v1.4.0

What's Changed

feat: Expand Knowledge Graph with Azure managed service classification, GCP trace/data flow integration, and enhanced node filtering and detail layouts
feat: Enhance trace observability with a root-span "By Traces" listing view, merged label mappings, and unified GCP trace routing
feat: Improve platform UI/UX with an in-app guided tour engine, product updates drawer, endpoint readiness visibility, and smoother navigation transitions
feat: Optimize container image scanning with incremental updates and clean-image status reporting
fix: Tune RabbitMQ alert thresholds to reduce noise and resolve orphaned GCP project node mapping issues

feat: Support user-attached images and auto-enable image collection for Nubi Slack conversations
feat: Add channel member management across Slack, MS Teams, and Google Chat
fix: Resolve event loop freezing and blocking issues in Google Chat and Discord integrations, and encrypt Discord bot tokens at rest
fix: Fix navigation links for resolution details and direct events summary email tips to the autopilot dashboard

feat: Trigger RCA write-back upon incident investigation completion
fix: Fix ticket action validation to support multiple assignees, correct GitHub/GitLab repository targeting, and isolate auth config settings

feat: Add session-scoped LLM conversation sharing across workflow runs
feat: Expand workflow controls with skipping of optional mitigation steps and granular audit event categorization
feat: Restructure Autopilot navigation to group Auto Optimize under the Optimize section and rename paths to /automation
fix: Resolve Temporal workflow deadlocks, relay agent connection serialization errors, and kubectl task execution failures
fix: Improve task management by recording failed auto-optimize runs and outputting human-readable validation messages

feat: Extend blast-radius safety bands and impact summaries to EC2 and RDS recommendations
feat: Batch CloudWatch metrics fetching for EC2 recommendations to improve performance
fix: Prevent BigQuery and collector OOM crashes by computing accurate GOMEMLIMITs, dropping large table schemas from metadata, and disabling per-table discovery by default
fix: Standardize GCP recommendation and alert attributes, including severity defaults, stopped status mapping, alarm metric recovery, and LB/latency log extraction
fix: Enhance cloud collector stability by resolving Azure pagination loops, preventing MQ retry storms on credential failures, optimizing ECS task metadata fetching, and avoiding duplicate resource insertion crashes

feat: Implement LLM gateway request pipeline (data-plane) with per-tenant provider credentials and egress filter configurations supporting tagging and redacting
feat: Enhance agent autonomy with background watch registration, interactive FinOps analytics, direct PR proposal from event analysis, and GitHub CI artifact triage
feat: Optimize agent execution with React3 role-split prompt overlays, service-facing memory injection APIs, input schema validation, and reasoning metric instrumentation
fix: Standardize the agent orchestrator layout on Orchestrating (retiring ReWoo) and resolve React3 execution anomalies, including dual-write/clone races, crash recovery, and validation assertions
fix: Fix token metric reporting and pricing calculations by persisting cost values, correcting tracking duration offsets, and enforcing provider-specific thinking-token spend limits
fix: Stabilize agent workspace tool executions by repairing ClickHouse client paths, injecting workspace env vars, resolving zero-byte stdout success validations, and adding shell CLI recovery hints
fix: Resolve caching, RAG retrieval, and citation formatting issues, including stale Google AI pointers, user-context ID forwarding, and FinOps deep-link formats

feat: Resolve Datadog pod subject ownership from live Kubernetes state
fix: Sanitize database connection names for PostgreSQL, MySQL, and ClickHouse relay connections