New features, improvements, and fixes across the IrisAgent platform. Updated regularly.
When an article contains multiple images, case answers now pick the ones that actually match the step being explained instead of the first image found. Images sourced from Confluence and Jira also render correctly in the Salesforce sidebar.
Agents can now send a case reply or leave an internal private note on a Salesforce case without leaving the IrisAgent sidebar. The sidebar also gets a single launcher for both Support Analyst and IrisGPT, and keeps your chat going as you move between cases.
Three rounds of benchmarking across 15 models led to a dozen production model swaps, including newly added Kimi K3 and DeepSeek V4 Flash. Every customer gets the higher-scoring model for each task with no configuration change on your side.
Every conversation now carries an editable contact record, and outbound email replies show up in the same thread as the chat. Conversations are also labeled by source, so you can tell chat, voice, and email traffic apart at a glance.
You can now set answer guidelines for Voice AI from Deploy settings, shaping tone and what the assistant should or should not say on a call. A new product filter scopes voice answers to a single product line, so callers only hear knowledge relevant to what they bought.
Chat and voice logs now show which procedure matched, the details the AI captured from the customer, and the workflow state it reached. New filters by chatbot configuration and agent request make it much faster to find the conversations you care about.
Install IrisAgent on your Shopify store through the Shopify app and add the chat widget with a theme app embed, no code required. The AI can then look up order status, products, and other read-only store data to answer "where is my order" questions directly.
Topics now cluster chat conversations alongside tickets, with a chat filter and a detail view for each topic showing volume, representative conversations, and insights. Chat-only teams finally get the same view of what customers are asking about.
A conversation only counts as a billable resolution when the answer was grounded in your knowledge. Chats that show red flags, such as a frustrated customer or a failed attempt, are now disqualified, so what you pay for lines up with what actually got resolved.
Support Analyst is no longer limited to ticket data. Teams running the AI chatbot can now investigate chat conversations in plain English, asking questions like why deflection dropped or which topics drive the most handoffs.
Internal data the bot collects during a conversation no longer leaks into the reply text, and clarifying questions or errors are no longer counted as answers in your metrics. When the bot genuinely has no answer, it now escalates instead of stalling, and the no-answer message is translated for the customer's language.
Support Analyst is reorganized into three clear steps: find an issue, prove it with data, then improve it. Suggested prompts are now grounded in what the system can actually answer, and you can jump straight into Support Analyst from an AutoQA review or a case with the context already loaded.
New AutoQA accounts now come with a labeled pack of default rules covering the fundamentals of a good support reply. You can start reviewing on day one and tune or replace the rules as you go, instead of writing a rule set from scratch.
Support Analyst now compiles a weekly briefing on what changed in your support data: emerging issues, notable shifts, and suggested next steps. It is read-only, so you get the summary without anyone needing to run a single query.
Each review now displays the case evidence and the surrounding messages behind every flagged moment, so you can see why a rule failed without opening the ticket. Admins can override a score when the AI got it wrong, human and AI reviews are shown separately, and you can filter reviews by sentiment, priority, and CSAT.
AutoQA scorecards now support per-rule weights, so a missed compliance step can count for more than a formatting nit. Scores reflect your team's real quality bar instead of treating every rule the same.
Point IrisAgent at your public status page from Data Sources, with an enable toggle, provider picker (Instatus today), and URL validation. Turning on live incident awareness no longer needs a setup request to our team.
Claude Opus 5 and Gemini 3.6 Flash are now supported, and our weekly benchmark moved seven production paths (chat, AutoQA, article generation, and batch case answers) onto the best-scoring model at each price point. You get the same answer quality or better, with faster responses.
Smart Operating Procedures can now include placeholders that resolve to your live status page: whether an incident is open, its name, its stage, and its link. Write the exact wording you want customers to see during an outage, and insert the placeholders with one click from the procedure editor.
During setup, your onboarding email now lists the support topics IrisAgent discovered in your data, each linking straight into the dashboard. You can see what the AI learned about your business before you go live.
The IrisAgent sidebar now runs on Experience Cloud Case record pages, and agents can copy a suggested resolution, dynamic resolution, or case summary with a single click. Asking IrisGPT a question also submits with Enter.
IrisAgent can now poll your public status page (starting with Instatus) and use active incidents to answer customers during an outage. Instead of waiting on an agent, people asking about a known issue get the current incident status right away.
Language handling got a broad upgrade: Polish and Ukrainian are now detected in the chatbot, fallback replies are written in the customer’s own language instead of defaulting to English, and long knowledge base articles and uploaded PDFs are translated in full rather than being cut off partway. Non-English customers get complete, native-language answers.
We now index each question and answer pair in an article separately, and long articles are processed in full instead of being skipped. The AI pulls the specific passage that answers the question, so both chat and case answers cite better sources.
The answer pipeline now runs its verification passes in parallel and follows a deterministic model cascade. Agents spend less time waiting for a suggested resolution to appear on a ticket.
We refreshed our model leaderboard with the latest frontier models, including OpenAI's GPT-5.6 family and Grok 4.5, so every AI answer runs on the most accurate model available. We also added automatic fallbacks so answers stay reliable even if a provider has an outage.
Impact performance cards now show the number and percentage of cases IrisAgent processed, so you can read deflection and resolution metrics against real coverage instead of guessing at the denominator.
Ask Support Analyst to grade an AI answer you already sent against the articles it cited. It checks whether the cited articles were relevant and whether every claim is supported, then tells you if a defect was a retrieval miss or a model hallucination.
The procedures table now shows impact summaries and feedback outcomes for each Smart Operating Procedure, so you can tell at a glance which procedures are helping and which need attention without opening each one.
AutoQA now shows each ticket's customer satisfaction score right alongside its AI quality score, so you can see where a low CSAT lines up with a QA miss and coach to the tickets that matter most.
A new rule-by-week heatmap shades each cell by how often that QA rule failed, so a manager can spot a problem rule or a bad week instantly and click straight into the failed reviews behind it.
You can now add reviewer notes to any AutoQA review (at the review level or on a specific rule) and share a review with teammates through a direct link and side panel, so discussing findings and coaching no longer means hunting for a ticket ID.
Map each procedure to the chatbots that should follow it, so different bots (for different products, brands, or audiences) run only the procedures that apply to them.
For customers on outcome-based pricing, resolution billing now excludes conversations that hand off to a human and internal dashboard preview chats, so you are only billed when the AI actually resolves a real customer's issue.
The ticket agent now streams its answer and shows each step live (searching tickets, drafting the answer) instead of leaving you on a spinner. It also searches your tickets by meaning, so a question like "customers frustrated about billing" finds the right conversations even without exact keyword matches.
AutoKB drafts knowledge base articles from resolved tickets, and you can now approve or reject each draft right from the dashboard, so getting new knowledge reviewed and published is a single click.
A custom API action can now run a prerequisite action first and feed its result into the next call, so you can build multi-step automations (look something up, then act on it) directly inside a procedure.
Each procedure's detail page now shows the real cases and chats where it matched and shaped the AI answer, with permalinks to those conversations, so you can see the impact of a procedure instead of guessing.
Support Analyst gained two new skills: it can suggest a Smart Operating Procedure from your tickets (with a one-click create card and how-to or action typing) and draft an AI answer to a ticket on demand. You get from raw tickets to reusable procedures and ready-to-send answers without leaving the analyst.
You can now set customer-level source URL filters so the AI chatbot only grounds answers in the pages and articles you trust, and excludes the ones you don't. Changes take effect immediately, giving you tighter control over where chat answers come from.
Customers can now attach screenshots or images to their question in the AI chatbot and AI search. IrisAgent reads the image and uses it to answer, so people can show a problem (an error screen, a receipt, a settings page) instead of struggling to describe it.
For customers on outcome-based pricing, the Billing page now shows your resolution usage, so you can see exactly how many resolutions IrisAgent delivered and what you are paying for. You only pay when the AI actually resolves a customer's issue.
AutoQA grades 100% of your tickets, and it no longer stops short on very long conversations. Reviews that previously failed to complete on lengthy tickets now finish cleanly, so your QA coverage stays complete.
Securely save and manage credit cards on file directly in your IrisAgent settings, powered by Stripe, so paying for IrisAgent is simple and seamless. Card details are handled entirely by Stripe and never touch IrisAgent's servers.
Add rewrite rules that map the words customers use (like "acct") to the terms your articles actually use (like "account") before the AI searches. This improves which articles get retrieved for both chat and case answers, and does not change the answer shown to the customer.
Support Analyst is an AI analyst that sits on top of your tickets. Ask a question in plain English and it plans and runs the right skills for you: natural-language search, ticket deep dives, sentiment analysis, instant analytics and agent performance, account overviews, conversation QA, proactive trending-issue alerts, AutoKB article drafting and knowledge-gap detection, plus audio-transcript and multilingual support.
IrisAgent now surfaces the questions your customers ask that your knowledge base can't answer, and can draft the missing article for you, so you can close coverage gaps faster.
We re-evaluated our model leaderboard and added the newest models from Anthropic, OpenAI, and Google, so every AI answer runs on the most accurate, up-to-date model available.
AutoKB drafts knowledge base articles from resolved tickets and now detects near-duplicates automatically, so your agents and customers see one clean, authoritative answer instead of several conflicting drafts.
Smart Operating Procedures now show an API panel when a connection exists so you can wire action steps without leaving the page, estimate the impact of each procedure, and catch duplicates as you create them.
Connect IrisAgent to Slack yourself in a few clicks with the new OAuth flow, deploy across multiple workspaces, and get AI case answers and updates delivered right in Slack.
Map cases to product lines directly from the UI with a new case product classifier, so routing, tagging, and analytics stay accurate without engineering help.
We redesigned the topics, team performance, and customer overview pages so you can see failed QA reviews, the biggest increases and decreases in ticket categories, and the most common tags at a glance.
AutoQA grades 100% of your tickets and now supports tag filters, date ranges, and a team performance view, so you can pinpoint quality issues and coaching opportunities across your whole team.
New customers can get set up faster with a step-by-step onboarding wizard that walks you through connecting your data sources and deploying your first AI agent.
Connect your HubSpot knowledge base and KnowledgeOwl so IrisAgent can ground answers in more of your existing content, with support for syncing deletions automatically.
IrisAgent now detects spikes in ticket categories and notifies you in Slack when a trending incident emerges, so you can get ahead of issues before they flood your queue.
