Get started

Core concepts

The six words the portal uses everywhere, workspace, credits, voice, dialect, agent and channel, each in a few sentences with the page that explains it in full.

Six words come up on every page of the portal and every page of these docs. Read this page once and you will know what each one means, which of them you own and which you share, and where to go when you need the detail. Each section ends with the page that covers the idea in full.

Workspace

A workspace is the container everything you build lives in: agents, knowledge, tools, conversations, speech generations, transcripts, API keys and the subscription. One is created for you when you sign up, named after you, with you as its Owner.

Your name, sign-in method, two-factor setting, theme, language and notifications are yours, not the workspace's.

Workspaces gives the full model.

Credits

Credits are how usage is paid for, from one balance. A generation, a transcription or a cleaning job costs credits in proportion to the size of the job, taken from the workspace's balance before the work runs and returned if the work fails. An agent call is charged for the time it is connected and for its AI usage. The balance belongs to the workspace, not to you: a new workspace starts on a trial with 60,000 credits and no time limit, a paid plan grants its credits every month, and unused plan credits roll over when the same plan renews.

Subscription & billing in the user menu opens the page that shows the balance and where it comes from.

Credits gives the price of every service, Credits, top-ups and rollover explains the plan credits, extra credits and when each ends, and Track your credit usage shows where to read what each job cost.

Voice

A voice is what speaks. The library under Voices holds every voice you can use, the same library in every workspace, and each voice has a name, a dialect and a style line, and most have a preview clip. You use a voice in two places: on Text to Speech, where it reads the text you type, and on an agent, where it reads what the agent composes during a conversation.

Browse the voice library shows the library page, and Choosing a voice explains what to listen for when a voice must narrate, and when it must hold a conversation.

Dialect

A dialect is the accent a voice speaks in, and every voice speaks exactly one. You never choose a dialect on its own: you choose a voice, and its dialect comes with it. Fusha is Modern Standard Arabic, understood across the region; the others, Egyptian, Saudi, Emirati and the rest, are the speech of one country. English voices sit in the same library.

Settle the dialect before you compare voices. A voice in the wrong dialect is the wrong voice, however good it sounds.

Dialects lists every dialect the library can carry and what the choice changes for a generation and for an agent.

Agent

An agent is one saved configuration that answers people on your behalf. It holds a persona (the instructions it follows and the first line it says), a model that thinks for it, one or more voices, the knowledge it may quote from, the tools it may call, and the channels it can be reached on. An idle agent costs nothing; each exchange in a conversation, one thing said and one answer, is a turn, and a conversation pays for its AI usage, and a call also for the time it is connected.

Build one from Agents in the sidebar, test it in the portal before anyone else talks to it, then connect it to a channel.

What is an agent describes every part, and What a turn costs shows how a turn becomes credits.

Channel

A channel is a place your customers already are, connected to one of your agents: a WhatsApp number, a Slack workspace, a phone number, a helpdesk account. Once connected, every message or call that arrives on it becomes a conversation with that agent, answered with the agent's persona, voice, knowledge and tools, and charged per turn like any other conversation. One channel is bound to one agent; one agent can carry many channels.

Web chat and web calls need no channel: they are the playground, opened with Test on the agent.

Channels overview lists every channel, what each one carries, and which one to launch first.

Worked example

Sara signs up. A workspace named Sara Ahmed's workspace is created with her as Owner and a trial with credits on it. She opens Voices, picks an Egyptian voice, and generates one sentence on Text to Speech; the credits for that generation leave the workspace balance. She builds an agent, gives it a persona and the same voice, tests it with Test, and connects a Slack workspace to it. The first message a colleague sends in Slack is a conversation on the agent, charged by the turn, and it appears on the agent's Conversations tab.

FAQ

Is a voice mine or the workspace's?

It depends. The library voices are the same in every workspace. A voice you clone is private to your workspace. What is the workspace's is the use you make of a voice: the generations it read and the agents it speaks for.

Do credits belong to me or to the workspace?

To the workspace. See Workspaces.

Does an agent need a channel?

No. An agent with no channel is still fully usable from Test in the portal, which runs the same agent a customer would reach. Connect a channel when you want people outside the portal to reach it.

Can one agent speak two dialects?

Yes, by carrying two voices, one per dialect. See Voices and language.