinates realtime information around the task; your service still runs and controls the work.

## Subjects identify participants

A Subject identifies a participant in a Task. It can represent a person, an agent, or a service. Subjects belong to the Task model; they are not a separate top-level primitive.

- `user:123` can represent a person using the application.
- `agent:researcher` can represent an agent performing part of the work.
- `agent:reviewer` can represent another agent reviewing or extending the work.
- `service:payments` can represent a participating service.

For a human-in-the-loop workflow, a research agent can show that it is analyzing sources, ask which market to prioritize, and wait for a person to answer. Your agent service continues its own work after the application records the response.

Lynta does not execute an AI agent, choose its model, or replace your orchestration system. Your backend remains authoritative and runs the workload. Use [State](https://lyntahq.com/state.md) for backend changes and [Channels](https://lyntahq.com/channels.md) for ephemeral bidirectional events.

The public site does not publish a customer-facing callable SDK, product API reference, or product-action MCP server. Its separate read-only [documentation MCP service](https://lyntahq.com/mcp) only searches and fetches public pages. Direct integration questions to the [early access form](https://tally.so/r/jaOogE).
