Documentation Index

Fetch the complete documentation index at: https://kb.expedient.com/llms.txt

Use this file to discover all available pages before exploring further.

AI CTRL - 0.10.2 Release Notes

Prev Next

# AI CTRL - 0.10.2 Release Notes

What's New - AI CTRL Platform Update to 0.10.2

This release is about giving the AI better context, giving admins finer control, and making everything feel faster. Here's everything that's new.

Highlights

  • Automatic Context Compaction - Long conversations summarize themselves so they keep running smoothly
  • Reworked Memory System - A clear split between what the AI remembers about you and what it remembers about a conversation
  • Streamed Reasoning - Watch the model's thinking render live as it works
  • Faster Search & Rendering - Noticeably quicker across large knowledge bases
  • Granular Admin Permissions - Separate controls for chat import, webhooks, and skills
  • Richer Usage Analytics - Filter by model tag and date range, and sort by any column

New Features

Automatic Context Compaction

Long conversations no longer slow to a crawl. Once a chat reaches a configurable length, the platform can automatically summarize what came before, keeping the conversation responsive without losing the thread. This is off by default; your administrator can enable it and tune when it kicks in.

Reworked Memory System

Memory now draws a clear line between long-term personal memories and per-conversation context, with structured controls to add, update, or delete what's stored. Automatic memories have also been tightened up, focusing on durable preferences and goals rather than one-off details that were never going to be useful again.

Administrators get a new option here too: memory tools can stay available to users while stored memories are excluded from being pulled into every conversation automatically. And if someone hasn't set a personal memory preference, it now follows your organization's configured default rather than falling back to a platform setting.

Streamed Reasoning Display

You can now watch the model's reasoning render live as it works through a problem, rather than waiting for a finished answer to appear. The reasoning is preserved in exports and conversation overviews, and tool calls and search steps now render more accurately and legibly alongside it.

Performance at Scale

Search across your knowledge base runs significantly faster, even on large collections. Math and Markdown content render noticeably quicker, reading and searching large files is lighter on the browser, and model and tool lists no longer stall when a connected service is slow or unreachable.

Admin Controls & Analytics

Usage analytics gain model tag filtering, sortable columns, and date-range filtering. Permissions become more granular, with separate controls for Chat Import, User Webhooks, and Skills Import/Export — plus one-click restore to put group and default permissions back to their original state. You can also set a custom prompt template for chat autocomplete, and optionally require users to confirm before a web search runs.

Chat Experience Improvements

A collection of quality-of-life upgrades that make every conversation smoother:

  • Last-Used Model Pre-Selected - New chats default to the model you used most recently
  • Recurring Calendar Events - The calendar event editor now supports repeat scheduling
  • Prompt & Automation Shortcuts - Quick Edit for prompts and Clone for automations
  • Draggable Sidebar Items - Reorder pinned notes, and drag folders, notes, and models straight into the chat input
  • Better Chat Search - A right-click context menu and content snippets make results easier to scan
  • Voice Suggestions in the Model Editor - Pick from a dropdown of available text-to-speech voices when configuring a model
  • Bulk Unshare - One-click "unshare all," with chat and file counts shown directly in dialog titles

Building & Configuring Models

  • Easier Tool and Knowledge Attachment - Tag suggestions, a unified model picker, and searchable pickers simplify attaching actions, filters, tools, knowledge, and skills
  • Improved Formatting Controls - Better Markdown rendering, and dropdown-style configuration for tool settings
  • Clearer High-Contrast Sidebar - Stronger visual distinction for the active chat

Changed

  • Native tool calling is now the default. Chats and models without an explicitly set mode now use the newer Native tool-calling behavior. The previous behavior is still available and can be re-selected manually as Legacy.
  • Beta labels removed. Memories, Notes, Channels, and High Contrast Mode are now fully supported features.

Security Enhancements

  • Hardened access controls across knowledge base uploads, file edits, image editing, folders, and API keys
  • Improved protections against spoofed activity in collaborative notes and channels
  • More consistent sign-in behavior, preventing account-existence detection
  • Automatic sign-out is no longer triggered by temporary, transient errors
  • Image generation prompts are no longer logged by default, improving data privacy
  • Channel member lists no longer expose private profile information
  • Strengthened protections against malicious links and unsafe web content

Stability & Reliability

  • Chat history recovers automatically from corrupted or malformed states
  • Chat saves now merge instead of overwrite, preventing message loss from concurrent edits
  • Channel messages and pinned items sync correctly and live for all members
  • Chats recover automatically after a missed connection, such as a backgrounded mobile app
  • Fixed false "unread" indicators triggered by automatic, non-user chat changes
  • Fixed sync and display issues affecting pinned notes, tags, dropdowns, and the code editor
  • Fixed image editing failures affecting certain iPhone photo formats
  • Fixed mobile input issues, including the iOS Enter key and horizontal scrolling with many tools enabled
  • Large or complex math no longer crashes chat rendering
  • Saved settings — model selection, theme, default parameters — now persist reliably across sessions and restarts
  • Tool dialogs no longer hang when you click outside them
  • More reliable code execution and sandbox startup
  • Chat search now correctly finds structured-output messages
  • Fixed blank notification previews
  • Sidebar chat timestamps now reflect last activity rather than creation time

Localization

Translation updates for Catalan, Brazilian Portuguese, Irish, German, and Spanish, with expanded Thai coverage.

Have questions about any of these features or need help getting started? Reach out to your AI CTRL account team; We're happy to walk you through it