2026.3.29 — March 2026 — Loquent Changelog

2026.3.29 is the biggest Loquent release yet. A built-in AI assistant — think ChatGPT or Claude, but with communication superpowers — lets you perform every action on the platform through natural language. After every call, an AI extracts follow-up tasks automatically. Every API endpoint and UI view now enforces role-based permissions. The entire platform is fully usable from mobile and looks stunning on any screen size. 60 PRs merged.

Loquent AI Assistant

A conversational AI that understands your data and can take action on your behalf — every action in Loquent, through natural language.

Every action, through conversation — search contacts, reply to leads, send emails and SMS, analyze call transcripts, create tasks, configure agents, provision phone numbers. Write actions show an inline preview with confirm/cancel before execution
50+ tools across every module — contacts, calls, messages, analytics, tasks, analyzers, voice agents, text agents, phone assignments, plan templates, and notifications, all permission-gated so the AI only offers actions you’re authorized to take
Multiple conversations — manage parallel assistant sessions, draft a reply to one contact while the AI researches another
Full-screen overlay mode — expand from a 400px sidebar to a viewport-filling chat with Cmd+J (sidebar) and Cmd+Shift+J (full-screen)
Page-aware context — the assistant knows which page you’re on and tailors responses accordingly
ABAC-enforced tools — the LLM only sees tools the user is authorized to use; analytics tools respect dashboard scoping
Typed tool registry — string tool names replaced with AssistantToolName enum for compile-time safety
Lazy persistence — conversations are only saved on first message, eliminating empty sessions
Icon-only trigger with breathing ring animation
UI polish — contained wide content in bubbles, removed raw JSON from tool cards, deep links on all tool results

Task Management & AI Follow-Ups

The legacy todo system is replaced with a streamlined Task model — with AI-generated follow-ups from calls and time-based reminders.

AI-generated follow-up tasks — after every call, an AI analyzes the transcript and creates actionable tasks with due dates, assigned to the contact owner. AI-generated tasks show a sparkle icon with violet badge and link back to the source call
Time-based reminders — set a due date and time on any task; a background job sends a notification when it’s due
Full task lifecycle — create, edit, and manage tasks with status (Open/Done/Dismissed), priority, category, source, due date, and assignee with status tab filtering

Complete Permission Enforcement

ABAC permissions now cover every API endpoint and every UI view across all modules.

Contact module — all 22 endpoints enforced with check_contact_access, including notes, messages, attachments, tags, and field values; granular can_edit/can_delete on notes
Call module — ViewAssigned/ListAssigned scoping so users only see calls from assigned contacts; dedicated MakeCall permission
Settings module — all 16 role, member, and phone management endpoints enforced
7 remaining modules — AccessDenied UI guards and button gating for agents, analyzers, knowledge bases, plans, plan templates, text agents, and messaging
Notifications & Twilio — all endpoints enforced including Twilio access tokens
Assistant tools — all 50+ tools gated by resource permissions; registry pre-filters by user role
Dashboard scoping — non-admin users see stats scoped to their assigned phone numbers

Mobile-Responsive UI

The entire platform is now fully usable from mobile — and it looks amazing.

Every view, rebuilt for mobile — messaging, contacts, tasks, plans, agents, analyzers, knowledge bases, settings, admin, and notifications all work beautifully down to 375px with route-based panel switching
Polished mobile navigation — sleek sidebar drawer with backdrop, auto-close on navigation. It feels native
Infinite scroll everywhere — messaging, calls, plans, tasks, and notifications use infinite scroll pagination
Correct viewport handling — replaced 100vh with 100dvh for proper mobile viewport height
Shared component library — new EntityGrid, EntityCard, and ConfirmDialog primitives standardize grids and delete confirmations across agents, analyzers, text agents, and knowledge bases

Messaging Improvements

Unanswered messages workflow — “Unanswered” filter chip with count badge, direction/time-range sub-filters, and a clickable shortcut from the dashboard KPI
Rich HTML email rendering — emails display their full HTML formatting inside message bubbles with XSS-safe server-side sanitization; toggle between rendered and plain text
Redesigned compose field — shared component with inline and expanded modal modes, SMS/Email channel toggle, character counter, markdown preview, and attachment handling

Contact Improvements

All your Oprah fields, now in Loquent — custom contact fields let you replicate every field from Oprah and beyond: text, number, date, URL, email, boolean, textarea, and select with color-coded options. Full type-specific validation on both client and server
Plans can update contacts — autonomous plans can modify contact data during execution as part of automated workflows

Plan Improvements

Smarter AI cost management — separate models for trigger evaluation (cheap/fast) vs plan execution (capable), reducing cost without sacrificing quality
Model override per template — pick a specific AI model per plan template; the AI generator suggests a model tier

UI & Visual Polish

Card depth — subtle shadows on base cards, hover elevation on clickable grids
Typography — page titles bumped to text-2xl, KPI values to text-3xl
Unified tabs — all tab navigation uses a single TabsTrigger component with underline indicators

Infrastructure

Docker fixes — replaced broken Dioxus install with cargo-binstall, fixed binary name in ENTRYPOINT
Build optimization — improved multi-worktree build performance
CI improvements — skip cargo-check hook when no .rs files changed
Workflow refactor — switched to main-as-dev with dedicated prod branch