Changelog

New Front Door 🚪

The app has been rebuilt around the thing that actually matters: how fast a new enquiry gets a call back. The old menu asked you to think in features — agents, calls, tools, inbound. The new one asks you to think in a loop: leads come in, an agent calls them, the result goes somewhere useful.

• Sources — where your leads come from. Connect your website form, your business phone number, HubSpot, or Zoho CRM, and every new enquiry is called back within about a minute. No list to upload, no button to press. Each source gets its own call-back rules: how many times to try, and the hours you're happy for us to call.
• Destinations — where the results go. Set up a place for each call's details to land — a message to your team, or your own webhook pointing anywhere you like — and it's written out after every call, answered or not.
• One agent screen, four tabs. Everything about an agent now lives in one place: Profile for voice, languages, and guardrails; Context for knowledge and variables; Prompts for the conversation itself; Actions for transfers, summaries, and destinations.
• One Save, everywhere. Edits across all four tabs collect into a single draft with one Save Changes. The old two-step save inside the flow editor is gone.
• Calls are now Conversations, with the lead, the source, the outcome, and the length on every row — and the transcript, recording, summary, key topics, and captured fields one click away.
• A dashboard to open on. Your workspace balance, plan limits, and the week's call numbers, on one screen.
• Setup that asks first. New workspaces now confirm the business name, what it does, and the country, currency, and timezone up front — so calling hours, rates, and number formats are right from the first call instead of being corrected later.
• Fixed: connecting Zoho CRM no longer installs anything inside your CRM. You approve the connection on Zoho's own screen and that's the whole of it.

Same product, far shorter path from "someone enquired" to "someone called them back".

Know Your Neighbours 🌍

Calling someone is governed by where they are, not where you are. A contact in California expects a call on California hours; one in New York, three hours earlier. Until now everything ran on a single workspace clock, and that only holds while your business and your customers share one. So the product now asks two separate questions instead of one — and once it knows the answer, calling hours, what your agent says, and what you can buy all follow from it.

• Setup asks where your customers are, not just where you are. The first-run screen now has three questions: Where is your business located?, Where are the customers you'll be calling?, and What should we call your company on calls? The first sets your currency, plan and terms. The second decides the rules your calls run under. The third is what your agent says out loud, so it is the one your customer actually hears.
• Calling hours follow the contact. Where local rules require it, your campaign's time slots are read in each contact's own local time — the schedule now says so, and shows the hours you must stay inside. You can narrow that window; you can't widen it. Set it in Campaigns → schedule.
• Nobody gets called at the wrong hour. The check runs per contact at dial time, not once per campaign, so a list spanning several time zones is handled contact by contact. It applies to single test calls too, not just campaigns.
• Your agent tells the truth about being an AI. Where the law requires it, your agent says plainly that it is an AI assistant when asked — and on outbound calls it says so up front, along with the fact that the call is recorded. You'll see that opening line in the workflow editor above your own greeting, marked Said first, automatically. It isn't editable, and it can't be removed.
• Contact lists keep the numbers you gave them. Uploading contacts no longer assumes a country code for numbers typed without one, so an international list arrives as you wrote it. When a number could belong to more than one country, add the country code — the test-call dialog now asks for it directly.
• Numbers, wherever you are. In countries where we don't sell numbers yet, the Buy tab is simply gone and Bring Your Own is the way in — no dead ends, and no verification steps that wouldn't apply to you.

One workspace clock was always going to run out. Now it doesn't have to.

Just Ask 💬

Your numbers used to live behind menus and filters. Now the home screen is a chat box: ask "how many calls did I make this week?" or "how do I set up inbound?" and get a straight answer — from your own data or the product guides — without hunting for a report. And while you're building, that same plain-language help now runs through the agent editor too.

• Chat is the new home. Ask about your calls, campaigns, spend, and credits in plain language and get the real numbers back — with the assistant showing its work, so you can see exactly where an answer came from before you trust it. Tap a suggested follow-up to keep digging, or rate any answer to help it get sharper.
• Answers from the manual, too. Ask how something works — "how do I add a phone number?" — and it replies straight from the product guides with a link, so you never leave the chat to find a setup step.
• Fix your agent by describing the problem. In the workflow editor, tell the assistant what's going wrong — "it hangs up too early," "stop it promising refunds" — and it proposes concrete fixes you review and apply in one click. Point it at your recent calls so the fix targets what actually happened.
• Version history and rollback. The editor now snapshots every save and every generation as a labelled version, so you can browse your agent's full history and restore an earlier one in a click — experiment freely, roll back instantly.
• Transfer a call to a human. When a conversation needs a person, your agent hands the live call off to your team — set it up per agent, and the caller is connected without dropping.
• Guardrails. Give an agent a hard list of things it must never say, so it stays on message on every call.
• More languages. Your agents now hold conversations across a growing set of languages — 20 at last count — so you can reach customers in more markets without a separate setup for each.
• Duplicate and re-point. Copy an agent — voice, flow, knowledge, and all — to spin up a variant in seconds, and switch an existing agent between outbound, inbound, and web without rebuilding it.

Less clicking, more asking.

You've Got Mail 📬

Not everyone lives in the dashboard. Your ops lead wants the gist of the call that just happened; your founder wants one number over morning coffee. Now both land in the inbox — automatically, per agent, with nobody refreshing a screen.

• A summary after every call. The moment a conversation wraps, the people you pick get an email with the whole story — what was discussed, the next steps, the details the agent captured, and a link to the recording when there is one. No opening the app to find out how it went.
• A morning roundup. Once a day, an executive overview of yesterday lands in the inbox: how many calls, how many connected, total talk time, the busiest hour — including the ones that never connected. The whole day in ten seconds.
• Per agent, your call. Set the recipients on each agent in the workflow editor's new Email reports panel. Two separate lists — one for the after-every-call summary, one for the daily roundup — so the right people get the right level of detail. Leave a list empty to switch that email off.

Set it up in Agents → your agent → Email summary. See Configure your agent.

Right on the Record 📇

Your voice agents now work right inside Zoho CRM. Sales teams live in the CRM — every lead, every note, every follow-up. Until now, calling meant hopping over to IntrLume, running the calls, and stitching the results back by hand. That whole round trip is gone: pick an agent from the record, let campaigns work your saved views, and watch every outcome land back on the lead before your next stand-up.

• One-click calls. Open a lead or contact, pick an agent, hit call. The agent dials the record's number, has the conversation, and files the outcome — while you move to the next thing.
• Campaigns that follow your filters. Point a campaign at any saved view — "New leads, not contacted in 7 days" — and IntrLume keeps working it on your schedule. New leads that match the view get picked up automatically; nobody rebuilds a list, ever.
• Inbound, captured. Incoming calls become CRM records on their own. A known number gets its conversation logged on the existing contact; an unknown one becomes a brand-new lead — so missed calls stop being missed opportunities.
• Everything written back. The outcome, a summary, and the recording land on the record where your team actually looks. Skip-and-retry logic reads those same fields, so follow-ups stay honest.
• Fixed: a rare case where an answered call could stay silent for certain phone-number formats.

Your CRM stays the source of truth. IntrLume just makes it talk.

Key to the City 🔑

IntrLume now plays well beyond its own walls. Until today it was an app you visit; from this release it's a capability you can build into your own product and website — trigger a call when someone signs up, confirm an order the minute it's placed, or let website visitors talk instead of type.

• API access. Admins can mint a secret key and wire IntrLume into your own systems, server to server — trigger calls from your backend, pull results into your dashboards, and automate what used to be clicks. Access is enabled per organization, and keys can be rotated the moment you want the old one dead.
• Your website can talk now. Embed a web agent on your site and visitors get a real conversation instead of a contact form and a wait. A domain allowlist keeps it yours alone — the agent only runs on the sites you approve.

Numbers Talk 📊

Meet Call Analytics — your whole calling operation on one dashboard. Call logs tell you what happened on a call; this tells you what's happening across all of them: whether answer rates are slipping, which campaign is actually converting conversations, and where the short "sorry, wrong number" calls are creeping in.

• The headline numbers. Total talk time, average call length, productive-call rate, and short-call rate — with every call bucketed by outcome: answered, voicemail, no answer, busy, failed. One glance tells you if the operation is healthy.
• Trends you can act on. Daily call volume with answer and productive-rate lines, a duration distribution that shows whether conversations are real or fizzling at hello, campaign-by-campaign performance, and your busiest agents — all filterable by direction and time range (today through 90 days).
• Show-your-work mode. Every card and chart explains its own formula with a worked example, one click away. No mystery math, no arguing about what "productive" means.

Second Opinion 🧪

How good is your agent, really? Listening to random recordings doesn't scale, and gut feel doesn't survive an argument. Evaluations replace both with a rubric: pick an agent, pick real calls, and get a graded report — what's working, what's broken, and exactly what to change.

• Scored on real calls. Evaluations grade the conversations your agent actually had — accuracy (did it say true things, use its knowledge correctly, finish the job?) and experience (was it concise, did the conversation move, did it take turns like a human?).
• Working well vs. issues found. Every scored call splits into wins and misses in plain English, each with the judge's reasoning one click away — so you're never guessing why a dimension failed.
• Fixes, pre-written. Concrete prompt improvements with before/after diffs, tagged with the scores they should lift. Apply one in the editor, re-run the evaluation, and watch the number move — improvement becomes a loop, not a hunch.

Tool Time 🧰

Talk is great. Doing things mid-call is better. Until now your agents could answer questions and capture details; from this release they can act — look up a record, book the slot, update a system — while the caller is still on the line, not after a human gets around to the follow-up.

• The tools marketplace is open. Browse a growing catalog of ready-made tools and hand them to your agents — no building from scratch. Each tool is a mid-call action: fetch the order status, check the account, create the ticket.
• Attach where needed. Wire a tool to the exact conversation step that should use it and pre-fill the fixed inputs, so the agent gathers only what's missing and calls the tool at the right moment — not whenever it feels inspired.
• Fixed: agents occasionally getting stuck in the wrong language after a mid-call switch.

Paper Trail 🧾

Every call now comes with receipts. Trusting an AI with your customers starts with being able to check its work — and fixing what it reads should never mean re-uploading a document and hoping the parser behaves this time.

• Call detail, deep-dive edition. Open any call for the full story, organized in clean tabs — the transcript turn by turn, the recording, the outcome, and every piece of data your agent captured along the way. When someone asks "what did the agent actually say?", you'll have the exact answer.
• Edit your knowledge base in place. See the exact text your agent reads on calls, fix mistakes right there in the viewer, and re-index in a click. Garbled table? Outdated price? Correct it and re-test in the same minute — no re-uploading, no waiting.

First Contact 👋

IntrLume is live — voice AI agents that pick up the phone, talk like actual humans, and get real work done in the languages your customers actually speak. Describe the job, get a working agent, and put it on a real phone number or your website — the distance from idea to first live call is measured in minutes, not sprints.

• Agents that write themselves. Describe the job in one line — "book product demos, confirm the slot, handle reschedules" — and IntrLume drafts the greeting, the instructions, and the entire conversation flow. Fine-tune anything later in a drag-and-drop editor where every step of the conversation is a box you can open.
• Fluent in your customer. A growing lineup of languages, each in its own script — and the agent switches mid-sentence the moment your caller does. No "press 2 for your language."
• Three channels, one agent. Dial out, pick up, or drop a talk-to-us button on your site. Outbound, inbound, and web — same brain, same knowledge, wherever the conversation happens.
• Campaigns on autopilot. Point one at a contact list, set the calling window and days, and IntrLume works the list on schedule — pacing the dials and retrying the no-answers so you don't have to babysit it.
• An agent that's done the reading. Hand it your documents and it answers from them live on the call — your pricing, your policies, your product — instead of improvising.
• Receipts included. Call logs with the transcript, recording, outcome, and captured details of every conversation, from day one.

Welcome aboard. The fun part starts now.