---
title: "Week of May 1–8, 2026"
slug: "platform-update-2026-05-08"
date: "2026-05-08"
tag: "Application"
summary: "Agent sandbox management, MCP connector lifecycle, field-level encryption, multimodal graph agents, dedicated notification queues, and one-click production deployments."
author: "ibl.ai Engineering"
---

## Frontend

### Agent Configuration Infrastructure

- **Agent Sandbox Management API** — `SandboxConfig` component with CRUD operations for agent instances, health checks, connectivity testing, and model picker integration via new data layer hooks for agent config (get/create/update/delete/push) under agent-nested URLs
- **Agent Skills Management** — `AgentSkills` component with skill CRUD operations plus per-agent enable/disable controls through new agent skill resource hooks and per-agent skill assignment hooks
- **Agent Prompt Configuration** — `AgentConfigPrompts` component exposing eight prompt fields with rich-text editor for comprehensive agent behavior configuration
- **Instance Health Monitoring** — automated health check and test connectivity mutation hooks with push gating, server error messaging via `readErrorMessage`, and optimistic cache writes for connect/disconnect operations enabling instant UI updates without refetch

### MCP Integration & Memory Management

- **MCP Connector Management** — comprehensive `connector-management-content` and `connector-dialogs` components for Model Control Protocol configuration with full connector lifecycle management
- **Enhanced Memory Architecture** — restructured memory management with `learners-memories`, `saved-memories-modal`, and `manage-categories-modal` components, including new memory categorization system and learner-specific memory workflows
- **Settings Integration** — moved memory toggle to settings tab with enhanced per-agent memory controls

### E-commerce & Credit Management

- **Credit Balance Component** — new `CreditBalance` component with real-time credit monitoring, auto-recharge configuration, and upgrade package modal integration for seamless payment workflows
- **Enhanced Billing Integration** — updated billing tab with credit coin component, e-commerce flow optimizations, and Stripe upgrade hook (`use-stripe-upgrade`) for streamlined payment processing
- **LLM Provider Visualization** — added provider icons and branding for Amazon, Claude, DeepSeek, Gemini, Groq, LLaMA, Microsoft, Mistral, NVIDIA, OpenAI, Perplexity, and xAI with `get-llm-provider-details` utility for consistent provider representation

### Accessibility & User Experience

- **Persistent Chat Input Labels** — added configurable `persistent_chat_input_label` tenant metadata setting enabling WCAG 3.3.2 compliant visible labels above chat textarea (opt-in via tenant configuration)
- **Enhanced ARIA Support** — comprehensive `aria-label` improvements across chat input components including voice call buttons, voice chat buttons, and upload menus for improved screen reader accessibility
- **Agent Display Enhancements** — unique ID configuration in agent settings, markdown rendering for suggested prompts in welcome and advanced chat, and configurable guided prompts per-agent

## Backend

### Agent Capabilities

- **Multimodal Graph Agent Support** — route multimodal (file-uploaded) chat turns through `LCBaseAgentGraphLLMRunner` when `ENABLE_GRAPH_LANGCHAIN_AGENTS=True`, enabling multimodal content to flow through the same graph (tools, RAG, middleware) as text-only conversations via new `_resolve_multimodal_llm` and `_configure_lc_ai_mentor` override hooks
- **Enhanced File Upload Grading** — `GradeSubmissionTool` now accepts file attachments alongside or instead of text submissions, routing to multimodal-capable LLMs via existing `get_multimodal_llm` resolution with structured output mechanisms for comprehensive assessment capabilities
- **Conversation-Mode Grading** — `MentorGraderConfiguration.grading_mode` now supports conversation mode via dedicated `grade_conversation` tool, enabling holistic assessment of entire chat sessions rather than individual submissions

### API Infrastructure

- **Grader Configuration CRUD API** — new admin endpoints `mentors/{uuid}/grader-config/` and `mentors/{uuid}/grader-config/criteria/` for managing `MentorGraderConfiguration` and `GraderCriterion` without Django admin, including paginated criteria lists with search capabilities
- **Enhanced Chat History Processing** — `aretrieve_truncated_message_history` now reattaches `ChatFiles` to historical human messages as OpenAI-format content blocks, preserving visual context across multi-turn multimodal conversations while maintaining text-only storage
- **Improved Document Processing** — upgraded `markitdown` to 0.1.5 fixing Google Docs exported `.docx` parsing in multimodal file uploads by resolving `_CustomMarkdownify` class parameter compatibility issues

### Security Capabilities

- **Field-Level Encryption Infrastructure** — implemented `EncryptedTextField`, `EncryptedJSONField`, and `EncryptedFieldsMixin` with Fernet symmetric encryption for transparent encrypted field access on all credential models including `LLMCredential`, `GlobalCredential`, `IntegrationCredential`, `ConnectedService`, `MCPServer`, and `MCPServerConnection`
- **Credential Masking in Admin** — enhanced Django admin with credential field masking for both plaintext and encrypted fields via `MASK_CREDENTIALS` `GlobalConfiguration` setting (default: `True`), protecting sensitive data exposure in admin interfaces
- **Encryption Management Commands** — new management commands `backfill_encrypted_fields`, `verify_encrypted_fields`, and `cleanup_plaintext_fields` for encryption lifecycle management with comprehensive migration tooling

### Notification System

- **Dedicated Notification Queues** — introduced `notifications.high` and `notifications.low` Celery queues with dedicated worker support, separating notification processing from AI inference and data processing workloads for improved system performance and scalability
- **Grade Posted Notifications** — new `SUBSECTION_GRADE_UPDATE` notification type automatically emailing learners when grades are posted or updated on assignments, quizzes, and exams with configurable debounce modes (`every` with 5-minute window or `first_only` per subsection)
- **Course Schedule Change Notifications** — `COURSE_SCHEDULE_CHANGE` notification type monitoring course start date, pacing mode, and certificate availability changes with automatic email dispatch to enrolled learners including old/new value details and course links

### Data Model Enhancements

- **Auto-Recharge Default Enablement** — credit accounts now enable auto-recharge by default at configurable `CREDITS_AUTO_RECHARGE_DEFAULT_AMOUNT_USD` (default $10), with payment method validation guards preventing failed Stripe charge attempts for accounts without stored payment methods
- **Report Generation Improvements** — new `MentorInteractionResearchGenerator` providing session-level engagement exports with prompt-by-prompt breakdown including user email, agent identity, session timing, token counts, and topic associations via batch-processed 1M row safety-capped pipeline
- **Enhanced API Response Handling** — reported-skills GET endpoints now return 200 with empty payload when users have no reported skills instead of 400 Bad Request, improving API consistency and client error handling

## Infrastructure

### Deployment & Operations

- **Production Service Update Workflow** — new `workflow_dispatch` workflow enabling one-click service updates across all VMs in production environments with concurrency controls, sequential execution (`max-parallel: 1`), and comprehensive rollup reporting
- **Non-Interactive Environment Setup** — `iblai infra setup-env -f .env` command enabling Ansible bootstrap from environment files without interactive prompts, supporting both provisioned and free-standing deployment modes with comprehensive `.env` schema validation
- **Environment Provisioning Automation** — `iblai infra provision-env -f .env` command for non-interactive Terraform provisioning supporting single-server deployments with auto VPN IP detection, SSH key generation/management, and ACM certificate provisioning

### Security & Configuration

- **Field Encryption Key Management** — auto-generated `FIELD_ENCRYPTION_KEY` using Fernet symmetric encryption during initial setup with rotation support via `ibl config rotate-secrets --only IBL_DM.FIELD_ENCRYPTION_KEY`
- **AI Call Health Monitoring** — exposed AI call healthcheck port via configurable `IBL_DM.AI_CALL_HEALTHCHECK_PORT` (default 8831) for enhanced service monitoring and load balancer integration
- **Dependency Management** — lockfile-driven dependency controls with explicit pins for high-drift packages (`redis`, `requests`, `charset-normalizer`, `sqlalchemy`) preventing silent version floating during dependency updates

### Integration Capabilities

- **Google SSO Configuration** — automated `google_sso_config` Ansible role creating `OAuth2ProviderConfig` rows on LMS for `python-social-auth` `google-oauth2` backend with optional organization binding and idempotent execution
- **Microsoft SSO Configuration** — `microsoft_sso_config` role handling `IBL_EDX_BASE_OAUTH_SSO_BACKEND` configuration patches, Django settings reload, and `OAuth2ProviderConfig` creation for `azuread-oauth2` integration with tenant-specific OIDC endpoints
- **Enhanced Open edX Compatibility** — patched discussions topic-link bulk updates avoiding N+1 queries, hardened bookmarks cache rebuild for orphaned blocks, and assessment mode support in mentor XBlocks with configurable UI element hiding

## Vibe Agent Skill Ecosystem

- **MCP Agent Integration** — new `iblai-agent-mcp` skill providing comprehensive Model Control Protocol connector management with add/edit/featured connector workflows and connection testing capabilities
- **Agent Audit Capabilities** — `iblai-agent-audit` skill enabling systematic agent behavior analysis, performance monitoring, and configuration validation with detailed audit trail generation
- **Credit Balance Management** — `iblai-credit` skill with real-time credit monitoring, balance display, and integration with auto-recharge workflows for seamless credit management user experiences
- **Enhanced Vibe-Starter Integration** — improved `iblai-ops-init` skill with automated platform and token configuration after vibe-starter cloning, streamlined setup workflows, and better CLI integration patterns

## REST API — New Endpoints

- `GET/POST/PUT/PATCH /api/ai-mentor/orgs/{org}/mentors/{mentor_unique_id}/grader-config/` — agent grader configuration management
- `GET/POST /api/ai-mentor/orgs/{org}/mentors/{mentor_unique_id}/grader-config/criteria/` — grader criteria collection management
- `GET/PUT/PATCH/DELETE /api/ai-mentor/orgs/{org}/mentors/{mentor_unique_id}/grader-config/criteria/{id}/` — individual grader criteria with validation and weight management
- `GET/POST /api/core/platforms/{platform_key}/demographics/fields/` — demographic field configuration per platform
- `GET /api/core/platforms/{platform_key}/demographics/fields/export/` — export demographic field configurations
- `POST /api/core/platforms/{platform_key}/demographics/fields/import/` — import demographic field configurations
- `GET/PUT/PATCH/DELETE /api/core/platforms/{platform_key}/demographics/fields/{slug}/` — individual demographic field management
- `GET /api/core/platforms/{platform_key}/{username}/demographics/` — user-specific demographic data access