On this page
What's Changed
🔍 Troubleshoot
feat: Group same-subject alert bursts under one incident leader and surface them in the incident triage UI
feat: Parse Knowledge Graph evidence format to correlate and group cross-service alerts by stored topology
feat: Provide rule-health verdicts for alerts without fetchable metrics to identify stale or misconfigured rules
feat: Introduce role-based dashboard templates and widget catalog for Custom Dashboards
feat: Add column display settings, custom links, and search-on-enter capabilities to dashboard table panels
feat: Replace legacy Troubleshoot event KPI cards with the interactive Nubi briefing strip
feat: Add Disk I/O metrics to Kubernetes cluster utilization charts
feat: Add MCP circuit breaker and real-time observability metrics
fix: Surface per-query Elasticsearch metric failures directly as structured tool errors
fix: Use real Elasticsearch metric index mappings and schemas instead of fallback wildcard queries
fix: Bound telemetry metric label values RPC round-trips to 60s to prevent hanging queries
fix: Enable metrics agent to read actual label values instead of inferring them
fix: Correct SLO evaluation logic, add NO_DATA status handling, and persist burn-rate statistics
fix: Direct metrics agent to suggest shell executions when specific queries require deeper introspection
⚡ Optimization
feat: Introduce centralized resolution coordinator enforcing legal lifecycle state transitions across recommendations
feat: Support end-to-end dismiss and snooze workflows for recommendations with full visibility in Optimize views
feat: Automatically settle recommendation resolutions from local ticket status via the resolution coordinator
feat: Update open rightsizing pull requests automatically when underlying workload recommendations change
feat: Redesign Optimize recommendation filters and surface resource ownership in the recommendation detail drawer
feat: Add self-hosted VM package inventory matching against CVEs with centralized vulnerability tracking
fix: Preserve claimed and settled recommendation states across recurring collector syncs and rightsizing upserts
fix: Ensure dismissed vulnerability scan findings remain dismissed across recurring scans
fix: Prevent merged rightsizing pull requests from blocking subsequent recommendation PRs
fix: Retire stale recommendations stuck in InProgress status and eliminate no-op rightsizing recommendations
fix: Categorize requests-unset pod recommendations under Configuration and key rightsizing behavior on rule name
fix: Correct rightsizing scan priority parsing as integer levels (0-4)
fix: Display bare Cloud SQL instance names in recommendation evidence and attach notification channels to generated alarms
fix: Report gross cloud spend instead of net-of-credits in FinOps analytics tables
fix: Asynchronously refresh FinOps account context to eliminate system prompt build latency
🔄 Workflow
feat: Make the Automations dashboard tenant-level with account filtering and reordered tabs
feat: Enable AI agents to autonomously discover and invoke opted-in automations from incident symptoms
feat: Allow automations to pre-approve specific tool permissions for downstream sub-agents
feat: Track caller and callee execution relationships, surfacing nested automation runs in execution histories
feat: Add state-aware Activate and Pause toggles directly in automation action menus
fix: Enforce SSRF protection with NAT64 and CGNAT CIDR block filtering on runbook-server network tasks
fix: Validate cron syntax and parameter definitions prior to saving automation trigger registrations
fix: Roll back workflow creation transactions when trigger registration fails
fix: Report accurate execution status for workflow steps conditionally selected by Switch branches
fix: Fix execution pagination and summary ranking in the Automations dashboard
fix: Skip external ticket writes during workflow dry-run executions
fix: Order MCP authentication configuration fields before tool names with descriptive tool-list error feedback
☸️ Kubernetes
feat: Add self-hosted VM fleet management page under Infra with provider-generic overview
feat: Detect Kubernetes Secrets exposed as plaintext pod environment variables
feat: Add GitHub Actions (ARC) self-hosted runner monitoring dashboard on cluster details
feat: Merge Clusters and Cloud navigation into a unified Infra sidebar with section flyouts
fix: Add SSRF protection to MCP direct handler connections
fix: Resolve pod workload ownership accurately from subject owner during event log analysis
fix: Prevent GCE metadata DNS resolution inside workspace pods
fix: Resolve Prometheus alert subjects that fallback to alertname during collector sync
fix: Prevent cluster rightsizing errors when an account lacks Kubernetes provider configurations
fix: Filter out deleted and archived integrations during resource discovery
☁️ Cloud
feat: Add GCP alarm-missing recommendations for Cloud Run, Pub/Sub, Memorystore, Cloud Functions, Load Balancing, and Cloud Storage
feat: Add metric-label alert filtering support for cloud monitoring collectors
fix: Widen GCP resource discovery crawl across account region footprints and seed crawl regions from billing exports
fix: Derive GCP Recommender locations dynamically from active account resources
fix: Skip unroutable AWS regions and cap AWS Config rule discovery to keep cloud synchronization within execution budgets
fix: Acknowledge EventBridge SQS messages when account tokens lack an active cloud account
fix: Resolve RDS database endpoints so VPC Flow Log edges map correctly in topology graphs
fix: Scope GCP log evidence accurately for native log alerts and HTTP load balancers
fix: Publish resolved event notifications properly from cloud alarm resolve paths
fix: Align AWS Savings Plan cost metrics to the same monthly basis as savings estimates
🤖 AI
feat: Wrap LLM clients with universal PII/PHI redaction and rehydration with tenant-scoped admin controls
feat: Add daily AI cost reports via Slack digests and executive dashboards with setup guides
feat: Add Google Vertex AI support to AI Gateway with generic endpoints, tenant BYO credentials, and OpenAI-compatible MaaS provider
feat: Add AWS Bedrock provider support to AI Gateway with Converse API tool-calling, IRSA, and credential forwarding to sandboxes (bare model IDs are supported for on-demand inference within the same region, with inference profiles recommended for cross-region routing)
feat: Support model ID aliasing (alias=served model IDs) allowing multiple gateway upstreams to serve the same model identifier
feat: Add embeddings endpoint on the generic /v1 AI Gateway lane and forward provider-specific parameters
feat: Support private and self-hosted LLM endpoints (vLLM, Ollama, SGLang) with cold-start warming detection and fast failover
feat: Support Gemini 3 native thinking_level parameter and thinking-aware TTFT watchdogs
feat: Add OAuth2 client-credentials authentication and custom request headers for external LLM configurations
feat: Optimize prompt caching with asynchronous cache creation, multi-breakpoint support, and zero-latency hits
feat: Standardize default lean orchestrator mode across K8s, AWS, GCP, and Azure with unified tool sets and delegation rules
feat: Introduce dedicated AWS observability sub-agents (aws_metrics, aws_logs, aws_traces)
feat: Add GitLab specialist agent backed by the glab CLI for issue and merge request workflows
feat: Add Events-v2 investigation agent with deterministic structured tools and direct summarization shortcuts
feat: Introduce typed resolution write tools and operator-readable approval cards on the FinOps agent
feat: Enable dynamic skill discovery from account Knowledge Bases via load_skills
feat: Introduce file-system evidence recall with compression handles and cross-message indexing
feat: Add tenant-scoped weekly reliability reviews and weekly event analysis digests (b-Cortex Digests)
feat: Add global Nubi AI chat drawer with sticky launcher across all pages
feat: Add chat action audit logging for web chat follow-ups and conversation deletions
fix: Eliminate PII scrubber false-positives on Kubernetes and DevOps operational payloads
fix: Consolidate agent loop-detection mechanisms into a unified stats-based convergence brake
fix: Deduplicate agent memory facts across all stores and scope RCA feedback to message subjects
fix: Preserve model tier and task type metadata on token usage tracking records
fix: Enforce time-to-first-token (TTFT) watchdogs across Google AI and custom providers, recording abandoned attempts
fix: Resolve pinned LLM configuration context windows and cap prompt token sizes before dispatch
fix: Offload oversized event RCA prompt inputs to workspace filesystem storage
fix: Format tool calls and responses to match strict OpenAI-compatible schema specifications
fix: Support vision-capability gating and image attachments in agent turns
fix: Prevent agent panics when custom prompts omit tool usage definitions and recover from empty responses via model fallbacks
fix: Synchronize action node status reads in the parallel planner dispatcher to eliminate race conditions
fix: Ensure git operations in code analysis clone with matching repository credentials and prevent empty-commit PRs
fix: Bound event analysis reuse freshness gates to prevent stale investigation caching
🔔 Notifications
feat: Show live Thinking Steps progress and real-time activity headers in Nubi Slack threads
feat: Render Mermaid diagrams, charts, tables, and formatted markdown as native Slack message blocks
feat: Ground Nubi conversations in watched Slack channel context, humanizing user mentions and omitting redundant greetings
feat: Deliver automated weekly digests and dynamic daily report emails to configured notification channels
feat: Display cited Slack channel conversations directly in Nubi web chat responses
fix: Bind incident Slack channels to account and conversation contexts, preventing cross-tenant message leaks
fix: Thread repeat finding notifications into the channel they were originally delivered to
fix: Prevent finding notification cards from getting stuck in Analyzing status
fix: Preserve notification alert cards when initiating Ask Nubi to Analyse actions
fix: Fix Slack markdown table rendering, link pipe-splitting, and bold/italic formatting
fix: Handle missing account context gracefully in native Slack interactive action payloads
fix: Surface clear diagnostic errors when adding Slack bot members to private channels
🎫 Tickets
feat: Add deletion support for ticketing integrations
fix: Schedule automated hourly Ticket Status Sync cron job to synchronize ticket states in the background
fix: Strip unsupported optional fields before dispatching ticket creation requests to Ticket Master
🔌 Integrations
feat: Support Confluence Data Center integration via Personal Access Tokens (PAT) and Basic authentication
feat: Add Elasticsearch and Kibana alerting webhook integrations with automated event closure and subject extraction
feat: Add public /status page and platform status aggregator
feat: Allow integrations to appear under multiple functional categories in the integrations catalog
feat: Register forager discovery datasources as active platform integrations
feat: Index custom dashboards, Automations, and newer tabs in global page search
fix: Display real-time modal progress loaders while user roles are being created or updated in RBAC
fix: Target link-role aware constraints when creating integration configurations