Platform Concepts

Everything you need to understand how Kajoo Agentic is structured — workspaces, agents, skills, recipes, connections, and more.

This page explains the core concepts behind Kajoo Agentic. You don't need to read it before using the platform — but it's here when you want to understand what's happening under the hood, customize a workspace, or build one from scratch.


Workspace

A workspace is the container for a delivery engagement. It holds everything needed to complete a specific job — agents, connections, skills, tool permissions, and prompt templates — organized around a single use case or project.

One workspace typically maps to one migration engagement, one client site, or one delivery workflow. Your organization can have multiple workspaces, and they are fully isolated from each other.

What a workspace contains:

ComponentWhat it is
AgentsThe AI workers that execute the workflow
ConnectionsMCP servers that give agents access to external systems
ProjectsWorking areas with chat sessions where you interact with agents
SkillsKnowledge packages loaded into agents
Prompt templatesReusable prompts you can invoke from chat

Workspaces can be created from a recipe (pre-configured for a specific use case) or from scratch (manually assembled). See Getting Started for how to create one.


Agents

Agents are the AI workers inside a workspace. Each agent has a specific role and a defined set of tools, skills, and an LLM assigned to it based on the nature of its work.

Orchestrator agents vs. sub-agents

Kajoo workspaces use a two-tier agent structure:

  • Orchestrator agents coordinate the overall workflow, manage your conversation, and delegate tasks to sub-agents. You interact with orchestrators directly.
  • Sub-agents execute specific tasks — analyzing code, migrating a component, pushing content, converting a rendering. They work in the background, called by orchestrators as needed.

What each agent has

PropertyWhat it does
System instructionsDefines the agent's role, knowledge, and behaviour
MCP connectionsWhich tools the agent can call
SkillsCurated knowledge packages the agent draws from
LLM settingWhich model the agent uses, provisioned through Azure AI Foundry
Tool permissionsPer-tool human-in-the-loop settings (auto-approve or approval required)
Prompt templatesReusable prompts that can be invoked from the chat window

LLM assignment

Each agent is assigned the right model for its task. Lighter planning and orchestration tasks use smaller, faster models. Code migration, synthesis, and reasoning tasks use more capable models. The platform routes automatically — you don't manage model selection per request. All models are provisioned and governed through Azure AI Foundry.


Skills

Skills are curated knowledge packages attached to agents. They contain the DXP-specific context, rules, and patterns an agent needs to do its job correctly — things like how Sitecore XP templates map to SitecoreAI components, how to structure a headless SXA layout, or how to generate a TypeScript GraphQL adapter.

Skills are what make Kajoo agents DXP-aware rather than generic. A code migration agent without the right skills is just an LLM. With the right skills, it understands Sitecore rendering models, Content SDK 2.0 patterns, and JSS component structure.

When you create a workspace from a recipe, the appropriate skills are loaded automatically. You can add, swap, or remove skills from any agent in the workspace configuration.


Recipes

A recipe is a pre-configured workspace template for a specific delivery use case. When you create a workspace from a recipe, Kajoo automatically provisions:

  • The full agent team for that workflow
  • All required MCP connections
  • The appropriate skills for each agent
  • Human-in-the-loop tool permissions
  • A setup wizard to connect your environments

Recipes encode hard-won delivery knowledge — the agent structure, tool permissions, phase sequencing, and guardrails for a particular type of work. You don't assemble this from scratch; you start from a recipe, connect your project, and begin working.

Available recipes:

RecipeWhat it delivers
XP → SitecoreAI MigrationFull Sitecore XP to SitecoreAI (formerly XM Cloud) migration — code, content, and presentation

You can also build a workspace from scratch for custom workflows, or use an existing recipe as a starting point and customize from there. See Recipes for the full library and roadmap.

Recipe Builder

A Recipe Builder is coming to Kajoo Agentic — allowing teams to remix existing recipes, customize agent configurations, and compose entirely new recipes without writing infrastructure code.


MCP Connections

MCP (Model Context Protocol) connections are the integrations that give agents access to external systems. Each connection exposes a set of tools that agents can call — reading from a repository, pushing content to a CMS, querying a GraphQL endpoint, creating a branch.

Connections are registered at the workspace level. All agents in a workspace can be granted access to any connection in that workspace.

Connections in a recipe workspace are pre-configured. When you create a workspace from a recipe, the required connections are provisioned automatically. You connect your specific environments (repos, CMS instances) through the setup wizard.

You can add custom connections to any workspace — for example, an Atlassian MCP server to link Jira tickets to migration work, or any other remote MCP server your workflow requires. Go to Settings → Configure Workspace → MCP Servers to manage connections.


Human-in-the-loop

Human-in-the-loop (HITL) is the control mechanism that determines how much agents do autonomously vs. how much requires your explicit approval.

Every tool an agent can call has a HITL setting:

SettingBehaviour
Auto-approveThe agent executes the tool call without asking. Used for low-risk read operations.
Approval requiredThe agent pauses and presents the proposed action for your review before executing. Used for write operations, commits, content pushes, and other high-impact actions.

In recipe workspaces, HITL settings are pre-configured based on the risk level of each tool. You can adjust any tool's setting at any time in the agent configuration panel.

This is what makes the reviewer model work in practice — agents propose and execute routine work automatically, but pause at every meaningful decision for your input.


Projects

Projects are the working areas inside a workspace. A workspace can have multiple projects — for example, one per client site or one per migration engagement.

Each project has its own chat sessions where you interact with agents directly. A project can have multiple chat sessions running simultaneously, which is how parallel workstreams (like running code migration and content migration at the same time) are supported.

⚠️

Don't run two code migration sessions in parallel within the same project. Code migration sessions share the same sandbox — concurrent sessions cause conflicts. Parallel sessions are safe for independent workstreams (e.g. code migration + content migration).


Agent Units (AU)

Agent Units (AU) are the usage and billing unit for Kajoo Agentic. AU is an outcome-based measure — it reflects the delivery work completed, not raw AI consumption (tokens, runtime minutes, or infrastructure).

You don't need to think about AU while working. It's reported in your workspace usage dashboard. Pricing and AU details are available from your Kajoo contact or on the Kajoo pricing page.


What's next

Ready to create a workspace?See Getting Started
Start the XP → SitecoreAI migration?See XP → SitecoreAI Migration