---
title: "Week of March 31 – April 3, 2026"
slug: "platform-update-2026-04-04"
date: "2026-04-04"
tag: "Application"
summary: "iblai-app-cli for scaffolding apps, voice calls and screen sharing in canvas, Copy Agent across tenants, agent evaluation with LLM-as-Judge, PII privacy router with Microsoft Presidio, metadata pass-through, ECR image mirroring, and new MCP authentication docs."
author: "ibl.ai Engineering"
---

## Frontend

The new iblai-app-cli gives developers a single command to scaffold a fully functional ibl.ai app with authentication, chat, and native desktop/mobile builds via Tauri v2. In MentorAI, agents now support live voice calls and screen sharing directly in the canvas, and the new Copy Agent feature lets teams duplicate agents across tenants — including or excluding training data. SkillsAI adds per-unit progress tracking in course outlines.

- **iblai-app-cli v1.0.0** — Developer CLI for scaffolding ibl.ai apps. `iblai startapp`, `iblai add auth|chat`, Tauri v2 desktop/mobile builds (iOS, Android, Windows), Next.js 16, AI-assisted development via Claude Code and Cursor
- **Screen sharing and voice calls in canvas view** — Users can initiate voice calls and share their screen directly from the agent canvas
- **Copy Agent** — "Allow Copies" toggle and cross-tenant agent duplication, with or without training data
- **Explore page and sidebar navigation** — Tenant explore page with admin/non-admin views, "New Chat" and "Workflows" sidebar buttons
- **CSV dataset uploads** — Agents can be trained on CSV files from the dataset management UI
- **Windows installer** — NSIS .exe installer for MentorAI desktop distribution
- **Course outline progress indicators** — SkillsAI shows per-unit completion checks instead of a single icon

## Backend

This week added three capabilities that matter for enterprise agent deployments: an evaluation framework that scores agent responses using LLM-as-Judge, a privacy router powered by Microsoft Presidio that can redact or block PII before it reaches any LLM, and a metadata pass-through that lets frontends inject arbitrary context into agent conversations.

- **Agent Evaluation System** — Evaluate agent quality using datasets, experiments, and LLM-as-Judge scoring
- **Privacy router (Microsoft Presidio)** — Per-agent PII detection: redact, mask, or block names, emails, SSNs, etc. All events logged to ModerationLog
- **Generic chat metadata pass-through** — Frontends send arbitrary context (product group, level, state) alongside messages; metadata injected into LLM prompt and persisted on sessions, exports, and analytics
- **Cost tracking for self-hosted Claws** — Full cost visibility for tenants running their own OpenClaw instances
- **Memory endpoint enhancements** — user_id, start_date, end_date filters on memory endpoints; auto-creation of default memory categories
- **Course outline summary endpoint** — Lightweight structural completion data for faster loading
- **Completion percentage on course units** — completion_percentage and complete boolean per unit
- **Course sync improvements** — Filtering, dry-run mode, and progress reporting
- **Tenant OpenClaw configuration advertising** — Tenants advertise their OpenClaw configuration to connected clients

## Infrastructure

All third-party container images (Prometheus, Grafana, cAdvisor, exporters) are now mirrored to IBL's private ECR via `ibl ecr sync`, removing runtime dependencies on upstream registries. The new `resetup` and `launch` commands in iblai-infra-cli enable non-interactive environment provisioning.

- **ECR image mirroring** (`ibl ecr sync`) — Mirror third-party images to IBL private ECR with --dry-run support
- **resetup command** — Re-configure existing environments with new domains and rotated secrets
- **launch command** — Non-interactive AMI-based provisioning for CI/CD
- **launch-env command** — Launch from .env files with interactive confirmation
- **Image version pinning** — All versions controlled by iblai-images package
- **Database password sync** — PostgreSQL and MySQL passwords synced after secret rotation
- **SPA health checks** — All SPA launches verified via HTTP with retries
- **Ingress system** — Pre-provisioned domain endpoint management with lock backends
- **Dead code removal** — Removed ~2,000+ lines of unused Terraform, Debezium, Schema Registry, and lazydocker code

## Documentation

- [Agent Evaluation System](https://ibl.ai/docs/developer/agents/evaluations) — Developer guide for datasets, experiments, and LLM-as-Judge scoring
- [MCP & OAuth Architecture](https://ibl.ai/docs/developer/agents/mcp-authentication/architecture)
- [MCP Server Connections](https://ibl.ai/docs/developer/agents/mcp-authentication/mcp-connections)
- [OAuth Connector API](https://ibl.ai/docs/developer/agents/mcp-authentication/oauth-connectors)