feat: Each event now shows whether it is open, has ended, or has come back
feat: Add a reason when you close or reopen an alert manually, and see that reason later
feat: Edit or disable a Prometheus alert rule where it is defined, or mute it in NudgeBee, directly from the investigation screen
feat: Disable GCP and Azure alert rules at the source, and undo cloud changes exactly
feat: Pod and provider logs are grouped into highlights and grouped log cards
feat: OOM kills that happen on the same node are linked, and investigations show what a container's memory limit was when it was killed
feat: Cluster over-commitment alerts now explain what they are about
feat: Utilization, Noisy Neighbours, anomaly and SLO evidence, and cluster tiles now work with Datadog, Elasticsearch, OpenObserve, and other metrics providers, not only Prometheus
feat: Edit SLOs directly from the SLO table
fix: Crash-looping pods keep their alert open between restarts
fix: Pods are counted and filtered by the status you see, not their internal phase
fix: An object that was edited several times is shown once in change history
fix: Latency anomalies are now detected correctly
fix: Investigations use the actual incident time window, rule definition, and metric values
fix: Recurring events are re-analyzed on their own, without carrying over history from earlier occurrences
feat: Nubi can run kubectl, Helm, and AWS commands against another connected cluster or account in a single request
feat: Ask Nubi chats are listed across all accounts, and you choose the account in the message box
feat: New VM orchestrator for investigating issues across a VM fleet
feat: Remediation plans are validated before you see them, can recover work left unfinished during an incident, and can recommend and apply alert-rule fixes
feat: Investigations record the check that proves their root cause
feat: Log queries can be written in your log provider's own query language
feat: Knowledge Bases can index PDFs, including their figures, and images on Confluence and ServiceNow pages
feat: Open a Knowledge Base's document count to see its documents and their full content
feat: Knowledge Base search uses the built-in embedding model by default, and ranks results instead of filtering them out
feat: Mark which models accept images; attached images are sent to OpenAI-compatible models
feat: The code agent follows the repository's own AGENTS.md or CLAUDE.md instructions and skills, and can run long commands in the background
feat: Usage & Limits now shows the history of budget changes
feat: Tenant-wide actions for agents, tools, and Knowledge Bases; the Functions tab has been removed
fix: Detection of high-entropy secrets in outgoing data is now off by default, and can be turned on in the UI (Enterprise)
fix: Read-only commands no longer ask for approval unnecessarily, and planning continues after you deny a command
fix: Fixes that were never applied are no longer reported as "already present", and NudgeBee checks whether a fix has already shipped before recommending it
fix: Remediation plans only offer commands that the Run button accepts, and show only on the incident they were created for
fix: Event summaries are written as a short on-call brief
fix: Watches describe what they monitor and check one last time before they expire
fix: Rate limits that stop an investigation now show the real cause
fix: Inconclusive hypotheses are now shown instead of hidden
fix: Audit records now include super-admin and automated actions
fix: Accounts with several LLM providers use the tenant default provider
fix: Tools stay bound to the account you selected
fix: Only one Code Analyzer & Fixer agent is listed
feat: Notification rules have been redesigned: one rule can cover several event types and accounts, add exceptions and per-source defaults, and pause sending for a set time ("don't send"). A route check shows which rule will match
feat: Automation events can be used in notification rules, with a new Automation tab in the rule editor
feat: Optimization digests are sent per channel, grouped by account
feat: Slack incoming webhooks and Google Chat webhooks as notification targets
feat: Follow-up questions with many options show a dropdown next to the web app link
fix: Slack alert links open the relevant page instead of Home
fix: Nubi now reads Slack thread context correctly
fix: The alert report is put together directly instead of being rewritten by a second AI pass
fix: The daily recap explains what the Highlights percentages are compared against
fix: Brand and assistant names are used consistently across chat platforms