Platform Overview

Bivariant is a platform for building and operating AI-powered applications. It provides the infrastructure, tooling, and services that teams need to go from idea to production — including AI agents, workflow automation, data management, external integrations, and multi-channel deployment.

What Bivariant does

Building production-grade AI applications requires more than a model API. It requires authentication, data storage, workflow orchestration, real-time messaging, third-party integrations, and multi-tenant isolation. Bivariant provides all of these as managed platform services.

The platform is organized around a few core primitives:

These primitives compose together: a blueprint defines agents that use flows, which call actions backed by integrations, and store data in collections. When a blueprint is installed as an app, each organization gets its own isolated runtime data.

Architecture overview

The platform is structured in layers:

Identity and tenancy

• Organizations — top-level tenants that isolate data and access
• Spaces — isolated environments within an organization for different projects
• Accounts — individual users (human or service accounts) with role-based permissions

Application layer

• Blueprints define what an application can do — its agents, flows, actions, integrations, data schemas, and screens
• Apps are per-organization installations of blueprints, with their own runtime data, connections, and service accounts

Execution layer

• Actions — the universal callable unit. Every operation (native service call, external API request, or flow execution) is invoked through the same interface
• Flows — visual workflows composed of steps, with support for sequential, parallel, and loop execution
• Assistants — the AI execution engine that powers agent conversations with LLM calls, RAG retrieval, and tool use

Data layer

• Collections — structured tables with typed fields, validation, relations, and views
• Datasets — large-scale data storage with partitioning and S3 integration
• Assets — file and media management with hierarchical organization
• Search Indexes — vector-based search for knowledge base and RAG

Communication layer

• Conversations — real-time messaging between humans and agents, with rooms, threads, and participants
• Channels — multi-platform delivery (web widget, WhatsApp, telephony)
• Events — distributed pub/sub system for internal and external events
• WebSockets — real-time event delivery to connected clients

Distribution layer

• Artifacts — packaged blueprints for cross-organization distribution
• Store — marketplace for publishing, discovering, and acquiring applications

Design principles

Everything is a blueprint. Every application on the platform — whether a third-party CRM connector, an AI assistant, or a platform-native service — is expressed as a blueprint composed of the same building blocks. This provides a uniform invocation model, composability across blueprints, and clean multi-tenant isolation.

Build-time and runtime are separated. Blueprints define application structure (agents, flows, schemas). Runtime data (records, conversations, credentials) is created per-organization when an app is used. This separation enables safe multi-tenant distribution.

One interface for all operations. Native platform operations and external API calls share the same CallAction interface. This means flows, agents, and other platform components can invoke any operation without knowing its underlying implementation.

Next steps

• Quick Start Guide — get your first application running
• Core Concepts — understand blueprints, apps, actions, and the data model
• Agents — configure AI assistants with knowledge bases and flows
• Flows — build automation workflows
• API Reference — complete API documentation