2026.08

This release adds a dedicated PII detection and redaction engine, AI-generated slide decks, multi-agent "swarms" for AI Mind, a major security hardening pass (password policies, idle logout, SSO-in-DB), a role-scoped BI redesign, and broad reliability work across the data warehouse, Lake connector, and CI pipeline.

PII Detection & Redaction

• Dedicated PII service: A new module detects and redacts personally identifiable information, with configurable detection modes and custom label sets.
• Chat integration: AI Mind conversations now run PII detection with an admin kill switch, confirmation codes before exposing flagged content, and safe error surfacing.
• Admin configuration: PII scanning is configurable, with custom regex patterns and audit logging of every detection.

AI Mind: Swarms & Smarter Models

• Swarms: Mind conversations can now run as multi-agent "swarms," with a configurable default and per-scheduled-task settings.
• Mind Benchmarks: A new question-bank based benchmarking system, with experiments and a runner engine to track mind quality over time.
• Model catalog overhaul: New available models now carry retirement flags, tier badges (frontier/pro/lite), and an admin-configurable tenant-wide allowlist.
• Reasoning effort control: Replaced the old reasoning mode with native, per-model reasoning effort settings.
• Configurable scheduled tasks: Tasks can set their own model, reasoning effort, web search, and swarm behavior.
• Batch PDF uploads: Upload PDFs to training cards in a batch, with cancel support.

Slides (New)

• AI-generated decks: A new Mind tool builds full slide decks end-to-end, seeded directly from your conversation.
• AI motion engine: Automated chart label placement, animation, and template design for generated decks.
• Email subscriptions: Subscribe to a deck and receive it on schedule by email.

Security & Sign-In

• Password policies: Configurable lockout after repeated failed attempts, reuse restrictions, and password expiry.
• Idle logout: Configurable automatic sign-out after inactivity.
• Two-factor authentication rework: Token handling around enabling, disabling, and resetting the two-factor authentication added earlier was reworked.
• SSO: SSO configuration is now admin-editable in-app, with its own permissions.

Introducing DataPrism

BI has been rebuilt from the ground up as DataPrism.

• Semantic Models
• Role-owned workspaces
• Row-level security
• PII-aware data modes
• Measures
• AI-assisted measure generation
• Consistent formatting
• Real-time platform sync
• Smarter charts
• Bridge-table cross-filtering
• Table picker

Data Warehouse & Connectors

• Excel import improvements: Batch import, better handling of messy column names, and native cell-type support for more accurate data types.
• Column reordering: A dedicated ordering engine keeps column positions consistent as you edit.
• Legacy SQL Server support: Optional support for SQL Server 2005 and older.
• MySQL/MariaDB improvements: New structure detection and full MySQL 8+ compatibility.
• ClickHouse: SSL support, inline syntax checking in the editor, and streaming exports that no longer risk memory exhaustion on large files.
• Lake connector: Protected credential files, revert/redo for connector-building AI chats, and partition-level management for Lake tables.

User Management

• Avatar uploads: Users - and permitted admins on their behalf - can upload and update profile avatars.
• Batch delete: Remove multiple users in a single action.
• Expanded profiles: Additional profile fields for a more complete user record.
• Custom color themes: Users can create, edit, and favorite their own color themes alongside the built-in Light and Dark options.
• Organization branding: Admins can publish a "brand" theme that sets the default palette for everyone, gated by its own permission; Company Name is configurable again in system settings and auto-fills the {{company}} placeholder platform-wide.

DataMind Platform

• Model providers (New): Every model your company can call now lists the providers that can serve it, with input, output, and cache-read prices, uptime, latency, throughput, and the discount where one runs. Your settings choose how DataMind routes a request: throughput, latency, price, or balanced.
• Lower pay-as-you-go rates: The rate for every pay-as-you-go model was reduced.
• Usage log detail: The request log now names the provider and the region behind each request, with its latency, billed tokens, billed cost, and caching saving. On a request to one of your plan's models it shows a region label instead of the provider.
• Standard plan overage: The tokens the standard plan includes are now billed like the premium plan's, with an overage rate on the invoice; the spend cap gates the overage.
• Credit (New): DataMind can grant your company credit, which appears on your billing timeline and is offset against your invoices.
• Free prompt tokens (New): The assistant instructions DataMind adds to a request for one of your plan's models are measured and kept off your bill.
• Sign-up requests (New): Request an account for a new company from the sign-up page; DataMind reviews every request, and the requester can check its status.
• Longer context and thinking (New): Your plan's models accept up to 524 288 tokens of context, and a request chooses whether the model thinks before it answers, with its own output ceiling.