The April–May release of Agentic studio focuses on extensibility, performance, and day-to-day usability. Custom MCP connectors let you bring your own tools into Agentic studio with OAuth-based authentication. Workflows now run faster and recover from upstream failures thanks to parallel per-item execution and provider-level retries with exponential backoff. The Contextual Assistant gains quicker loading of languages and brand kits, chat history search, and tighter integration with the host Sitecore Pages window.
Smarter chat and signals
Chat and signals have been refined for speed, traceability, and continuity across long sessions.
Instant language and brand kit loading - Site languages and brand kits associated with your space appear immediately when you open the assistant, rather than after a separate fetch finishes. Brand kit detection recognizes additional identifier sources and uses friendly display names throughout the assistant flow.
Synchronized assistant shell - When you open an artifact, the assistant tells the host Sitecore Pages window to expand; when you close it, it collapses. The host page gets out of your way when you need the room and tucks back when you don't.
Reliable streaming recovery - If a streaming response is interrupted, reopening the conversation refetches the authoritative message state from the server rather than leaving a stale half-response on screen. Active run lifecycle handling has been centralized so chat resume works consistently across long-running sessions.
Chat history search - Search across your chat history to find past conversations and pick up where you left off.
Cleaner brief generation - Form labels are stripped of internal section prefixes for readability, pre-filled context parameters are sanitized before display, and artifact cards appear more cleanly in the brief flow.
Preserved view state - Returning to a space from an artifact drops you back into the view (list, cards, etc.) you were using before opening it.
Signal pulses - A new type of generated market signal, signal pulses, is wired through Agentic studio end to end, with full API coverage.
Signal artifact selection - The Signals view supports richer artifact selection and action handling, so working with the artifacts generated from a signal is more direct.
Expanded agent capabilities
Agents gain new ways to integrate with your tools, work with your content, and stay under your governance.
Custom MCP connectors - Bring your own MCP (Model Connector Provider) endpoints into Agentic studio. Add, authenticate via OAuth, enable/disable, and remove them through the tools settings UI. Each user authenticates separately, so private connectors stay private.
Page HTML context variant - SEO, AEO, GEO, and accessibility agents can request the rendered HTML of a Sitecore page as a first-class context variant, so they can analyze the actual markup visitors see. Configurable in the agent setup panel.
Scoped page search - The Sitecore page search tool now requires a site name up front and offers better client-side filtering. Results are scoped correctly and are returned faster.
Widget and report outputs, with a visual schema editor - Agents can produce JSON-rendered widget outputs (with a dynamic catalog of widget types) and report-style outputs alongside text, code, and image artifacts. The widget editor gained preview data, schema diagnostics, and a visual schema editor tab so building widgets is click-driven rather than hand-edited JSON.
Clickable prompt actions in agent responses - Model responses can include actionable prompt suggestions that insert text into your next message rather than sending it immediately. Site and page actions from agent tool outputs can be inserted as removable pills.
Agent name conflict validation - Creating or renaming an agent now checks for duplicate names across your space and blocks conflicts up front, instead of letting two same-named agents coexist and confuse downstream lookups.
Language version auto-creation - When an agent requests page content for a language that doesn't yet have a version, the system creates it automatically and surfaces clearer messages when something genuinely fails.
Skill sharing with visibility controls - Skills now have private / team / public visibility plus a collaborator system. You can share an in-progress skill with specific colleagues without exposing it to the entire organization.
Image display in artifacts - Agent-generated images now render properly in the artifact panel, and image-generation model selection is centralized so agents and workflows produce consistent results.
Faster and more reliable workflows
Workflows now run faster, recover from upstream failures, and behave consistently across long sessions.
Parallel per-item workflow execution - Workflows that fan out across many items such as bulk translations, content updates, multi-page operations, execute per-item branches in parallel rather than sequentially. Each branch streams its progress back independently, and fan-out window size is tunable.
Provider retries with exponential backoff - When a model provider hits a rate limit, timeout, or transient error mid-generation, workflows automatically retry with exponential backoff and surface detailed error information if all attempts fail. Per-node retry policies can be configured in the workflow graph.
Connected workspaces as first-class workflows - Linked agents in a space compile and execute through the same workflow runtime as everything else, with a live contract preview in the configuration UI, persistent agent connections, and run output surfaced alongside the connected workspace view.
Faster artifact rendering - Fixed a memoization issue in the Artifact and ArtifactModal components that was silently disabling React's render skipping. The affected views now skip re-renders as intended.
Smaller model context on long conversation - Chat history is compacted and summarized server-side before being sent to the model, so long conversations don't keep ballooning the prompt and slowing down responses.
Smarter document persistence - The save workflow only persists documents when something has actually changed, preventing duplicate writes during routine edits.
HTML template previews preserve <head> styles - Styles and stylesheet links that you place in the document head are correctly carried through into the iframe preview. Scripts and inline JavaScript continue to be removed for safety.
Reliable workspace run persistence - Workspace run persistence is now reliable, with runs saving and resuming as expected after fixes to the underlying logic.
Search reasoning fix - Search queries containing capital-S words (Sitecore, SEO, SSL, and similar) are no longer truncated by the reasoning-step extractor.
URL updates - /chat now permanently redirects to /conversations, and the long-standing /enviroments → /environments typo has been corrected with a permanent redirect. Existing links keep working.
Previews and early access
Capabilities still being rolled out and are available for early evaluation.
Virtual Filesystem (Preview) - A bash-like interface that lets agents browse and edit a snapshot of your product data through familiar commands (ls, cd, cat), with an explicit commit step and permission checks before changes touch real data.
Sitecore Context tool (Preview) — A new tool, sitecore_context, that gives agents off / read / write access to your Sitecore environment, configurable per agent.
Admin Configuration Packages (Early Access) — Initial groundwork for an upcoming admin feature to export and import configurable Agentic studio items (agents, signals, schemas, HTML templates, skills, widgets) as zip packages.