• Added a dedicated AI Web API extension under /api/ai with endpoints for providers, requests, resources and configured skills.
• Added an OpenAI-compatible chat completions proxy at /api/ai/chat/completions and /api/ai/v1/chat/completions that forwards to the configured default provider.
• Added provider selection by unique provider name for AI requests and resource uploads, while still supporting the configured default provider.
• Added the new permission ai.api for AI Web API access, including Web API help integration and permission labels.
• Added configurable AI slash commands for the CoWork bot, including provider selection and command-specific system prompts; adding, renaming or deleting commands updates CoWork command availability accordingly.
• Added the ability to determine the desired response format of a Task Planner AI Query Job: Plain Text, HTML, or Markdown.
• Added per-job AI tool selection to Task Planner AI Query Jobs. Only tools selected for the job and enabled in the Task Planner AI settings are provided to the request.
• Added TypeSafe Jev support, including a Jev job for the Task Planner and AI configuration for the encrypted API key and configurable base URL. Jev is TypeSafe's decision model: it evaluates natural-language context and returns structured choices, scores or yes/no probabilities that software can use directly, instead of generating prose. It can support semantic workflow steps such as classifying or routing tickets, checking conditions and ranking relevant knowledge base content.

Fixed Bugs

• Added handling for tools such as AI Triggers which have spaces in their names.
• The AI Query job in the Task Planner no longer allows empty request texts.
• The AI Trigger's placeholders are now offered to subsequent task components.
• The "Classify Ticket with AI" action now reports missing AI configuration instead of a misleading error message.
• Added connection and read timeouts for AI communication to prevent potential system hangs if an AI provider stops responding.

• Added a "Max Context Length (tokens)" property to AI providers which will intelligently attempt to shorten AI requests in case they would exceed a token limit.
• Added activation toggles and a "Default" provider option for AI translations, with automatic migration of the legacy "use default provider" settings.
• Removed all MCP client integration from the AI plugin, including MCP configuration, validation and MCP-specific documentation.
• Added a GPT provider setting "Reasoning Level" to explicitly control reasoning_effort; automatic model-based defaults are no longer sent.

Fixed Bugs

• Disabling the CoWork AI bot now fully disables all AI bot responses, including when replying to an existing AI message.
• The link to the AI provider configuration dialog was incorrect in a couple of Task Planner components.
• Fixed icons for the AI Trigger and Job in the Task Planner.

• Added a model dropdown list for the three primary providers (OpenAI, Gemini, and Claude) to the provider configuration in order to make it easier to choose your desired AI model.
• New AI Provider option: Azure OpenAI
• New OAuth Token option for hosting local LLMs via Ollama with authentication.
• Added support for streaming APIs, providing chunked output as it arrives from the AI Provider.
• Added support for function calling, providing the ability to provide the AI with tools it can call to fulfill its tasks.
• Added support for MCP clients either via npm or python. These can be set in the Configuration by system administrators and they can be provided to the AI when sending requests - caution is advised in the usage of MCP clients.
• As the automatic anonymization could not be guaranteed to work in all cases, it has been removed. If you require data privacy for your AI calls, we recommend using a local AI.
• New AI Provider Type "custom" which makes it possible to provide any OpenAI-compatible endpoint to connect to.

• Added DeepSeek support as an AI Provider

• Added Gemini, Claude, and Mistral support as AI Providers.

• New Clear Reports formula function "gpt" which takes any string query as a parameter and returns the GPT response.
• Added obfuscation to storage of OpenAI API Key in configuration.
• HelpDesk spam filter capability (off by default) which can check incoming emails for whether GPT would categorize them as spam.
• Anonymization of any telephone numbers and email addresses to avoid sending personally identifiable data to OpenAI.