docs

Getting started

Welcome

What this platform does and how the pieces fit together.

What this platform does

This platform lets you create AI-powered customer support agents that answer from your own knowledge — your docs, policies, and product info — rather than the model's general training. Each agent can also call tools (to look up an order, check a price, or trigger an action in another system) and be published to the channels your customers already use: a widget on your website, Telegram, WhatsApp, Instagram, and email.

One agent, configured once, answers consistently everywhere you publish it.

The pieces

  • Agent — the assistant itself: a name, a model, a system prompt, and the settings that control how it behaves.
  • Knowledge base — the documents and text you give the agent to ground its answers in. This is where facts should live, not the system prompt.
  • Tools — actions the agent can take mid-conversation: call an API, run a lookup, or fire an integration. Built manually, generated from a goal with AI, or connected via a remote MCP server.
  • Channels — where customers actually talk to the agent: your website widget, Telegram, WhatsApp, Instagram, or email.
  • Triggers — automatic behavior: a greeting when a chat starts, a canned reply for a keyword, or a scheduled action that runs on a timer.
  • Conversations & Ops — where you read what customers asked, take over a conversation from the AI when needed, and see how the agent is performing.

Who can do what

Every teammate you invite is either a member (can create and edit agents, knowledge, tools, channels, and triggers) or an admin (everything a member can do, plus destructive actions, managing teammates, and organization-wide settings like credentials and data retention). Team invites and roles are managed from the account menu in the app header, not inside this documentation's settings pages — see Team & roles.

Where to go next

If this is your first time here, the Quickstart walks through creating an agent, giving it knowledge, and publishing it to a channel in a few minutes. If you'd rather understand the model first, read Core concepts.