Glossary of Chat Thing terms

Chat Thing uses a few words of its own, and other products use different words for the same things. Each entry gives a short definition, the other names people search for, and the page with the detail. Names in code are what you'll see in the API, the MCP tools or the JavaScript SDK.

Bots and how they answer

Bot

One AI agent with its own instructions, knowledge, power-ups and channels. Also called an agent, AI agent, chatbot or assistant; the app says bot. In the API and MCP tools it's identified by botId. See How Chat Thing works.

Prompt (instructions)

The instructions that tell your bot its role, tone, rules and what to do when it can't help. Older bots show System message and Initial prompt instead, and the API, SDK and MCP tools call it systemMessage. See Write your bot's instructions.

Model

The AI model that writes your bot's replies. You can choose one on the Standard plan and above. See Models.

Creativity (temperature)

How varied your bot's wording is. Lower means more consistent answers that stick to your content. Called temperature in the API and MCP tools. See Settings reference.

Knowledge (data sources)

The content your bot answers from: your website, files, Notion, YouTube, RSS feeds or text you type in. Each connected source is a data source. People often call adding knowledge "training" the bot, but Chat Thing doesn't retrain the AI model: it searches your content for each question. See Knowledge.

Document (row, chunk)

A data source is made of pages, files or entries, which the API and MCP tools call rows. When a source syncs, each one is split into smaller pieces called documents (sometimes called chunks), and those are what the bot searches. See Retrieval settings.

Sync

Reading a data source's content so the bot can search it. A re-sync picks up changes; Auto-syncing does it every day, week or month on the Standard plan and above. See Keep your bot's knowledge up to date.

Retrieval settings

The settings that control how your bot searches its knowledge: Document relevance, Max context documents, Enhanced knowledge retrieval and others. Also called RAG settings. See Retrieval settings.

Sources (citations)

Links under an answer showing which pages it came from, when Include sources for responses is on. See Retrieval settings.

Power-up (tool, action, function)

Something your bot can do during a chat, such as email your team, book a meeting or call your API. Other products call these tools, actions or functions. Parts of the API still use those words: the Chat API reports power-up activity as tool events and messages. See How power-ups work.

Client-side power-up

A power-up that runs in your web page rather than on Chat Thing's servers, registered with the JavaScript SDK's registerPowerUp(). See JavaScript SDK.

Display

A power-up that shows an answer as cards, a table, a chart, a map or a diagram instead of plain text. See Displays.

Scheduled task (task)

An instruction your bot carries out on a schedule, without anyone chatting to it, such as a weekly summary email. Shown on the bot's Tasks tab, on the Standard plan and above. See Scheduled tasks.

Test case and test run (evals)

A test case is a question plus checks its answer must pass; a test run runs all of a bot's test cases and grades the answers. Other tools call these evals. On the Enterprise plan, on the bot's Tests tab. See Test your bot's answers.

Channels and conversations

Channel

A place people talk to your bot: the web channel, WhatsApp, Slack, Discord, Telegram, email or the API. One bot can use several channels at once. See Channels.

Web channel (widget, bot page)

The channel for chatting in a browser. It gives you a chat widget (a chat bubble on your site, also called a chat launcher or live chat widget), an embedded chat for inside a page, and a bot page at its own link. Every bot has it, on every plan. See Website.

Pre-chat form (lead form)

A form visitors fill in before their first message, such as name and email. Its answers are saved with the conversation and the bot can see them. See Ask visitors for their details first.

Conversation (chat)

One thread of messages between someone and your bot. The app lists them under Conversations; the API and MCP tools often say chat (chatId). See Review customer conversations.

User data

Details about the person chatting, such as their name, email or account ID, passed in by your pre-chat form, the SDK's identifyUser or the API's userData. The bot sees them on every reply. See JavaScript SDK.

Handoff (escalation)

When the bot passes a customer to your team. The Talk to a human power-up takes the customer's email address, emails your team and marks the conversation Escalated. See Hand off to a human.

Human takeover (live chat)

A teammate stepping into a live web chat and replying themselves while the bot stays quiet, then handing it back. Standard plan and above, web channel only. See Human takeover.

My Chats

Your own private chats with your bots, kept separate from customer conversations. See My Chats.

Analytics

Topics, sentiment, common questions and recommendations worked out from a bot's conversations, on its Overview tab. Pro and Enterprise. See Analytics.

Topics

Groups of conversations about the same subject, found automatically or set by you. Part of analytics. See Analytics.

Tokens and plans

Message tokens (credits)

What your bots use when they reply, measured in tokens (roughly three-quarters of a word each). Other products call these credits or message credits. Each plan includes a monthly allowance. See Message tokens.

Storage tokens

What syncing knowledge uses, based on how much content is processed. Only new or changed content counts. Each plan includes a monthly allowance. See Storage tokens.

Add-on

A paid extra that raises one limit, such as more message tokens or bots, without changing plan. See Add-ons.

Team

The account that owns your bots, plan and billing. A personal team is just you; an organisation team has members with roles such as Owner, Admin and Viewer. See Teams and roles.

Early access

A feature that's being rolled out to some teams before everyone. Its docs page has an Early access banner and isn't listed in the sidebar or search. To ask for it, email support@chatthing.ai. New features are announced in the changelog.

Building with Chat Thing

Otto

Chat Thing's built-in assistant. Click Ask Otto on a bot to change settings, add knowledge or test answers in plain words, or Ask Otto about this chat to find out why an answer went wrong. See Otto.

MCP (Model Context Protocol)

An open standard that lets AI assistants such as Claude, ChatGPT and Cursor use tools in other apps. Chat Thing's MCP server lets your assistant build and manage your bots. See AI assistants (MCP).

WebMCP

An experimental browser standard that lets a web page offer tools to AI agents. The Chat Thing widget can pass those tools to your bot as client-side power-ups. See WebMCP.

JavaScript SDK

The window.chatThing object that comes with the chat widget, for controlling the widget and passing user details and page context from your site. See JavaScript SDK.

API channel

The channel that turns on the REST API for a bot, so your own app can chat with it and manage its manual knowledge. Standard plan and above. See API overview.

Webhook (hook)

A URL that Chat Thing calls when something happens, such as a new or escalated conversation, or a URL you call to start a sync. The API, MCP tools and the duplicate-bot dialog call these hooks. See Webhooks.