Rick Butterfield explores Chrome's WebMCP API, testing its application in the Umbraco 18 backoffice. This API allows AI agents to interact with named tools directly, enhancing efficiency in navigating the interface and executing tasks without relying on visual elements. The results indicate significant speed improvements in automated testing processes.
Kaspar Boel Kjeldsen's analysis reveals that Haiku 5.5 significantly improved its performance on the Umbraco MCP benchmark, achieving a score of 93 compared to Haiku 4.5's 47. The study also evaluates budget and open models, highlighting the importance of verified recipes for consistent results and honesty in outputs.
Matt Brailsford introduces the Umbraco.Community.AI.Carbon package, which adds a CO2 tab to Umbraco AI’s Analytics, estimating energy and CO2e for AI chat usage. Utilizing EcoLogits' method, it provides insights without new tracking. The beta version is available, aiming to enhance understanding of AI's environmental impact.
Microsoft Clarity analytics, behavioural insights and AI-assisted SEO reporting for the Umbraco backoffice.
Makes the Flowcourier Analytics data answerable in the Umbraco AI Copilot
uSync handlers and serializers for Umbraco.AI connections, guardrails, contexts, profiles and settings
Push and pull Umbraco.AI connections, profiles, prompts and agents between servers with uSync.Complete, and Umbraco.AI agent tools that publish content, pull content and take restore points, each checked against the acting user's permissions. Needs uSync.Complete.
Umbraco.AI agent tools that run uSync reports, exports and imports, each checked against the acting user's uSync access.
uSync handler and serializer for Umbraco.AI.Agent agents, including their tools, contexts and per-user-group tool permissions.
uSync handler and serializer for Umbraco.AI.Prompt prompts, so they travel between servers with the rest of your uSync files.
uSync.Complete for Umbraco.AI. Everything in uSync.AI, plus push and pull of AI items between servers and AI agent tools that publish content, pull content and take restore points. Needs uSync.Complete.
Søren Kottal discusses creating a PDF metadata dashboard in Umbraco, utilizing AI to streamline the process. This approach improved the editor's experience by providing an accessible interface for managing PDF titles. Kottal emphasizes the balance between using standard tools and custom extensions, advocating for solutions that enhance usability and efficiency.
Kaspar Boel Kjeldsen conducted a benchmark comparing eleven AI models from Anthropic and OpenAI by having them build the same Umbraco blog. The study assessed grades, cost, time, honesty, and temperament. Sonnet 5.5 emerged as the most effective model, while Haiku 4.5 demonstrated unpredictable results despite occasional success.
Copilot Workspace for Umbraco AI — full-section persisted chat with projects and reusable context attachments
Matt Brailsford discusses the introduction of Jev, a new "System One" model by TypeSafe AI, designed for quick, precise decision-making in automation. Unlike traditional chat models, Jev provides structured outputs (yes/no, pick-one, score) with confidence values, enhancing efficiency and reliability in workflows like Umbraco Automate, where accurate decisions are crucial.
Umbraco HQ introduces the Copilot Workspace, enhancing user interaction with AI by allowing persistent conversations and project-based context. This update also features knowledge sets for package developers, enabling embedded contextual information for improved AI responses. The new Copilot offers a more focused approach to content editing, streamlining user experience.
AI-powered search, chat, enquiry capture and external content indexing for Umbraco 17