Core Concepts
This page explains the building blocks of CubePlex. Understanding these will help you get the most out of the platform.
Organizations and workspaces
CubePlex uses a hierarchical structure:
Organization is the top-level account. It owns billing, provider API keys, and org-wide policies. Every user belongs to exactly one organization.
Workspaces live inside an organization. Each workspace is an isolated collaboration space with its own conversations, skills, memory, and MCP tool grants. A user can be a member of multiple workspaces.
Roles control what you can do:
| Scope | Roles | Can do |
|---|---|---|
| Organization | Owner, Admin, Member | Owner/Admin: manage providers, models, members, org settings. Member: use assigned workspaces. |
| Workspace | Admin, Member | Admin: manage workspace settings, tools, skills. Member: chat, use tools. |
Conversations
A conversation is a message thread between you and an AI agent. Each conversation is tied to a specific model (which you can change mid-conversation). The agent can:
- Respond with text and thinking (extended reasoning).
- Make tool calls to external services via MCP connectors.
- Execute code in a sandboxed environment.
- Generate artifacts — files, websites, images, or other deliverables.
You can attach documents, images, and code files to your messages. Conversations are saved automatically and appear in the sidebar history.
Skills
Skills are packaged capabilities you install to extend what the agent can do. Think of them as plugins that give the agent new knowledge or behavior patterns.
Three sources of skills:
- Built-in — ship with CubePlex.
- Org-uploaded — your organization creates and shares custom skills.
- Remote registries — community skills hosted on registries like skills.sh.
You discover and install skills from within a conversation or from the Skills page in the workspace sidebar. Once installed, the agent can use a skill whenever it is relevant to the conversation.
See the Skills guide for details.
Memory
Memory lets the agent remember information across conversations. CubePlex uses a three-tier system:
| Tier | Visibility | Example use |
|---|---|---|
| Personal | Only you, across all your workspaces | "I prefer Python over JavaScript" |
| Workspace | All members of the workspace | "Our API uses snake_case for JSON keys" |
| Organization | All members across all workspaces | "Company name is Acme Corp, founded 2020" |
Memory types describe what is being remembered:
- preference — how you like things done.
- project_fact — a fact about a project or codebase.
- procedure — a step-by-step process.
- correction — something the agent got wrong that you corrected.
- decision — an agreed-upon decision.
- org_policy — an organization-wide rule.
The agent reads relevant memories automatically at the start of each conversation. You can also view, edit, and delete memories manually.
See the Memory guide for details.
MCP tools
Model Context Protocol (MCP) connectors let the agent call external APIs — databases, SaaS products, internal services, and more.
How the tool lifecycle works:
- Templates — the catalog of available connectors (e.g., "GitHub", "Slack", "PostgreSQL"). Templates live in the catalog with visibility scoped globally (all orgs), per-org, or per-workspace.
- Enable — a workspace admin (or org admin via distribute) enables a template for the workspace, making it available for credentialing.
- Credentials — the person using the tool connects credentials (org-, workspace-, or user-scoped), depending on the connector's authentication mode.
- Runtime — credentialed tools are available to the agent during conversations.
Authentication modes:
- Static credential — you provide a fixed secret (an API key or bearer token), sent as an
Authorizationheader, a custom header, or a query parameter depending on the connector. - OAuth — the connector walks you through an OAuth flow.
- None — the connector needs no credential.
See the MCP Tools guide for details.
Artifacts
Artifacts are deliverables the agent produces during a conversation. They go beyond plain text — an artifact might be a downloadable file, a live website preview, a code snippet, an image, or a data table.
Artifacts are:
- Versioned — the agent can iterate on an artifact across multiple turns.
- Previewable — rendered inline in the conversation.
- Downloadable — save artifacts to your local machine.
Automation
Automation lets you run agent tasks without manual interaction.
- Scheduled tasks — run on a cron schedule, at a fixed interval, or as a one-shot at a specific time.
- Event triggers — a webhook fires and kicks off an agent run with the webhook payload as context.
See the Automation guides for details.
Next steps
- Workspace Setup — Configure a workspace for your team.
- Conversations guide — Learn the full set of chat features.