Skip to main content

CubePlex 0.9.3

Features and improvements​

  • Mermaid diagrams in Markdown previews. Chat messages, artifact previews, skills, fetched pages, and trace messages render fenced mermaid blocks as diagrams that follow the current theme. Incomplete or invalid diagrams show their source, and downloads keep the original Markdown. See artifact previews.
  • Prometheus metrics for backend operations. The backend exposes HTTP and SSE activity, agent run outcomes, LLM calls and token usage, and tool call outcomes at /metrics. Docker Compose labels and Kubernetes Pod annotations support discovery when configured in your Prometheus installation. See the metric list and counting rules.
  • File previews appear earlier during writing. A write tool row and its preview can appear while file content is still being generated. Concurrent writes keep the file you selected open as other calls update or finish.
  • Conversation refreshes preserve reading state. Unchanged conversations avoid downloading history again or remounting message cards. Refresh checks pause in hidden tabs and resume when you return.

Fixes​

  • More reliable subagent commands. Subagent shell commands stay in the foreground, stream progress, and return a bounded result or timeout. Transient polling failures are retried without replaying commands whose start is uncertain; missing command records and provider failures are reported. Completed output is preserved around deadlines. See sandbox command behavior.
  • Original images in chat. Presented images load the original file, preserve its decoded proportions, and defer loading until needed. Click an image to open the original; failed images offer a download.
  • Internal background task payloads stay hidden. Raw task delivery inputs no longer appear in chat during active streams or in the background task list.
  • Citations remain visible beside diagrams. Rendering Mermaid no longer drops citations from surrounding Markdown.
  • Clearer shutdown logs. Expected request and SSE cancellations caused by Uvicorn's graceful shutdown timeout log at INFO without a traceback. The forced-shutdown summary remains ERROR, and unrelated cancellations and application errors retain their original reporting.

Before upgrading​

Upgrade the backend, frontend, and sandbox image together to v0.9.3, or install Helm chart 0.9.3. This release adds no database migration or required configuration change beyond the normal image version update.

If your organization has a saved sandbox policy, update its default image to ghcr.io/cubeplexai/cubeplex-sandbox:v0.9.3. The saved policy takes precedence over the deployment setting. Existing running sandboxes keep their current image until restarted after their tasks finish.

The new /metrics endpoint has no application authentication. Keep it accessible only to your monitoring network or restrict access at your proxy or firewall. Prometheus must be configured to scrape each backend instance; deployment labels and annotations alone do not enable scraping. See the Docker Compose and Kubernetes guides.

GitHub Release · Full changes since 0.9.2.