---
title: Realtime for AI Agents | Lynta
description: Give people and agents a live view of long-running work with State, Channels, and Tasks.
canonical: https://lyntahq.com/realtime-for-ai-agents
last-updated: 2026-10-06
---

# Realtime for AI agents

Agent work can take seconds or hours. It can call tools, produce partial results, wait for a person, or hand work to another agent. The original HTTP request may end before the user-facing work is done. Realtime infrastructure lets an application show progress, questions, decisions, and completion after that request returns.

Lynta carries realtime information around agent execution. Your application continues to run the agent and own its data.

## State, Channels, and Tasks

- [State](https://lyntahq.com/state.md) syncs backend changes such as job status, progress, and latest results to connected clients.
- [Channels](https://lyntahq.com/channels.md) carry ephemeral events such as messages, signals, notifications, or tool activity.
- [Tasks](https://lyntahq.com/tasks.md) represent a live unit of work and make its progress and requests visible to people and agents.

A Task can include Subjects that identify participants such as `user:123`, `agent:researcher`, or `agent:reviewer`. A research agent might ask a person which market to prioritize, wait for a response, and continue its own work afterward.

Lynta does not replace your database, execute an AI agent, choose its model, or steer its orchestration. Your backend decides what is authoritative and runs the workload; Lynta makes related changes and events live for connected clients.

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