What's new in TR-AI

More accurate auto-sends, a new dashboard layout and production hardening

• ImprovedContradiction check: an answer that says the opposite of your help article no longer reaches auto-send
• ImprovedTickets with several questions must have every question answered before a reply is sent automatically
• ImprovedPast tickets are cleaned of names, emails, phone numbers and amounts before TR-AI learns from them, and holding replies or reopened tickets are no longer learned
• ImprovedBetter non-English handling, dated sources (newer information wins) and text log attachments read into the answer
• ImprovedError reports now go through your knowledge base and site search instead of generic diagnostic questions
• NewNew dashboard layout: sidebar navigation grouped by Daily, Setup, Improve and Advanced, with sign-out and workspace switcher
• ImprovedSign-up checks your help-desk login before you continue, and the overview now leads with your review queue and ticket log
• ImprovedPlain-language error messages, a single place to set confidence thresholds, and a live “first ticket received” check for every help desk
• ImprovedSafer deploys, automatic data clean-up for more record types, admin alerts for billing and job failures, and lower memory use

Hardening release: safer known-issue handling, clarifying questions on every channel, connected-app controls and production stability

• ImprovedKnown-issue replies now run after safety screening, PII redaction and escalation rules, and never promise updates on channels that can't receive them
• ImprovedChat, SMS and voice customers now get clarifying questions (order number, identity checks) instead of an immediate hand-off
• ImprovedShopping assistant steps aside for order problems, billing and complaints so the support engine answers them
• NewSettings → Connected AI apps: see and revoke every AI assistant connected over MCP; access ends automatically when a member is removed
• NewAI-suggested knowledge base articles from MCP wait in a review queue before they are used in answers
• ImprovedAnswer cache respects channel style, supervisor checks, release incidents and reopen tracking
• ImprovedStability: one browser crawl at a time server-wide, bounded database connections, size-capped chat screenshots
• SecurityDaily caps on the ChatGPT app, single-use calendar sign-in links, expiring booking links and safer token refresh
• FixedPay-per-resolution now bills only verified AI resolutions across every channel

Known issues with automatic fix updates, domain learning that maps your product and writes workflows, and more accurate retrieval

• NewKnown issues — tickets matching an active issue get a “known issue, being worked on” reply with workaround / ETA on every channel
• NewBack-matching: find affected open tickets when an issue is created, approve before anything is sent
• NewAutomatic update to every affected customer when an issue is resolved, with auto-close policies
• NewAuto-detected known-issue suggestions from spikes, the AI analyst and repeated failing answers (Bronze+)
• NewKnown issues API, outbound webhooks, MCP tool and public JSON status feed; earlier incidents move into known issues automatically
• ImprovedDomain learning explores menus, tabs, “load more” and detail dialogs, and records a card per screen with menu path, fields and buttons
• NewAI-written step-by-step product workflows with exact click-paths, a product map and an editor; re-training only processes changed screens
• ImprovedRetrieval: keyword + meaning scoring and ranked prompt order — right article ranked first 91.7% → 94.5% (internal 109-question benchmark)
• ImprovedCrawling removes repeated boilerplate and link lists, keeps short sections with their headings and skips canonical duplicates — answer passage in top 3 93.6% → 95.4% (same benchmark)

A sales & revenue agent, screenshots on every desk, customer memory everywhere, opt-in SMS/WhatsApp notices and OAuth for AI tools

• NewSales & revenue agent — lead qualification with scoring and HubSpot / Salesforce sync
• NewMeeting booking on your own Google Workspace or Microsoft 365 (organisation-wide or per-rep sign-in, Meet / Teams links), or Calendly / HubSpot Meetings links
• NewRetention offers before a subscription is cancelled (pause, downgrade, skip, capped coupon)
• NewHosted payment links (Stripe, Shopify, Chargebee) — card details never collected in chat or on calls
• NewRevenue page: leads, meetings, saves and payment conversions
• NewScreenshots read on HubSpot, Salesforce, Zoho Desk, ServiceNow and Jira, plus image upload in the chat widget
• ImprovedCustomer memory now works with all eight help desks
• NewOpt-in proactive incident notices and check-ins on SMS and WhatsApp (numbers stored encrypted only while on)
• NewOAuth sign-in for the MCP server and the ChatGPT app
• ImprovedDomain learning explores your product in a headless browser — JavaScript screens, tabs and every area of the app

Screenshots, simulated customers, staged releases, Slack / Teams / Instagram, proactive notices, an AI analyst, an MCP server, phone-agent upgrades, commerce connectors and more

• NewScreenshot understanding — error messages and labels in customers' screenshots are read and used to find the answer
• NewLive translation for agents in the copilot and the browser extension
• NewOne-click publishing of drafted articles to Zendesk Guide, Freshdesk, Intercom and Salesforce Knowledge
• NewCustomer memory built only from tickets your team resolved
• NewSimulated customers: AI customers hold full conversations with TR-AI and every check is graded
• NewStaged releases: test a bundle of setting changes, roll out by topic, publish and roll back
• NewSlack and Microsoft Teams channels, and Instagram DMs (Bronze+)
• NewProactive incident notices to affected customers, and optional check-ins on quiet conversations (Bronze+)
• NewAI analyst: ask your data, an hourly monitor with likely causes and fixes, and e-mail digests (Bronze+)
• NewMCP server for Claude, ChatGPT and Cursor (Growth Pro), and a customer-facing ChatGPT app
• NewPhone agent upgrades: interruptions, speech models, pronunciation glossary, language switching, sentiment hand-off, outbound calls, live take-over and caller ratings (Growth Pro)
• NewConnectors for Loop Returns, Recharge, Skio, WooCommerce, BigCommerce, Chargebee, Paddle and Recurly
• NewShopping assistant in the chat widget with product cards, cart links and discount codes (Growth Pro)
• NewKnowledge sources: SharePoint / OneDrive, Google Drive, Guru and Document360
• NewProcedures drafted from your SOP documents
• NewDecision traces for every ticket, an optional supervisor check before sending and SIEM audit streaming (Growth Pro)
• NewAI-agent detection, predicted CSAT, several support inboxes, IT helpdesk mode and industry template packs
• NewAgent assist in the copilot: a step-by-step plan, suggested replies and ticket field values
• NewWorkforce planning, replies to Facebook / Instagram comments and Google / Trustpilot reviews (Bronze+)
• NewOptional pay-per-resolution plan: answers that didn't work are never billed
• ImprovedGetting-live checklist, automatic Zendesk webhook setup, review queue always visible, and banners for broken help-desk logins and the daily limit
• ImprovedLower AI cost: paid web search and conflict checks only when they can change the outcome, and real per-ticket cost tracking
• ImprovedFollow-ups: several customer messages answered together, e-mail replies answered with the conversation in view
• SecurityStricter checks on every URL you enter, customer-data lookups tied to the verified customer, encrypted webhook keys
• FixedWebhook deliveries are saved and retried, so a restart never loses a ticket

Live chat takeover, SMS / WhatsApp / Messenger, brand voice and multibrand, A/B experiments, a copilot browser extension and AI-drafted procedures

• NewLive agent takeover in the chat widget — during your live-chat hours agents reply on the Zendesk or Freshdesk ticket and the reply appears in the chat
• NewHand-off tickets with the full transcript in all 8 help desks, for the chat widget, voice agent and messaging channels
• NewSMS and WhatsApp (your Twilio account) and Facebook Messenger channels, with STOP / START opt-outs and hand-off to a person (Bronze+)
• NewVoice agent extras — text the caller the links from an answer (with their consent) and voicemail transcripts on the ticket
• NewBrand voice — tone, formality, reply length, words to use and avoid, greeting, sign-off and emoji, with a live preview
• NewMultiple brands, each with its own voice and knowledge sources, matched by help-desk brand, recipient address, widget or phone number (Growth Pro)
• NewLive A/B experiments of engine settings with significance tests, auto-stop guardrails and one-click roll-out (Growth Pro)
• NewCopilot browser extension for Chrome and Edge — suggested replies and sources next to the ticket in all 8 help desks
• NewAI-drafted procedures — from a description, from recurring tickets in your history, or improvements from real runs; test on past tickets before enabling

Procedures with live-data connectors, a website chat widget and a phone voice agent

• NewProcedures — plain-English, multi-step policies TR-AI follows using live data; lookups run instantly, changes and refunds wait for approval with amount caps
• NewConnectors for Shopify (orders, tracking, cancel, refund), Stripe (customers, invoices, subscriptions, refunds) and Jira (create and find issues); any MCP server on Growth Pro
• NewWebsite chat widget — one script tag, answers from your knowledge base, hand-off to Zendesk or Freshdesk (or email) with the full transcript
• NewPhone voice agent — spoken answers on your Twilio number, transfer to your team in business hours, voicemail tickets after hours (Growth Pro)
• FixedZendesk follow-up replies are answered again — outdated follow-up triggers are repaired automatically

Regression tests, smart triage, quality scoring, topic insights and a sharper, more consistent resolution engine

• NewRegression tests & settings history — every setting change is versioned; rerun saved test sets after any change and compare answers case by case
• NewSmart triage — custom intents, sentiment, urgency, language and extracted details (order numbers, error codes, versions, plans) written into Zendesk and Freshdesk fields and tags
• NewQuality score — every sent AI and human reply scored on an editable rubric, with per-agent views and a needs-attention list
• NewTopic insights — tickets grouped into topics with volume, trend, auto-resolution rate, knowledge coverage, estimated CSAT and an “Automate next” list
• ImprovedHybrid retrieval: keyword search alongside meaning-based search for the knowledge base and live site search; past resolutions are also matched question-to-question
• ImprovedCalibrated confidence scoring, and replies are sent automatically when the AI is at least 82% confident on a grounded, on-topic answer
• ImprovedRepeated questions get the same answer and score; paid web search runs only when the knowledge base, past resolutions and free site search all fall short of 82%
• FixedOnly tickets resolved by people are learned from — replies TR-AI posted are recognised and skipped

ServiceNow and Jira Service Management, per-customer encryption keys, live status page and full audit trail

• NewServiceNow integration — Incident table support: auto-answer, internal work notes, resolve with a close code, triage to an assignment group, and history import including work notes
• NewJira Service Management integration — requests in a service desk: public and internal comments, resolve via the Done transition, and a triage label for human hand-off
• SecurityPer-customer encryption keys: each account's secrets are encrypted under its own data key, wrapped by AWS KMS and rotated every 90 days
• NewFull audit trail per ticket — sources with links, confidence breakdown, outcome and every agent override — with CSV export
• NewLive status page with per-minute monitoring, measured 30- and 90-day uptime and automatically recorded incidents
• NewConfigurable data retention (24 months by default, 1 to 84 months) and a 12-month security audit log
• ImprovedChurn prevention: rule-based risk levels (negative sentiment trend across the last 10 tickets, more than two escalations in 30 days, unresolved ticket clusters) with the account manager CC'd on alerts
• ImprovedKB UI drift detection now compares article screenshots with live screenshots of your product
• ImprovedRelease feeds are ingested automatically every 15 minutes

Multi-source intelligence upgrade and confidence scoring overhaul

• NewReal-time web intelligence layer now queries live documentation and vendor release notes before resolving tickets
• NewConfidence scoring v2 — per-sentence confidence weighting replaces single-score model, reducing false positives by 31%
• FixedZendesk webhook retry logic now correctly handles 429 rate-limit responses without dropping tickets

Stability fixes and Freshdesk compatibility update

• FixedFreshdesk API v2 pagination now correctly handles cursors beyond 10,000 ticket history records
• FixedResolved edge case where tickets with empty subject lines caused classification to hang
• ImprovedDashboard load time reduced by 40% through query optimisation on the resolution analytics panel

Sentiment-based escalation routing and multilingual expansion

• NewSentiment analysis engine automatically flags emotionally charged tickets for immediate human escalation
• NewEscalation reason codes now included in all handoff payloads — agents see exactly why the AI escalated
• ImprovedMultilingual resolution expanded to 30+ languages, including Arabic, Hindi, and Japanese

Security patch and API rate limit improvements

• SecurityPatched potential information disclosure in ticket history API endpoint — all customers auto-updated
• ImprovedAPI rate limits increased (now up to 600 requests/minute per key on Growth Pro)
• FixedWebhook signature verification now correctly handles HMAC-SHA256 payloads from Intercom

Analytics dashboard v2 and custom escalation rules

• NewAnalytics dashboard rebuilt from scratch — automation rate trends, CSAT by resolution type, and agent time savings in one view
• NewCustom escalation rules engine — define your own conditions for when AI should hand off to specific agent groups
• ImprovedOnboarding wizard now completes knowledge base indexing in the background — no waiting before go-live