Key Concepts
This page defines the core terms used across Flametree and these docs. Skim it once now, and come back whenever a term is unclear.
How the pieces fit together: inside your tenant, AI agents talk to customers over channels. Agents ground their answers in knowledge sources, record every dialog as a session, fill in conversation results, and hand over to human agents when needed. Campaigns start outbound conversations at scale.
Tenant
A tenant is your organization's workspace. It contains everything your team configures and collects: AI agents, knowledge sources, channels, integrations, campaigns, customer data, and billing.
A tenant is created automatically when you register. Each user belongs to exactly one tenant.
User roles
A user is a person who works inside a tenant. Every user has one role that defines what they can see and do:
| Role | What it is for |
|---|---|
| Tenant Admin | Full access: build agents, manage knowledge and channels, invite users, manage billing |
| Tenant User | Testing access: try agents in the Playground |
| Tenant Viewer | Read-only access: open the Playground, browse knowledge sources, and review sessions and human-agent conversations without changing configuration |
| Tenant Operator | Live operations: handle escalated conversations in Human Agents, work with sessions, and monitor Analytics |
See the full permission matrix in Reference.
AI agent
An AI agent is an LLM-powered assistant that talks to your customers. You define who it is, how it speaks, what it must accomplish, and what knowledge it can use. Flametree offers three agent types in the Create new agent menu:
- Inbound Agent — replies to messages customers send first (website chat, Telegram, WhatsApp, email, inbound calls). Best for support and lead qualification.
- Outbound Agent — starts the conversation with your contacts, typically as part of a campaign. Best for outreach, reminders, and reactivation.
- Call Analytics — does not chat live; it analyzes recorded calls you import and extracts structured insights. Best for quality assurance.
Agents can work over text and voice channels, depending on the channels you attach.
Session
A session is one conversation between a customer and an agent (or a human operator). It stores the full message history, the parameters it started with, the results collected, and its status.
Sessions begin when a customer writes in or an agent reaches out, and end when the conversation is resolved or goes inactive. You review them on the Sessions screen.
For billing, usage is counted in billable sessions: each covers a 24-hour window. A single long conversation can count as several billable sessions. See Profile and Billing.
Channel
A channel is the pathway customers use to talk to your agent: the web widget on your site, WhatsApp, Telegram, email, voice telephony, and others.
You connect and manage channels in one place — Settings > Channels — and then attach a connected channel to an agent on the agent's own Channels tab.
Knowledge Source
A knowledge source is a reusable unit of knowledge an agent draws on to answer accurately. There are five types:
- FAQs — question–answer pairs for predictable, repeated questions.
- Plain Text — free-form reference text.
- Websites — links to public web pages.
- Files — uploaded documents.
- YouTube — public videos, transcribed for the agent.
Sources live in a tenant-wide library (AI Agents > Knowledge Sources), so you can connect one source to many agents and update it in one place.
Older tenants may also see Knowledge Bases — the previous knowledge system. New setups should use Knowledge Sources. If you still use Knowledge Bases, see the migration guide.
Conversation Result
A conversation result is a structured field the agent fills in during a conversation — for example the customer's name, phone, or selected service. You define which fields to collect on the agent's Conversation Results tab; the agent extracts them automatically as it chats. Collected results appear with the session and on the customer's profile in 360 View.
Campaign
A campaign is an automated outbound communication process: you upload participants, and a visual flow decides who gets contacted, when, on which channel, and what happens based on the outcome. You build campaigns in Campaigns > Flows with triggers, conditions, and actions — no code required.
Human agent and Copilot
A human agent (operator) is a person who takes over conversations the AI cannot or should not finish — for example when a customer asks for a human, or a handoff condition triggers.
Operators work in the Human Agents workspace, where they see escalated conversations, the customer's context, and Copilot suggestions — AI-drafted replies the operator can review, edit, and send.
Customer
A customer is the person on the other side of the conversation. Customers are not platform users — they never log in to Flametree; they chat through a channel. Customer profiles, history, and collected data live in 360 View.
When writing agent instructions, refer to the person the agent talks to as human — for example, "Collect the human's contact details." Language models handle this wording consistently. Use client, guest, or another term only if your business requires it.
Related pages
- Quick start — put these concepts into practice.
- Portal tour — where each concept lives in the UI.
- Reference — the full glossary and permission tables.