Platform Fundamentals

The InteractiveAI Platform organizes resources through a hierarchical structure designed for enterprise teams. Understanding this structure is essential for configuring access control, separating environments, and managing resources at scale.

Architecture Overview

The platform follows a two-tier hierarchy that mirrors how organizations actually operate. At the outermost level sits the Organization, which contains all billing, membership, and administrative settings. Within the organization, multiple Projects exist as isolated workspaces, each containing its own Interactive Agents, traces, prompts, datasets, secrets, and API keys.

This hierarchy serves a practical purpose: it allows enterprise teams to maintain strict isolation where needed while sharing administrative overhead where appropriate. A single organization can house dozens of projects without duplicating user management or billing configuration, yet each project remains completely independent in terms of data and access credentials. Every project owns its own set of Interactive Agents, and an agent built in one project is never visible from another.

---

Interactive Agents

The Interactive Agent is the central artifact of the platform. An agent is a configurable AI assistant that lives inside a project: you describe what it is, attach the behavioural building blocks it should follow, and deploy it as a running service reachable at its own URL. Every project hosts one or more Interactive Agents, and most of the resources in a project, whether prompts, glossaries, policies, routines, traces, or evaluators, exist to define how those agents behave.

Two parts make up an agent. The configuration describes the agent's behavior: its identity, the system prompt that defines its role, the context items it pulls from the project catalog, optional extra rules, and the external services it talks to. The deployment describes how the agent runs as a service: the runtime image, the public endpoint, environment variables, API keys, attached secrets, and an optional uptime schedule. The same configuration can be deployed multiple times in parallel, each instance with its own runtime settings.

Configuration and deployment live in different sections of the platform: Build → Agent Configuration for the behaviour, Deploy → Interactive Agent for the runtime. Both are described in detail in their respective pages.

Organizations

An Organization represents the top-level container in InteractiveAI. It typically corresponds to a company or business unit and establishes the boundary for billing, membership, and administrative control. Every resource you create, every team member you invite, and every project you manage exists within the context of an organization. When you sign into the platform, you land at the organization level.

What You Will Find Inside an Organization

Five sections govern how your organization functions. However, only Interactive Agents are visible to everyone, while Team, Usage, Billing, and Settings are reserved for those with Owner permissions.

Interactive Agents

Interactive Agents is the landing view of every organization. It displays each project in your organization as a collapsible row, showing the project name and the number of agents inside. Expanding a row reveals the agents in that project, each with its current status. From here you can create a new agent with + New Interactive Agent, create a new empty project with + Create project, or jump into any project to work on its resources. Projects without agents still appear in the list, ready to be populated.

Team

Team is where you manage who has access to your organization and what they can do. Every member appears in a table showing their name, email, role, invitation date, and who invited them. You can modify roles directly from this view and invite new members using the button in the top right corner. The platform implements role-based access control with four levels:

• Owners have complete administrative control including billing.

• Admins handle operational management without billing access.

• Members create and edit resources but cannot modify protected configurations.

• Viewers have read-only access to everything.

Usage

Usage provides visibility into resource consumption against your plan limits. The screen displays your Current Plan information alongside metrics for users, projects, datasets, agents, evaluators, and prompts, the LLM Token Consumption with its Token Breakdown and the Storage Usage alongside metrics for projects, traces, datasets, data volume, users, and token consumption. Each metric help you understand costs and usage before they become surprises.

From here you can access directly to the Upgrade Plan section where you can unlock even more possibilities by upgrading your plan.

Billing

Billing handles the financial side of your subscription. Here you see your current plan and monthly cost, manage payment methods, and access your complete invoice history. The interface shows your payment card on file and provides options to upgrade your plan or edit payment details.

Settings

Settings covers organization-level configuration. The General tab lets you modify your organization's name and type, while the Danger Zone at the bottom contains the option to permanently delete your organization, an irreversible action that removes all projects, traces, prompts, datasets, and associated data.

Key Characteristics

A user account can belong to multiple organizations, which proves useful for consultants working with several clients, agencies managing multiple brands, or employees working across subsidiaries. When you switch between organizations using the dropdown in the top navigation, the available projects and resources change accordingly because each organization maintains complete isolation from others.

Organization administrators control membership and permissions for all users within their organization. However, API keys and secrets are scoped to individual projects rather than to the organization itself. This means a developer can have access to your organization but still need project-specific credentials.

---

Projects

A Project serves as the primary unit of resource isolation within InteractiveAI. All operational data belongs to a single project: agents, traces, prompts, datasets, secrets, evaluators, and dashboards. Nothing crosses project boundaries, which means you can grant a contractor access to a development project without any risk of them seeing production data in a different project.

What You Will Find Inside a Project

When you open a project from the organization view, you land on the Dashboard, which serves as the operational command center providing a real-time snapshot of project health. Here you see total counts for traces, sessions, observations, and scores alongside performance metrics like average latency and average session duration. The cost section breaks down model cost, model cost per trace, and per user, giving you immediate visibility into spending.

Above the Dashboard in the sidebar sits Copilot, currently in beta and available on Enterprise plans, an AI assistant that lets you operate the platform through natural language.

Below the Dashboard, the sidebar reorganizes into all the functional categories that reflect the platform's core workflow:

Build

Build houses the components that define how your AI system behaves. Agent Configuration is where you create, version, and edit the agents themselves, describing their identity, description, context, extra rules, and connections. Context provides a version-controlled repository for the building blocks agents reference at runtime, including policies, routines, glossaries, macros and prompts. Playground offers an interactive environment for testing prompts with different models and parameters before committing changes.

Connect

Connect manages external integrations and and sensitive credentials for your project. LLMs displays the catalog of models available through the InteractiveAI Router, showing providers, capabilities, context windows, and pricing. Secrets Manager provides secure storage for API keys, tokens, and other credentials that your workflows require. Knowledge Base and Integrations are coming soon.

Deploy

Deploy runs your agents and services in production. Interactive Agent lists every deployed agent in the project alongside its status and last update, and is where you bring a configuration to production, edit a live deployment, or roll back to a previous revision. Service handles custom containerized workloads through the InteractiveAI CLI. Monitor displays real-time operational dashboards for your deployed workloads, including request volume, status code distribution, response time percentiles, and CPU and memory usage. Logs is where you view real-time container logs for your deployed agents and services.

Govern

Govern provides complete visibility into your AI system's behavior. This section contains Sessions, which group related traces into conversations or workflows; Traces, which capture end-to-end records of individual AI operations; Observations, which are the granular building blocks within traces recording each discrete step like model calls, tool invocations, or retrieval operations; and Scores, which attach quality measurements to your activity.

Improve

Improve contains the tools for systematically enhancing your AI system's quality. Datasets let you build collections of test cases to evaluate your application against consistent inputs. Annotations capture expert judgment through manual review, establishing ground truth for your evaluation infrastructure. Evaluators automate quality assessment using LLM-as-a-Judge, where a secondary model grades outputs against predefined criteria.

Report

Report surfaces insights about your platform usage and user behavior. Reporting delivers pre-configured visualizations of performance metrics, quality trends, and cost data designed for production AI monitoring. User Tracking connects LLM activity to individual end users, enabling you to debug issues, analyze consumption patterns, and understand how different segments interact with your system. Usage Control redirects to the organization-level Usage page for monitoring resource consumption.

Structuring Your Projects

Teams commonly structure their projects according to one of three patterns, each with distinct advantages.

• Agent-based structures create a separate project for each agent or closely related group of agents, such as support-bot, internal-search, and document-processor. This works well when each agent has its own team, its own release cadence, or its own set of policies, glossaries, and routines that don't need to be shared. Each agent gets a clean workspace with full isolation of context, secrets, and traces.

• Product-based structures create separate projects for distinct applications, such as customer-support-bot, internal-search, and document-processor. This works well when different products have different teams, different release cycles, or genuinely distinct operational characteristics. Each product gets its own complete workspace without interference from others.

• Team-based structures create separate projects for autonomous teams, such as team-alpha and team-beta. This pattern suits organizations where teams operate independently with their own tooling choices and deployment schedules. Each team owns their project entirely.

Some teams choose to create separate projects for production and non-production workloads, such as myapp-production and myapp-dev. This provides hard isolation between environments but requires managing multiple sets of API keys. Alternatively, you can use a single project and distinguish deployment stages using the environment attribute on traces, which allows filtering by environment in dashboards and trace views without the overhead of multiple projects.

Resources cannot be transferred between projects after creation. Consider your project structure carefully before deploying to production, as restructuring later requires re-ingesting historical data. A project rename is possible, but moving traces or prompts from one project to another is not.