Agents belong in the apps you already open
Most AI tools ask you to go somewhere new. The ones that stick show up where you already are, in the thread you were already reading.
Every AI product of the last two years has shipped the same thing: a box, on a page, behind a tab. It is a good box. It is also one more place to go, one more login, one more habit to build from scratch. Most people never build it.
We started Diali from the opposite end. Not "where should the agent live", but "where is the person already". For almost everyone we talked to, the answer was the same handful of messaging apps they check without thinking.
The tab you never open
There is a quiet failure mode in AI tooling that nobody puts in a changelog. The product works, the answers are good, and usage drops to zero in three weeks. Not because it got worse, but because using it required remembering it existed.
A tool you have to remember to use is competing with your memory, not with the alternatives.
Messaging apps do not have that problem. They are already open, already on the home screen, already the place a notification lands. An agent that lives there inherits all of that for free. You do not adopt it, you just reply to it. That is why we support every major messaging app rather than one: the right app is whichever one you were going to open anyway.
What changes when the agent is in the thread
The interesting part is not convenience, it is context. When the agent sits inside the conversation, three things become natural that were awkward before:
- Forwarding. You can push a message, a photo or a voice note straight to the agent instead of describing it.
- Interruption. It can reach you when something needs a decision, in the same place you would answer a colleague.
- Continuity. The thread is the history. There is no separate session to resume and no context to re-explain.
None of these are features you can bolt onto a web app. They come from the medium. That is why Diali agents reach Telegram, WhatsApp, Discord, Slack, Mattermost and Matrix, plus SMS and phone calls, and the dashboard chat is there for when you are at a desk.
The part that is genuinely hard
Living in someone's messaging app is a privilege, and it comes with obligations. An agent there has to know when not to speak. It has to be reachable only by the people you allow. It has to identify itself as an AI to anyone it messages or calls, because the alternative is a trick.
Those constraints shaped more of the product than any model choice did. Being in the room is easy. Being worth having in the room is the work.
Stop reading about it, build one
Set up an agent, pick a channel, and have it working inside the app you already keep open.