CastrelCastrel AI
Thursday, June 4, 2026 · v1.13.0

Chat Experience Overhaul & AI Capabilities Expansion

Chat Experience Overhaul & AI Capabilities Expansion
This release comprehensively redesigns the Chat interaction experience, introduces interrupt-style AI questioning with conversation steering, and adds application initialization, dashboard generation, and data traceability for more efficient and trustworthy troubleshooting.

New Feature: Chat Experience Overhaul

The Chat page sees its most comprehensive interaction redesign, covering the message reading area, task process folding, tool call presentation, session info card, side panel, sidebar, and input box.

The reading area is narrowed for denser information in long conversations. Agent execution steps now collapse by default — each round shows a one-line summary with step count and elapsed time. Tool calls use a lighter display style; clicking opens details in the right panel without inline expansion. Repeated calls to the same tool (e.g., Prometheus × 3) merge into a single entry. A new right-side session info card aggregates sources, sub-agents, artifacts, and todos in one panel, adapting visibility to screen width.

The input box now supports a / shortcut for mode switching — operators type / to quickly select a Chat scenario, replacing the previous button-and-dropdown flow. Other refinements include auto-focusing the last artifact in historical sessions, document tabs that show actual names, sub-agent indentation in the sidebar, and a cleaner footer with a context-compression indicator.

New Feature: AI Question Panel & Conversation Steering

When the Agent needs input during an investigation, the question prompt now appears directly in the input box — no more popping open a side panel. The input box transforms into a lightweight panel with recommended options, single-select, free-text input, and multi-step progress navigation. The right panel only opens when the operator clicks a tool call for details.

Conversation steering lets operators guide the Agent's analysis direction mid-conversation — pointing out priorities, clarifying assumptions, or asking for deeper investigation into a specific component — turning the investigation into a collaborative dialogue rather than a one-shot Q&A.

New Feature: Application Initialization, Dashboard & Data Traceability

Application initialization preloads connectors, resources, and relationships in one step, with quick binding for one-click data association and precise binding for field-level mapping and filtering.

Operators can generate dashboards in HTML and YAML formats during analysis sessions, inspect specific segments, and manage Grafana dashboards — create, update, and rollback — without leaving Castrel. Every conclusion is traceable to its source: clicking any finding reveals the exact metric, log line, or event that drove it.

Full Release Checklist

feat

  • Narrowed the Chat message reading column; added user avatar alongside hover timestamp and copy actions on messages.
  • Agent execution steps now collapse by default; each round shows a one-line summary with step count and elapsed time.
  • Reworked tool call display to a lighter style; clicking opens details in the right panel; repeated calls to the same tool are merged into a single entry.
  • Added a right-side session info card aggregating sources, sub-agents, artifacts, and todos with adaptive visibility.
  • Side panel tabs now auto-focus the last artifact when entering a historical session; tab close uses a hover-to-reveal × icon.
  • Document and knowledge tabs display the actual document name instead of a generic label.
  • Removed icon placeholders from sidebar conversation rows; sub-agent rows are indented to express task hierarchy.
  • Chat input box footer reworked to a lighter style; horizontal indicator appears on automatic context compression.
  • Added / shortcut for Chat mode switching, replacing the previous button-and-dropdown selector.
  • Moved Agent question prompts into the input box as an interrupt-style panel with recommended options and free-text input.
  • Added conversation steering for collaborative Agent direction control.
  • Added application initialization with quick and precise data binding modes.
  • Added dashboard generation in HTML and YAML formats with segment-level inspection.
  • Added Grafana dashboard lifecycle management — create, update, and rollback.
  • Added data traceability to reveal the exact data source behind each conclusion.
  • Overlaid global Application topology on the RCA fault path view with three visual tiers for fault nodes, affected objects, and normal nodes.
  • Launched Incident Pattern MVP for extraction, matching, and reuse of historical fault patterns.
  • Redesigned RCA context management with unified App-aware context propagation across all investigation nodes.
  • Refactored RCA Agent as a plugin and integrated it with the RCA home Copilot for incident queries and trend analysis.

bugfix

  • Fixed an issue where answer content was not displayed when clicking a tool call after submitting a response to an Agent question prompt without refreshing the page.