Kai Versioni

Scopri le novità in ogni release, con dettagli su funzionalità, miglioramenti e problemi risolti.

gen
feb
mar
apr
mag
giu
lug
ago
set
ott
nov
dic
Kai 1.1.23 nov 2026
Funzionalità

Authentication Moves into the Chat

  • Change made to the way Kai handles authentication with AI agents. In previous versions, authentication status and login operations were managed through the IDE’s Tools Options dialog. With Kai 1.1.2, authentication takes place directly in the chat window.
  • For agents that support these operations, you can also access Reauthenticate and Logout commands from the context menu of the chat tabs.

Images and File Attachments

  • Added the ability to attach images directly to a Kai conversation.
    • For example, you can provide an image of a user interface mockup and ask an AI agent to help build the corresponding Delphi form.
    • Or you might share a screenshot of an application to discuss a UI change or a problem you are trying to resolve.
  • Drag and drop images into the chat or use the new attachment button.
  • The same mechanism also supports other types of files, including documents and audio files, provided the agent you are using supports these attachments.

More Chat Improvements

  • Collapsible reasoning blocks. Agent reasoning blocks are now collapsible and automatically collapsed, keeping the conversation more compact and easier to follow.
  • Permissions inside the chat. When an agent requests permission to perform an operation, the request now appears directly in the chat instead of opening a separate IDE message box.

Expanded MCP Integration with RAD Studio

  • The compileProjects operation now supports additional build parameters, allowing agents to specify:
    • Target platform, for projects supporting multiple platforms.
    • Build configuration, including Debug, Release, and custom configurations.
    • Build mode, including Make, Build, Check, MakeUnit, Clean, and Link.
    • Also improved the consistency of the evaluateExpression operation used during debugging.

Better Support for Local AI Engines

  • Tested and improved the integration with LM Studio and Ollama, focusing particularly on the behavior of local models when used for agentic operations.
  • Unlike a simple chat request, an agentic operation can involve multiple tool calls and take considerably longer to complete. Local models can also have very different performance characteristics depending on the model and the hardware being used.
  • This release includes an updated Native LM Studio ACP agent, with improvements to the configuration and tuning of parameters such as request timeouts and the maximum number of agent operation turns.

Additional Quality Improvements

  • One improvement concerns how the RAD Studio IDE detects changes made to source files by Kai agents. Improved the handling of files already open in the editor, including cases where agents modify those files directly through the file system.
  • Addressed several issues in the chat interface, including painting and layout problems, particularly when undocking and redocking the chat pane. The chat context menu now offers better options for copying conversation content.

Other fixes and Improvements

  • Correct display of tool names in permission requests when using non-English versions of the IDE.
  • Improved Codestral FIM support for inline code completion.
  • Support for restoring Codex sessions when working with mapped VirtualBox network drives, addressing UNC path mismatches.
Kai 1.1.125 set 2026
Funzionalità

Enhanced Support for Local AI Engines

  • A major focus of Kai 1.1.1 is continued enhancement of the integration with Ollama and LM Studio.
  • Kai 1.1 introduced a significantly more flexible architecture for connecting to AI providers and local models. With Kai 1.1.1, Embarcadero expanded their testing across a wider range of local models and configurations and made additional refinements to the local-model experience.
  • This release also enhances how Kai provides instructions and available tools to local models, along with support for longer sequences of tool calls. These improvements are especially valuable for agentic workflows, where Kai may inspect a project, read files, make changes, and carry out multiple operations as part of completing a request.
  • The result is improved reliability and broader compatibility when using Kai with Ollama and LM Studio across a range of local models and configurations. Testing included multiple Ollama models as well as successful code-modification workflows with LM Studio.
  • Local models can vary considerably in context-window size, tool-calling capabilities, and configuration, so results will depend on the selected model and local inference environment. For more advanced agentic development workflows, a capable model with a sufficiently large context window will provide the best experience.

Other Quality Improvements in Kai 1.1.1

  • Improved reliability when creating files and modifying units.
  • Improved the display of long MCP tool descriptions.
  • Refined Select All / Deselect All behavior in the MCP Tools configuration.
  • Improved MCP Tools column header display when using a dark IDE theme.
Kai 1.18 set 2026
Funzionalità

New AI integrations

  • Access 40+ AI agents through the ACP Registry, including existing integrations like Claude, GPT-5 and Gemini CLI, and newer options such as Antigravity, Grok, Mistral, Cursor, Devin, Kilo Code, and Junie.

Agent management

  • Search, install, and update supported AI agents directly from Kai, with new ACP Registry agents becoming available without requiring a Kai update.

UI

  • Enjoy a redesigned Kai experience with improved spacing, monochrome icons, a new chat toolbar, cleaner configuration pages, and more consistent controls.

Local AI

  • Use the new Native LM Studio ACP agent to work more smoothly with local models through both LM Studio and Ollama.

Sessions

  • Session handling now runs through ACP, with an Import Sessions option for bringing active-engine sessions into the Kai chat environment.

Improved Authentication

  • Authentication is improved and reworked to be handled through ACP-based authentication and OAUTH support, allowing supported login methods to be pulled directly from each connected AI agent.

Quality

  • Kai 1.1 addresses numerous customer-reported issues affecting chat behavior, scrolling, popups, performance, UI responsiveness, authentication, and model availability.
Kai 1.0.28 lug 2026
Funzionalità

Expanded Completions Support

  • Kai support for completions (available as Ghost test, Code Completions and Code Suggestions) has been expanded to offer integration with additional local AI engines.
  • Considering that different model families use different special tokens, Kai makes it now possible to pick predefined options or manually customize the FIM tokens.

Improved Compatibility and Reliability

  • This release addresses a locale issue made more visible by the 1.0.1 release and introduces an improved IME model in the chat.
  • Expanded support for additional AI deployment scenarios, including improvements for LiteLLM environments and GitHub Enterprise Cloud authentication when using GitHub Copilot.
  • Added small but useful enhancements including the ability to copy content directly from the Kai chat window and improved handling of Windows user profiles containing spaces.
  • This release includes internal security improvements that strengthen Kai's MCP communication infrastructure.
Kai 1.0.1Versione principale1 lug 2026
Funzionalità
  • Authenticated Connections for 'Local' AI Engines - This 1.0.1 update adds support for authenticated local AI servers. Kai now supports authentication when connecting to local AI servers such as Ollama and LM Studio. This enables secure connections to AI models hosted on remote or cloud-based infrastructure where authentication is required, making Kai suitable for shared AI environments.
  • Kai Quality and Stability Improvement
    • Resolved issues that could lead to IDE instability, making Kai more robust in everyday use.
    • Improved authentication and connectivity for both local and cloud AI models, including Ollama, LM Studio, and Claude.
    • Improved inline completions, ghost text, suggestion rendering, clipboard operations, and compatibility with both modern and Classic Code Insight.
    • Visual and user interface issues have been addressed, including layout, theming, font sizing, icon rendering, resizing behavior, navigation controls, and clearer terminology.