Week of May 15–22, 2026

Frontend

• AgentTasksTab for periodic agent management — comprehensive CRUD interface for scheduling agent tasks with date/time/cron repeat configurations, log-details modal, and status badges (Completed/Failed/Running/Scheduled/Disabled) via new useGetPeriodicAgentsQuery and useCreatePeriodicAgentMutation hooks
• AgentPrivacyTab component — master toggle with action dropdown (redact/mask/block), block-message textarea for block action, entity-type chips for PII detection, and output-filter toggle, persisted via useEditMentorJsonMutation to the /settings/ endpoint
• Advanced Chat SDK integration — full agent surface (message stream, canvas, file attach, voice, prompt gallery) rendered directly in Next.js host apps via in-process Chat component with ServiceWorkerProvider wrapping and Redux slice integration
• Initial prompt auto-injection — new boolean agent_autoplay configuration with autoplay icon on course content pages for automatic TTS playback of agent responses

Data Model

• Privacy router types and constants — PrivacyAction, PrivacyEntityType, MentorPrivacySettings schemas with SUGGESTED_PRIVACY_ENTITIES and DEFAULT_PRIVACY_RESPONSE configurations
• Periodic agents API slice — useGetPeriodicAgentLogsListQuery and useDeletePeriodicAgentMutation for task lifecycle management with cron parsing utilities
• Monetization owner logic updates — refined enable_monetization flag enforcement with management tab descriptions for paywall configuration

Security

• Dependency updates — refreshed npm and Rust dependencies to current patched releases as part of routine maintenance
• Node.js 20 LTS upgrade — theme build infrastructure moved from EOL Node 16 to active LTS, with dependency overrides aligned to patched releases

Backend

• CRM Identity slice v1.0.0 — new /api/crm/ namespace with Person and Organization CRUD endpoints, POST /api/crm/persons/{id}/invite/ for platform invitations, POST /api/crm/persons/{id}/link-user/ for manual User binding, and POST /api/crm/persons/merge/ for duplicate consolidation
• Text-to-speech endpoint — GET /api/ai-mentor/orgs/{org}/users/{user_id}/chat-messages/{chat_message_id}/tts/ returns MP3 audio synthesized from message text using the mentor's configured TTS provider (OpenAI/Google)
• Evaluation system enhancements — dataset items include trace_input/trace_output via ?include_trace=true; experiment detail carries input/expected_output/actual_output/trace_id/error/scores in unified responses
• LLM-as-Judge endpoints — GET /evaluations/datasets/{ds}/runs/{run}/evaluate/ for per-run status and GET /evaluations/judges/ for tenant-wide judge monitoring with full lifecycle tracking via EvaluationTaskRecord table

Data Model

• RBAC action catalog — GET /api/core/rbac/actions/tree/ hierarchical action tree for autocomplete UIs and GET /api/core/rbac/actions/definitions/ flat action dictionary with descriptions and assignable resources
• Platform monetization separation — new Platform.enable_monetization boolean distinct from show_paywall for Stripe Connect merchant capability with owner change audit logging via data["owner_change_log"] JSONField
• Notification template taxonomy — Content/Audience/Action metadata dimensions stored as prefixed tags with admin filters for template catalog organization
• Voice provider configuration — Mentor.voice_provider field (OpenAI/Google) with read/write access on mentor settings endpoints for TTS and LiveKit call integration

Security

• Integration credential access control — sensitivity-aware serializers that mask sensitive credential fields and scope access by role
• Backend security review — routine hardening across data-access and endpoint authorization paths, with added input validation and Docker base-image package updates to current patched releases

Infrastructure

• LiveKit agent tool integration — call agents receive mentor's configured Langchain tools via CallAgentToolFetcher with BaseTool.ainvoke async execution and ToolError translation
• Managed claw cost reconciliation — ReconciliationService fetches usage via WebSocket RPC alongside ClickHouse data for complete billing reconciliation with source breakdown metadata
• Langfuse direct database access — performance optimization using managed=False reflection models eliminating per-row HTTP fan-out for experiment detail queries
• Tenant platform launcher — ibl_tenant_platform ansible role creates Platform + admin User + UserPlatformLink via run_launch_steps when PLATFORM_NAME set, with generated admin password surfaced through IBLAI_FIXTURE_OUTPUT pipeline
• Single-server safety defaults — 100GB minimum volume floor with Pydantic validation, 32GB memory warnings suggesting 64GB instances, and IBL_SPA.MENTOR.STRIPE_ENABLED=false defaults
• Microsoft SSO completeness — microsoft_sso_config patches IBL_SPA.AUTH.EXTERNAL_IDP_LOGOUT_URL and IBL_DIRECT_SSO_URL with tenant ID fallback and Auth/Mentor SPA restarts
• SPA wait budget optimization — increased from 150s to 450s (7.5 min) for cold-start pnpm install completion preventing false-negative provision failures

API Changes

• DM field encryption documentation — comprehensive operator guide for IBL_DM.FIELD_ENCRYPTION_KEY provisioning, Fernet dual-write pattern, and data-loss-aware key rotation runbook
• Langfuse project ID propagation — IBL_DM.LANGFUSE_PROJECT_ID config key with default ibl_ai injected into all DM containers for project identification
• Course metadata flags — agent_content_mode, course_content_mode, and agent_autoplay added to COURSE_METADATA_KEYS allowlist and public API exposure

Documentation

• Agent chat integration skill — comprehensive documentation for rendering the SDK's in-process Chat component with peer dependency setup, ServiceWorkerProvider wrapping, Redux slice integration, and full props reference
• Agent privacy skill — PII detection and filtering documentation with redact/mask/block action configurations and UI workflow examples
• Monetization skill implementation — complete skill covering Stripe Connect onboarding, paywall configuration, pricing management, and admin interface workflows
• Skill infrastructure improvements — updated agent chat skill with known-issues solutions; removed deprecated iblai-chat skill in favor of direct SDK integration

REST API Changelog

New Endpoints

• GET /api/ai-mentor/orgs/{org}/users/{user_id}/chat-messages/{chat_message_id}/tts/ — Text-to-speech synthesis for stored chat messages using mentor's configured voice provider
• GET /api/ai-mentor/orgs/{org}/users/{user_id}/evaluations/judges/ — LLM-as-Judge endpoint for tenant-wide evaluation monitoring