Getting started
Core concepts
The vocabulary used throughout these docs, and how the pieces relate.
Agent
The unit you build and publish. An agent has a model, a system prompt, its own knowledge base, its own tools, and its own channels — nothing is shared between agents except organization-level resources like credentials. You can run as many agents as you need (e.g. one for support, one for sales).
Grounded answers
Rather than answering purely from what the model already knows, the agent retrieves relevant passages from your knowledge base for each question and answers from those passages, citing them internally. This is what "grounding" means throughout these docs, and it's why what you put in Knowledge matters more than what you put in the system prompt. See Retrieval & grounded answers.
Knowledge source vs. chunk
A source is a document or pasted text you added. Behind the scenes it's split into chunks — smaller passages the agent actually retrieves and cites from. You manage sources; chunks are automatic, though you can preview them per source to sanity-check how a document was split.
Tool
An action the agent can call mid-conversation — usually an HTTP request to your own API or a third-party one (check an order status, look up a price, create a ticket). Tools can be added manually, generated from a plain-English goal with AI (a "toolkit"), picked from a small set of no-auth prebuilt tools, or connected via a remote MCP server. See Tools.
Channel
A place a customer can talk to the agent. Each channel is its own connection — you can have two Telegram bots pointed at two different agents, for example. See Channels.
Trigger
Automatic behavior that doesn't require the model to decide anything: a greeting shown when a widget conversation starts, a fixed reply for a keyword, or a scheduled action that fires on a timer (run a prompt, or run a tool) without a customer needing to say anything. See Triggers & automation.
Conversation
One session between a customer and an agent on a given channel. Conversations have a status (open with the AI, escalated, taken over by a human, or resolved) and are where transcripts, citations, and long-term customer memory live. See Conversations & Ops.
Organization vs. agent settings
Some things are configured per agent (model, prompt, knowledge, tools, channels, triggers). Others are configured once per organization and then reused across every agent — messaging account credentials, tool API-key credentials, AI provider connections, email sending domains, and data retention. Organization settings mostly require an admin role. See Organization settings.