Kai Overview
Go Up to Kai
Kai is Embarcadero’s agentic AI platform for Delphi, C++Builder, and RAD Studio, bringing automation into your development workflow.
Contents
Overview
Kai transforms RAD Studio into an agentic development environment, pairing best-in-class code suggestions with an AI agent that can generate, build, fix, and evolve your applications. More than a coding assistant, Kai acts as a collaborator: generating code and UIs, compiling projects, resolving errors, and executing tasks inside and outside the IDE using MCP-powered workflows.
From accelerating new applications to modernizing and refactoring existing codebases, Kai serves as an agentic companion, assisting with complex work and adapting to your process. Built on a flexible BYOK model, Kai supports leading LLMs, including GitHub Copilot, OpenAI, Claude, Gemini, and local models, giving developers control over performance, cost, and data privacy.
Why Kai
Kai allows users to stay in flow, to work entirely within RAD Studio. Easily turn prompts into working code, builds, and fixes without external disruptions.
Kai offers AI-powered inline autocomplete and rich code suggestions to improve coding and boost productivity. This easy-to-use tool lets users generate UIs, working prototypes, or even apps by describing an idea, then have Kai create a ready-to-use platform or a functioning application.
Users with existing applications can use Kai to modernize the code and UI, and help refactor existing codebases. Using the best model for specific tasks and leveraging new language and framework features.
When it comes to security measures, Kai does not use any proxy layer via Embarcadero services, nor does it have any mandatory cloud dependencies; your code stays under your control through local storage and direct model communication. In addition, Kai can be run offline with local LLMs for maximum IP protection and compliance.
Kai Architecture
Kai features an integrated AI agent that interacts directly with the development environment. Kai connects to leading AI models through a flexible BYOK (Bring Your Own Key) approach, supporting providers such as GitHub Copilot, OpenAI, Anthropic, Google, and various local LLMs.
It interfaces with local CLI engines via an MCP (Model Context Protocol) layer, enabling not only interaction with the IDE itself but also integration with external MCP servers. The Kai tool is installed as an IDE package that provides IDE integration and a conduit for LLM integration. The following list of CLI tools is used:
- github/copilot-language-server
- github/copilot
- openai/codex
- google-gemini/gemini
- Codex and Cloud Agent ACP interfaces
Each of the tools mentioned above offers integration with multiple models. The solution can also integrate local LLMs. The image below displays how Kai integrates with RAD Studio and the local LLMs.
MCP Support
MCP support is an integral part of the solution. When the tool makes a request, it also provides endpoints in the IDE so that external AI engines can interact with the IDE in multiple ways, with instructions detailed in the AGENTS.md file, passed along with each request.
MCP Agentic Workflows
Kai includes an MCP server interface that lets users integrate other MCP tools. The MCP server understands the file system, allowing users to read, write, and manage project files without extra configurations.
Kai provides an MCP interface for RAD Studio that automates compile-fix workflows by enabling the AI engine to compile an application, read error messages, attempt fixes, and recompile automatically. Optionally, use your own MCP server’s version control capabilities to create and manage pull requests directly from the IDE.
Feature Summary
Kai integration to RAD Studio introduces an agentic experience within the IDE, combining a dockable chat interface, inline code completion, and advanced code suggestions that enhance the RAD Studio editing experience.
Kai offers inline autocomplete, which augments standard code completion, while expanded suggestions provide alternative code snippets and refactoring options. It also offers Ghost mode, which lets users add code directly in the editor and accept it with a single keystroke. Use the Chat window to request coding assistance on creating UI forms, modifying source code, compiling projects, analyzing compiler output, or automatically resolving errors and warnings.
With model selection available per project or even per prompt, Kai offers a highly customizable AI experience, bringing powerful automation and intelligent assistance directly into the RAD Studio development lifecycle.
See all Kai features in more detail in the Kai Features page.