Glossary

Definitions of key terms used throughout the botts.ai platform. A reference guide for understanding agents, knowledge bases, deployments, credits, roles, and other concepts.

A reference of key terms used throughout the botts.ai platform and this documentation.

A

Agent

An AI assistant you configure on the Agents page. An agent combines a System Prompt, linked Knowledge Bases, built-in and custom Tools, Conversation History, and Customer Memory. The agent page groups Configure, Deploy, the Testing pages (Tests, Predictions and History, see Testing) and the Insights pages (Chats, Forms, Usage). An agent goes live through one or more Deployments. The number of agents in your organization is limited by your Plan. See Agents.

Auto-Recharge

An optional billing setting that tops up your Credits automatically. You pick a Credit Package and a threshold in credits, below the package size; whenever your balance falls to the threshold, that package is charged to your saved payment method (the check runs every few minutes). Auto-recharge does not run during a free trial; it starts once the plan is paid. After three consecutive failed charges, auto-recharge is disabled and stays off until you explicitly re-enable it in the billing settings (re-enabling resets the failure counter). Updating the payment method alone does not turn it back on. Low-balance warning emails are only sent when auto-recharge is not enabled. See Team & Billing.

B

Backup Model

An optional second AI model on a Website or Internal Chat deployment. It serves traffic only when the primary model fails (see Failover). The backup must be a text model and must differ from the primary; picking a backup from the same provider is allowed but the form shows a reliability warning: a different provider is recommended. If the primary understands images and the backup does not, a message with an image attached does not fail over (the form warns about this). The backup is validated against your Data Residency Tier because it handles real traffic during an outage. Phone deployments do not support backup models. See Deployments.

C

Channel

The way a Deployment reaches its users. Currently selectable channels are Website (chat Widget, optionally with AI voice), Phone (voice calls on a real phone number), and Chat (Internal) (Internal Chat for your team). Telegram, WhatsApp, and Email appear in the channel selector but are not yet available ("Coming Soon").

Chat Memory

Your personal memory in the dashboard Chat: what you have shared about your role, projects and preferences, plus a short summary of what you have been working on, used by every model you chat with. It is on by default and visible only to you. Turn it off, delete single facts (Forget this) or clear it (Forget everything) under Settings → Chat memory. Not to be confused with Customer Memory, which an agent keeps about the people it talks to.

Chunk

A segment of text from a document, sized for optimal retrieval. When a document is processed, it is broken into chunks that are individually indexed. This allows the search system to find the most relevant section of a document, not just the most relevant document.

Connection

One external system your organization has wired up, listed under Integrations → Your integrations. A connection holds the URL and the credentials once, and contributes one or more Tools that agents can call. Two kinds exist: an MCP Server and an HTTP API (a base URL plus endpoints you define or import from an OpenAPI spec). A one-click integration is one that botts.ai prepares and you connect by signing in. A connection's credential is either shared organization-wide or connected per user, so each member acts as themselves in the external system. See Integrations.

Contact (End User)

A person who interacts with your agent through a deployment: a website visitor chatting in the widget or a caller on your phone line. Contacts are tracked in the agent's Chats page (Grouped view), where you can see each person's conversations and any facts the agent has remembered about them (see Memory). Contacts are not team members of your organization. See People & Memory.

Conversation

A single interaction session between a Contact and an Agent. A conversation contains all messages exchanged during one chat session or phone call. Conversations are listed in the agent's Chats page.

Conversation History

An agent feature (toggle in Configure) that lets the agent see recent messages from previous conversations with the same Contact on the same channel. Contact identities are per channel: a website visitor, a phone caller, and an internal chat user are separate contacts even if they are the same human, so history does not carry over between channels. You can set how many past messages are included. Unlike Memory, which stores distilled learnings, Conversation History passes raw recent messages to the agent.

Crawl

The automated process of visiting web pages, extracting their text content, and indexing it in a Knowledge Base. Crawls follow links from the start URL and can be run manually or on a daily, weekly, or monthly schedule. Crawling costs 1 credit per page crawled. See Knowledge Base.

Credit Package

A fixed bundle of Credits you can buy on top of your monthly plan credits, priced in CHF, EUR, and USD and paid as a one-time charge. Purchased credits expire 12 months after purchase. Credit packages are also what Auto-Recharge charges automatically.

Credits

The usage currency of botts.ai. Credits are consumed by:

Each plan includes a monthly credit allowance that resets every billing cycle; unused plan credits expire at the end of the cycle. Purchased credits (see Credit Package) last 12 months; plan credits are always spent first. When your balance reaches zero, new chat messages (website widget, internal Chat, Try Chat) are refused, widget voice and phone calls are rejected (calls with a spoken message), crawls, file uploads and test runs are blocked, and scheduled runs are skipped.

D

Data Residency Tier

An organization-wide setting (labelled Server Location on the Organization → General page) that restricts which AI models your organization can use, based on where data is processed and which company runs the servers:

Switzerland and the EU both count as "European." Model pickers only list compliant models, and saving a non-compliant model is rejected. The tier also applies to the website widget's AI voice (GPT Live 1, hosted in Sweden by a US company): under the two Swiss tiers and European server by European company, the voice option is unavailable to visitors. If you tighten the tier, existing deployments on now non-compliant models keep running but are flagged with a warning badge on their agent. Knowledge-base storage and platform infrastructure are hosted in Switzerland for every organization, regardless of tier. See Deployments.

Deployment

A live instance of an Agent connected to a specific Channel, managed in the agent's Deploy tab. Each deployment has its own AI Model, optional Backup Model, and channel-specific settings. Website deployments additionally get a unique API key that authenticates the embedded widget; internal chat and phone deployments have none: internal chat reaches signed-in members, and phone calls arrive over the telephony network, where a widget key plays no part. An agent can have multiple deployments (for example, one on your website and another on your phone line), and all of them share the same agent configuration and knowledge bases. Your plan limits how many active deployments you can have; deactivating a deployment with its toggle frees up headroom. See Deployments.

E

Embedding

A mathematical representation of text (a vector) that captures its semantic meaning. Embeddings enable semantic search: finding content by meaning rather than exact keyword matches. botts.ai generates embeddings on Swiss infrastructure.

End User

See Contact (End User).

Expectation

The required part of a Saved Test: a plain-language description of what a good next reply must do, such as the facts it has to state, its tone or a length limit. It is read only by the evaluator, never by the agent, so it does not change how the agent replies. Facts can only be checked when the expectation itself supplies them. See Testing.

F

Failover

The automatic switch from a deployment's primary model to its Backup Model when the primary fails. Failover triggers on provider server errors (5xx), rate limits (429), connection timeouts and network errors, a model that sends no first token within 30 seconds, the case where the provider reports the model itself as invalid or decommissioned, and an empty generation (a successful response carrying no reply, which is retried before it fails over). A conversation too long for the primary's context window is also served by a backup with a larger one. After three failures in a row a model is skipped for a while (about 12 minutes): requests go straight to the backup, or, without one, end quickly with "The selected model isn't responding right now." Failover does not trigger on other client errors, after the agent has already started streaming a reply in a live chat (scheduled runs can still switch mid-turn), or after a side-effecting tool (such as a form submission or HTTP POST) has already run in that turn. Failover applies to text channels only.

Form / Lead Capture

A built-in agent tool (toggle in the Configure tab) that lets the agent collect structured data from users: contact details, support tickets, booking requests. You define one or more forms with named fields; during a conversation the agent gathers the values naturally and submits the form. Submissions appear in the agent's Forms page, and you can optionally receive an email notification at your organization's notification email address for each submission.

H

Hybrid Search

A search method that combines semantic search (meaning-based) and full-text search (keyword-based) to deliver comprehensive results. This is the default search mode used by agents and by the knowledge-base test search.

I

Integration

The umbrella term for what your agents can reach outside botts.ai, managed on the Integrations page (Build section of the sidebar) with three tabs: Your integrations (the Connections shared by all agents in your organization), Browse (one-click integrations and a catalogue of servers and APIs to add) and Secrets (encrypted credentials those connections reference). See Integrations.

Internal Chat

The Chat (Internal) channel: a deployment that makes an agent available to your own team members inside the dashboard, via Chat in the sidebar. Each agent can have at most one Internal Chat deployment. Access can be limited to the whole organization, to a minimum Role, or to an invite list. Options include a Chat Picture, a Description shown when a chat starts, a Show Tool Use toggle that controls whether users see which tools the agent is using, and a File upload toggle that lets users attach documents.

K

Knowledge Base

A collection of information that an Agent uses to answer questions, managed under Knowledge in the sidebar. Knowledge bases are built by adding Sources: crawled websites or uploaded files. Content is extracted, chunked, embedded, and indexed for retrieval. One agent can link multiple knowledge bases, and one knowledge base can be shared by multiple agents. All knowledge-base data is stored in Switzerland. See Knowledge Base.

M

MCP Server

A remote server speaking the Model Context Protocol, an open standard for exposing tools to AI assistants. Rather than describing endpoints yourself, you point botts.ai at the server and it discovers what the server offers; you then import the tools you want. Imported tools are namespaced servername__toolname. If the server later changes its tool list, botts.ai shows a diff and keeps running your previously approved set until you re-approve, so a server cannot silently redefine what your agent does. See MCP servers.

Memory

The Customer Memory feature (toggle in the agent's Configure tab, off by default). After each conversation, the agent extracts learnings about the person (preferred name, communication style, standing instructions) and applies them automatically in future conversations. Memory is stored per person and per agent. A retention setting controls how long learnings are kept: Keep indefinitely (default), or 30, 60, 90, 180 days, or 1 year. You can review and delete remembered facts in the agent's Chats page: switch to the Grouped view and select a person. For the memory of your own dashboard chats, see Chat Memory. See People & Memory.

Model

An AI language model that powers a deployment. Every deployment selects a primary model; Website and Internal Chat deployments may add a Backup Model. The model picker shows each model's provider, hosting location, speed and intelligence ratings, and credit rates. Text models are billed per token (rates shown per ~1,000 words as an approximation); the voice model GPT Live 1 is billed per minute (shown as "Voice N credits / min"). Models that understand attached images carry an Images badge. Phone deployments require a voice model; all other channels use text models, and a website widget's optional AI voice always runs on GPT Live 1. Which models are available depends on your Data Residency Tier.

O

Organization

A shared workspace on botts.ai where team members collaborate. Organizations contain agents, knowledge bases, integrations, and team members with Roles. The Plan, credit balance, and Data Residency Tier are all organization-level. All resources are scoped to an organization.

P

Page

The unit used for knowledge-base size limits: one page = 2,000 characters of stored extracted text. A document's page count is its character count divided by 2,000, rounded up. The page limit covers crawled and uploaded content combined, across all knowledge bases in your organization.

Plan

Your organization's subscription tier, which sets the monthly credit allowance and resource limits:

The Personal plan includes the internal AI chat workspace only (no external channel deployments). Personal and Business start with a free trial (1 month and 14 days respectively, with 150 and 1,500 starter credits), no credit card required, one per person. You can also choose Pay now instead, without the trial to get the full monthly credits right away and keep the trial unused. The legacy Free and Lite plans are no longer offered to new customers. See Team & Billing.

R

Role

The permission level of a team member within an organization. The hierarchy is Member < Builder < Admin < Owner; each role includes everything below it.

Roles also drive role-based deployment access: a deployment can require a minimum role (e.g. Builder or Admin) to use it.

S

Saved Test

A conversation that ends with a customer message, stored on an agent together with an Expectation and optional tags. Run All replays every saved test of an agent against its current configuration on a chosen platform, model and tool set, and grades each reply as Passed, Failed or Couldn't assess. Up to 200 per agent; created by hand, from Try Chat, from a customer conversation, or drafted on the Predictions page (Predict Future Chats). See Testing.

Scheduled Task

A personal recurring agent run, managed under Scheduled in the sidebar. It runs a deployed agent or a general-purpose model on a daily, weekly, or monthly rhythm, optionally from a start date (Start on) and until an end date (Stop after), or just once on a chosen day, and delivers the result by email or into your chat, marked unread. Tasks belong to the account that created them, not the organization, so no one else sees them. Every failed run is emailed to you, and three consecutive failures pause a task automatically; a run with no credits available is skipped rather than queued. See Scheduled Tasks.

Secret

An encrypted credential (API key, token, password) stored on the Integrations → Secrets tab. Secrets are referenced in connection URLs, headers, auth fields, and fixed parameter values with the syntax $secret:key, resolved at call time. Values are encrypted at rest and never returned by the API; to rotate a secret, delete and recreate it. Secrets are organization-scoped by default. Two narrower scopes exist for per-user connections: your own personal secret, and one an admin stores on a named colleague's behalf. Use organization scope for anything an agent calls on a public channel. See Integrations.

Semantic Search

A search method that finds content based on the meaning of a query, not the exact words used. For example, a search for "opening hours" would also find content that says "business hours" or "when we are open."

Source

An origin of content in a Knowledge Base. A source is either a website (a URL that gets crawled, with per-source crawl settings and an optional schedule) or a file (an uploaded .pdf, .doc, .docx, .txt, or .md document, up to 100 MB per file). Each source contributes documents and chunks to the knowledge base; deleting a source deletes all of its indexed content. See Knowledge Base.

System Prompt

The instructions that define an Agent's personality, behavior, and rules, edited in the agent's Configure tab. The system prompt tells the AI model who it is, how to communicate, what topics to handle, and what boundaries to respect.

T

Three I's

The framework that describes the three pillars of every AI agent on botts.ai: Information (Knowledge Base), Intelligence (Agent configuration and reasoning), and Interaction (Deployments and channels).

Tool

A capability that an Agent can invoke during a conversation. Built-in tools include knowledge search (enabled automatically when knowledge bases are linked), Advanced Knowledge Base Browsing (lets the agent list and read full documents), Web Search, and Form / Lead Capture. Appointment booking (find free slots and book in a Cal.com calendar, in chat and on voice) is set up by botts.ai on request; see Agents. In the dashboard Chat the assistant can also run code in an isolated sandbox to analyse spreadsheets and answer with charts and downloadable files. Tools from a Connection let the agent call external systems, for example to check stock or look up an order. The agent decides when to use a tool based on the conversation and the tool's description. Each tool is attached per agent, and every attached tool adds text to the model's prompt, so a short, deliberate list works better than a long one. See Integrations.

V

Vector Search

See Semantic Search. Internally, semantic search is powered by vector similarity: comparing the mathematical embeddings of the query and the stored content to find the closest matches.

W

Web Search

A built-in tool, off by default, that lets an agent search the live web and open individual pages to read them in full, citing its sources. Enabled per agent in the Built-in Tools section, and optionally restricted to (or excluded from) up to 10 domains. Billed per search plus per page opened. Web search queries are handled by a provider in France, which is disclosed on the deployment's Data Assessment panel whenever it is on. See Web search.

Widget

The chat window that a Website deployment embeds on your site via a one-line script tag containing the deployment's API key; it works on any website. Appearance and behavior (colors, fonts, welcome message, suggested messages, launcher style (Bubble, Bubble with message or Card with buttons), language, position, size, an optional AI voice so visitors can talk to the agent, and more) are customized after creation in an editor with a live preview, and applied with an explicit Save Changes. The key is public by design (it sits in your page source and only works for this deployment's widget). The dashboard cannot replace it: for a fresh key, create a new website deployment and swap the embed code. See Deployments.

Last updated on October 28, 2026