# Channels & continuity

> Agents are reached where teams already work; threads continue and approvals live in-channel.

Source: https://openap.org/docs/channels

An agent is only useful if a human can reach it, follow what it's doing, and step in. The usual failures: the
agent lives where nobody is (a separate web app instead of Slack or the ticket queue), there's no real thread
to follow (only a transcript the framework owns), and anyone who can type can drive it. OAP reaches agents
**where teams already work**, with durable threads and approvals that live in-channel.

[Video: Reached with an @mention in a real thread; the App Home tab is the agent's at-a-glance hub.](https://openap.org/media/channels-demo.mp4)

## How it works

A **channel** is a pluggable transport: Slack, a browser tab, a terminal, a webhook. The same agent is reachable
through any of them, and the conversation is a real, durable thread — it continues across turns, and everyone in
it sees the same plan, the same status, and the same approvals. In Slack you can even **@-mention the bot into a
thread that's already going**, bringing the prior discussion and the people in it into the session. Crucially, the approval cards a human acts on
render *in the channel*, next to the work, rather than in some separate console — and who is allowed to drive or
approve is an authorization question, not "whoever can type."

## Go deeper

- **[Slack](https://openap.org/docs/slack)** — the richest surface: the app-provisioning wizard, threads, in-channel interactions, and
  the App Home hub. (Most of the [Authorization](https://openap.org/docs/authorization) guides are shown through Slack.)
- **[Browser web chat](https://openap.org/docs/browser-chat)** — the built-in `/sessions` chat served by the web daemon.
- **[Local terminal](https://openap.org/docs/terminal-chat)** — a full-screen TUI (`oap agent chat`) with zero external setup.
- **[Continuity & interactions](https://openap.org/docs/continuity-interactions)** — how threads persist and how approvals, notices, and status render across kinds.

> **Security: How this addresses the OWASP Agentic Top 10**
>
> - [ASI09 — Human-Agent Trust Exploitation](https://openap.org/docs/owasp-top10#asi09) (substantial): the in-channel approval design — explicit confirmation on risky actions, an injection-isolated summarizer, and the approver derived from the session, not the sender.
> - [ASI07 — Insecure Inter-Agent Communication](https://openap.org/docs/owasp-top10#asi07) (partial): agent-to-agent messages travel only between delegating sessions, as typed, authorized directions over a Channel joining the two, with the sender authenticated by its own session subject.
