DocsTerminology
Definitions of key terms and concepts used across the Bivariant platform and documentation.
This page defines the terms used throughout the Bivariant documentation and platform interface.
Platform primitives
| Term | Definition |
|---|---|
| Blueprint | A versioned, reusable template that defines the structure of an application — its agents, flows, actions, integrations, data schemas, and screens. Blueprints are the build-time definition; they describe what an app can do. |
| App | An installed instance of a blueprint within an organization. Each app has its own runtime data (connections, records, conversations) isolated from other organizations using the same blueprint. |
| Action | The universal callable unit on the platform. Every operation — whether a native service call, an external API request, or a flow step — is represented as an action with typed input and output schemas. |
| Agent | An AI assistant defined within a blueprint. Agents have an identity (name, avatar), a default flow, and optionally a knowledge base for retrieval-augmented generation. |
| Flow | A visual workflow composed of steps organized in groups. Flows define multi-step automation logic with support for sequential, parallel, and loop execution. |
| Integration | A typed binding to an external API. Integrations define operations, authentication methods, event channels, and data types that become available as actions within a blueprint. |
Organization and access
| Term | Definition |
|---|---|
| Organization | The top-level tenant entity. All resources, users, and data belong to an organization. Organizations provide data isolation between different companies or teams. |
| Space | An isolated environment within an organization. Spaces separate projects, teams, or deployment stages. Apps are installed into spaces. |
| Account | A user identity on the platform. Accounts can be human users or service accounts (used by apps and agents for automated operations). |
| Role | A named set of permissions that can be assigned to accounts. Roles control what operations a user can perform within an organization. |
| Permission | A specific right to perform an action on a resource (for example, conversations:create or flows:update). Permissions are grouped into roles. |
| Service account | An account created for automated systems — specifically for apps and agents — that allows them to act within conversations and execute operations on behalf of the app. |
| API key | A token that provides programmatic access to the Bivariant API, scoped to an organization with configurable permissions. |
Data
| Term | Definition |
|---|---|
| Base | A logical container for related collections, similar to a database schema. Each base maps to an isolated data store. |
| Collection | A structured data table within a base. Collections have typed fields, validation rules, and support relations between tables. |
| Field | A column definition within a collection. Fields have a type (text, number, date, select, link, formula, etc.) and optional validation rules. |
| Record | An individual data entry (row) within a collection. |
| View | A saved configuration of filters, sorting, and field visibility applied to a collection. |
| Dataset | A large-scale data container with support for partitioning, multiple storage formats (CSV, JSON, Parquet), and S3-backed storage. |
| Asset | A file or media resource stored on the platform (images, documents, videos). Assets are organized hierarchically in folders. |
AI and knowledge
| Term | Definition |
|---|---|
| Assistant | The AI execution engine that processes agent conversations. It handles LLM calls, knowledge base retrieval, response streaming, and tool use. |
| Knowledge base | A vector search index associated with an agent. Documents are processed, chunked, embedded, and stored for semantic retrieval during conversations. |
| Search index | The underlying vector store that enables semantic search. Documents are indexed with embeddings for similarity-based retrieval. |
| Embedding | A numerical vector representation of text that captures semantic meaning. Used for similarity search in knowledge bases. |
| RAG | Retrieval-augmented generation. A technique where relevant documents are retrieved from a knowledge base and provided as context to the AI model before generating a response. |
Workflows and execution
| Term | Definition |
|---|---|
| Step | An individual unit of work within a flow. Steps can call actions, invoke assistants, wait for events, or branch based on conditions. |
| Step group | A container for steps that defines their execution pattern: sequential, parallel, or loop. |
| Edge | A connection between steps in a flow that defines execution order and data passing. |
| Task | A tracked execution record. Every action call, flow run, or assistant invocation creates a task with status tracking (created → running → completed/failed), input, output, and timing data. |
| Trigger | A condition that starts a flow. Triggers can be manual, event-based (for example, a webhook), or conversation-based (a new message to an agent). |
Integrations and connectivity
| Term | Definition |
|---|---|
| Connection | An authenticated credential set that links an app to an external service. Connections store OAuth tokens, API keys, or other credentials in encrypted storage. |
| Operation | An individual API endpoint defined within an integration, with typed parameters, input/output schemas, and authentication configuration. |
| Auth method | An authentication scheme defined in an integration (Basic, Bearer, API Key, OAuth2, or Custom). Auth methods specify what credentials are needed and how they are sent. |
| Event channel | An endpoint that receives events from external systems. Channels support webhooks (inbound POST), SSE (server-sent events), HTTP polling, and CloudEvents. |
| Event definition | A typed event specification within a channel, describing how to identify and extract event data from incoming payloads. |
| Data type | A shared JSON schema defined by an integration and available across the blueprint for consistent type references. |
Communication
| Term | Definition |
|---|---|
| Conversation | A threaded exchange of messages between participants (users and/or agents). Conversations maintain context and message history. |
| Room | A persistent container for conversations. Rooms can have multiple participants and support direct messaging with deduplication. |
| Participant | A user or agent that is part of a room or conversation. Participants with autoRespond enabled automatically trigger the agent's flow when messages arrive. |
| Channel | A platform-specific communication endpoint (web widget, WhatsApp, telephony) that connects external users to agents on the platform. |
| Presence widget | An embeddable JavaScript widget that adds a chat interface to any website, connecting visitors to a configured agent. |
Distribution
| Term | Definition |
|---|---|
| Artifact | A packaged, distributable unit — a published blueprint or notebook that can be shared across organizations through the marketplace. |
| Store | A marketplace where artifacts are listed, discovered, and acquired. Stores support categories, pricing plans, reviews, and order management. |
| Version | A specific release of a blueprint following semantic versioning (major.minor). Publishing a version freezes the current draft, deep-copies all components, and creates a new draft. |
Events and real-time
| Term | Definition |
|---|---|
| Event bus | The platform's distributed pub/sub messaging system, backed by NATS. All services publish and subscribe to events through the event bus. |
| Topic | A hierarchical event routing address (for example, events.{orgId}.conversations.conversation.message_posted.*). Topics support wildcard subscriptions. |
| Notification | A user-facing alert triggered by platform events. Notifications are delivered through WebSocket (real-time), push (mobile), or email channels. |
| WebSocket | A persistent, bidirectional connection used to deliver real-time events and notifications to connected clients. |
Quick Start
Get a working AI application running on Bivariant in under 15 minutes. This guide walks through installing an app, configuring an agent, building a flow, and deploying a web widget.
Core Concepts
Understand the foundational architecture of the Bivariant platform — blueprints, apps, actions, organizations, and the separation between build-time and runtime data.