Quiet conversations go back to your AI Agent
When a teammate replied and the customer has not answered for 24 hours, the conversation goes back to the AI Agent by itself, so a customer who writes again later is answered straight away.
- NewChange the wait or turn it off for each AI Agent under Agent automation, Follow ups. A team note in the Inbox says when a conversation was handed back.
- NewWith AI follow ups on, the AI Agent also sends one follow up written from the conversation, where the channel still allows it.
Answer text messages with your AI Agent
Your AI Agent can now answer text messages sent to your US or Canadian number, and your team can text a contact from their page.
- NewRegister your business with the carriers once from Settings, Channels, SMS. We send the registration for you and tell you when it is approved.
- NewTurn texting on for a number from the Voice tab, and texts land in the Text tab of the Inbox, answered by your AI Agent.
Call one person, with your AI Agent or yourself
Ask your AI Agent to call one person now, or ring them yourself from your browser. Both are on the Voice page and on a contact's page.
- NewCall someone has your AI Agent ring one person now, and say what you asked it to, without making a campaign.
- NewCall from browser lets a teammate talk to a person through the microphone, with the AI Agent's number showing on their phone.
Teams, satisfaction ratings and broadcasts
Put teammates in teams, ask contacts to rate a conversation when it closes, and send one message to a whole contact folder. All of it is in Automations, Templates.
- NewTeams group your teammates, and workflows can send a conversation to a team by the contact's choice, their language or who is on shift.
- NewAsk contacts to rate a conversation when your team closes it, and send each score to Google Sheets or your own system.
- NewBroadcasts send one message to a contact folder by email, or in the chats of people who wrote in the last 24 hours.
- ImprovedA new conversation now hears your welcome, privacy question and language question in one message, ahead of the reply.
Contact folders, a Voice page and check ins for quiet visitors
Keep imported contacts in folders, read every call and what it cost in one place, and let an AI Agent check in when a visitor goes quiet.
- NewContacts now has Contacts and Visitors tabs. Import a CSV into a folder, then start a campaign from that folder.
- NewThe Voice page shows your calls, minutes and AI credits, with a searchable call history and transcripts you can download.
- NewAn AI Agent can hold several phone numbers, each with its own label, and you can ask for a number in a country that needs documents.
- NewCheck in when a visitor goes quiet sends your own message after a few minutes of silence, then closes the chat if they stay quiet.
- ImprovedAn applied template now shows on Sources with a way to remove it, and workflows can be turned on from one AI Agent's own page.
Your AI Agent can call your contacts
Outbound call campaigns let an AI Agent call a list of US and Canadian contacts with your script, and every call lands in the Inbox with its transcript and summary.
- NewOpen an AI Agent's Campaigns page, upload a CSV or pick contacts, write a script with {{variables}} from your columns, add a voicemail message, choose the hours, caller ID and retries, and start.
- NewContacts are called only in their own local calling hours, every call says it is an AI assistant calling for your business, and anyone who asks not to be called again goes on your Do Not Call list, which you can also edit yourself.
- NewWatch answered calls, voicemails, no answers and busy lines add up live on the campaign page, and export the results as a CSV.
Your AI Agent answers the phone
The Voice Agent answers phone calls with your AI Agent's knowledge, takes messages, books meetings and transfers callers to your team.
- NewGet a US or Canada number on an AI Agent's Channels > Voice page. Calls to it are answered at once, and every call lands in the agent's Inbox with its transcript.
- NewKeep your own number by forwarding it to the new one, and check the forwarding with one test call.
- NewSet the greeting, the voice, the questions it asks and who it transfers to, and try it from your browser with a test call before anyone rings.
Four plans, each with a 14 day free trial
OrhanAI now has four plans, from website chat on its own up to agencies running AI Agents for many clients, and every one starts with a 14 day free trial.
- NewChoose any plan and try it free for 14 days. You add a card at checkout and nothing is charged until the trial ends, and you can cancel before then.
- NewBuy an AI credit pack from the billing page on any plan when the month's AI credits run low, without changing plan.
- ImprovedThe Free plan has been retired. A workspace without a plan keeps everything it has: its AI Agents stop answering and new messages wait in the inbox until a plan is chosen.
- ImprovedYearly billing saves 20% on every plan, and replies your team writes are still always free.
Answer customer email from your own address
Business email lets an AI Agent answer the email customers send to your company address, and reply from that address in the same thread.
- NewConnect an address on your own domain, such as support@yourcompany.com, from an agent's Email channel page. Verify the domain with three DNS records, forward the inbox, and every email that arrives is answered and lands in the agent's Inbox.
- NewChoose the From name, add a signature, add a line saying the reply was written by AI, get a copy of every reply in your own mailbox, and list senders the AI should never answer.
- NewA sender address set in Branding is verified along with its domain, so email your workspace sends to its leads and visitors goes out from your own address.
Answer with Claude or GPT
On paid plans, each AI Agent can now answer with a Claude model from Anthropic or a GPT model from OpenAI instead of the default model.
- NewPick an AI Agent's model in its Settings, General: Claude Haiku, Sonnet or Opus 5.5, or GPT-6 Luna, GPT-6.1 Sol or GPT-5.6 Sol. The picker shows how many AI credits each answer spends.
- NewSave your own Anthropic or OpenAI API key under Settings, AI models, and answers on it spend no AI credits. The provider bills you directly, and the key is stored encrypted.
- NewIf the chosen model cannot answer, for example because a saved key stopped working, the default model answers instead, and whoever manages developer settings is told about the key.
Pay yearly, and use a coupon at checkout
Every paid plan can now be billed once a year as well as monthly, and a coupon code goes straight into the Stripe checkout page.
- NewChoose monthly or yearly billing on any paid plan, from the pricing page or the billing page. You can switch between them at any time.
- NewHave a coupon code? Enter it on the Stripe checkout page and the discount is shown before you pay.
- NewA yearly plan gets an email a week before it renews, so the charge never arrives unannounced.
- ImprovedPaid plans have one straightforward price on each interval. The launch price has ended, and anyone already paying keeps the price they signed up at until their renewal.
Start free, and pay for AI credits
Every workspace now starts on a Free plan with no time limit, and every plan is counted in AI credits: the responses your AI Agent sends, which you can see in your inbox.
- NewStart on the Free plan and stay on it for as long as it suits you. There is no trial clock to run out.
- ImprovedPlans count AI credits, any response your AI Agent sends on any channel. Replies your team writes are always free.
- ImprovedWhen the month's AI credits run out, the AI pauses and conversations are handed to your team automatically, so visitors reach a person instead of an error.
- NewOn a paid plan, add extra AI credits from the billing page without changing plan.
- NewNew paid subscriptions start at a launch price for their first months, and you get an email 14 days before the standard price begins.
- ImprovedBilling is monthly, with no contracts and no cancellation fees. Upgrade, downgrade or cancel from the billing page at any time.
Reliability, and the documents your legal team asks for
Every service OrhanAI depends on now has a timeout, a retry rule and a fallback, so a slow provider costs a feature rather than your whole widget. And the trust pack is published rather than available on request.
- ImprovedIf vector search is unavailable, answers fall back to keyword retrieval instead of failing.
- ImprovedIf the AI provider is unreachable, visitors get a clear message saying so and their question is saved. No credits are charged for an answer that was never generated.
- ImprovedIf file storage is unavailable, uploads are refused with a message that says text messages still work, instead of an error with no explanation.
- NewSub-processor list, data processing addendum and incident response process, all published in full.
- FixedThe inbox and connected channels loaded conversation previews with a query that read far more of the database than it needed. Both are faster, and stay fast as conversation history grows.
- NewA per workspace ceiling on public widget traffic, so one unusually busy site cannot slow everyone else's.
Messaging channels, connectable end to end
WhatsApp, Messenger and Instagram can all be connected from an agent's Channels tab.
- NewMessenger and Instagram connect through one Meta app and one webhook. Connecting Instagram against a Page with no professional account is refused at the form, with the reason.
Data retention, and a truthful status page
A workspace can set how long conversations and audit records are kept, and the public status page reports every dependency rather than three of them.
- NewRetention windows for conversations and audit records, off unless you set one. Floors apply so a policy cannot delete live data.
- FixedThe status page could report all systems operational while file storage was down. It now derives from the same checks the application itself uses.
- FixedAuth pages loaded slowly on mobile because the logo was being served at its full source resolution.
Answer quality, and clearer plan limits
Visitors can rate an answer, and a bad one can become approved knowledge in a couple of clicks.
- NewThumbs up and down on widget answers, a review queue for the unhelpful ones, and one click to turn a bad answer into a training source.
- NewA failed payment now starts a grace period and emails you, rather than silently marking the subscription past due.
- ImprovedLimit messages name your plan and the tier that lifts the limit.
- ImprovedThe brand green moved to a shade that meets WCAG AA contrast as link text and as a button fill.

