Features

Conversations, leads captured and handoffs to your team, each next to the period before. You also see your most common questions, your busiest hours and how many chats the assistant handled without your team. Top questions are only those asked at least twice, and never anything that contains an email address or a number that looks like a phone or account number. Totals come from daily counters, so deleting a conversation does not lower them.

The dashboard gives you a link that runs the real widget on your own website in preview mode. Preview chats do not use your monthly credits, are limited to 60 visitor messages and are erased when you close the chat. A checklist on the same page walks you through verifying your website, connecting a key, adjusting the widget and installing it.

When a visitor asks for a person, the assistant stops answering that conversation at once, and everything the visitor writes goes to your team. When your team marks it resolved, the assistant takes over again. Outside your business hours the visitor is told that someone will follow up. If nobody has replied within 5 minutes, Serlion sends a second alert. You can also choose that the assistant handles everything, in which case visitors who ask for a person can leave a message.

Off until you switch it on. The assistant offers to connect a person once per conversation when a message looks urgent or angry. It decides with a list of English keywords, such as “unacceptable” or “lawyer”, or several exclamation marks in a row. It is not sentiment analysis and it does not recognise other languages.

Open, unassigned, mine and resolved views. Assign a conversation to a team member, or let the first person who replies take it. Write private notes that neither the visitor nor the assistant sees, and keep up to 50 saved replies. Updates arrive live. The visitor sees the first name of whoever replies. On Free and Starter there are no seats besides the owner, so only the owner works the Inbox.

Each person on the team can get a private Telegram message, and Pro adds alerts in Slack, Teams, Google Chat or Discord. Both are described on the Connectors page.

Every conversation is kept for a period your plan sets: 15 days on Free, 90 on Starter, 180 on Pro, 365 on Scale, and 180 or 365 on Custom. Lanus Agent can shorten that to 7 or 30 days. The list is paged, and search and filters work on the pages you have loaded.

When a visitor writes an email address or phone number in a message, Serlion saves it as a lead tied to that conversation, up to five per conversation. Email addresses are always recognised. Phone numbers are recognised in US and Canadian formats, not in European ones, so for those use the pre-chat form, which Lanus Agent can switch on. A lead has a name, email, phone, the message it came with, its source and the time. Every new lead emails the account owner, up to 10 emails an hour, and any more in that hour arrive in one summary.

Webhooks, your CRM, your newsletter (only with consent), team chat alerts and the owner email all get a new lead, and the chat never waits for them. The Connectors page explains each one.

Visitors can mark each reply as helpful or not helpful and add a reason. The dashboard shows the share of each and the latest ten of both kinds with the question that led to them. The written reason is stored with the conversation but not shown in the dashboard.

The owner, co-owners and admins can delete a conversation. That erases the transcript, its files, its lead and its shop lookups, and queues deletion of the copies in your CRM, help desk and Lanus Agent. Visitors can delete their own chat too, which does the same and is on for everyone for now.

A payment card number typed into a message is replaced by its last four digits before the message is stored or reaches the AI. This is always on. It recognises normal 13 to 19 digit card numbers, so ask visitors not to type card details into the chat.

Download conversations as a CSV with one row per message, and leads as a CSV. Both contain only what is still within your retention period. There is no export of the whole account.

Choose the colours of the chat bubble, the messages and the window, upload a logo, and pick one of five launcher icons. Place the chat in any of the four corners as a floating bubble, or put it inside your page as a section. Sizes are small, medium and large, or your own width and height. You can switch the copy, like and dislike buttons and the notification sound on or off.

A key can override the assistant’s name, greeting, voice, colours, position, size, mode, logo and icon, which helps when one account runs several websites. Without the override, every key follows the account settings.

The footer badge is off on every plan until you turn it on. If you do, you can change its text and link.

The widget’s buttons and labels are translated into nine languages: English, German, French, Spanish, Portuguese, Hindi, Japanese, Chinese and Arabic. The assistant can answer in 47 languages, and by default it replies in the language the visitor writes in. For a language without translated buttons, the buttons follow the visitor’s browser language if it is one of the nine, and are English otherwise. Arabic, Hebrew, Persian and Urdu are laid out right to left.

Give different pages a different greeting, for one page or for everything under it, such as /blog/*, with up to 10 rules. Add up to four quick-reply buttons under the greeting for the questions you already expect. A greeting is chosen when the page loads, so on a site that changes address without reloading, it stays the one for the first page.

A small bubble beside the chat button appears after a delay, after a scroll depth you choose, or when a desktop visitor’s mouse leaves through the top of the window. It never opens the chat by itself. The visitor has to click it, and it shows once per browser tab. From Starter, the AI can write a message for each page (5 pages on Starter, 25 on Pro, 100 on Scale). You approve each draft unless you turn on auto-publish. Before anything is shown, code drops messages that state a price, deadline or guarantee you never gave, contain a link or phone number, claim to be watching the visitor or claim to be human. A bubble alone does not start or count a conversation.

A weekly schedule with a time zone and an optional message for outside hours. It changes only what visitors read when they ask for a person. The assistant answers around the clock either way.

Pick one of five voices, Professional, Warm, Playful, Concise or Concierge, or write your own, and give the assistant a name. How Lanus behaves, and what it will not do, is on the Lanus page.

Add knowledge by pasting text, importing one web page by its link, or uploading a PDF or Word file of up to 10 MB. Each entry holds up to 8,000 characters, and the number of entries depends on the plan: 3 on Free, 10 on Starter, 25 on Pro, 50 on Scale and 25 to 500 on Custom. A link import reads one page and does not run JavaScript, so it is not a crawler, and it updates only when you ask it to read the page again.

Your text is cut into passages. For each visitor message the closest passages are looked up, the best five go to the assistant, and it is told never to invent facts about your business. The answer does not show sources to the visitor. If you have more entries than your plan allows, nothing is deleted, and the assistant answers from the oldest up to the limit. Help Center articles and Notion pages count separately, on Pro and up. Lanus Agent can list the questions your assistant could not answer well, so you know what to add.

A flow is made of steps: a message, a question with buttons, a product card, log in and create account buttons, an order status form, a checklist, a handoff to a person, a return to the AI, and a closing message. There are templates for a start menu with common questions, a product page helper, an order tracking form and a checklist. You can have 10 flows with up to 30 steps each.

A flow can open from a button under the greeting, on the product pages you list, or right after a visitor signs in to your site, which needs your site to tell the widget who signed in. For sign-in flows you choose who sees them: everyone who signs in, only people who came through the chat’s link, or only people who talked to the assistant.

These three steps need Pro. On a lower plan the flow is kept and shown without them. Order status also needs a WooCommerce connection. The Connectors page explains shops.

Telegram, WhatsApp, LINE, Discord and email, HubSpot and Pipedrive, Zendesk and Freshdesk, Mailchimp and Brevo, WooCommerce, Notion, Cal.com, Telegram alerts, team chat alerts, and webhooks. Shopify shops get product cards without connecting anything. The chat apps and Telegram alerts start on Starter. The rest start on Pro, and Cal.com is on every plan. Each is explained on the Connectors page, with how it works and where it stops.

You verify a website in one of three ways: a DNS record, a meta tag on your home page, or a small file on your site. One verified owner per domain, and the proof covers the www address and subdomains.

The lock lets only your verified websites run your widget, and it switches on when your first website is verified. Each key can also list up to 20 allowed domains. The lock reads the website address the browser reports, which stops other sites from using your key in normal use but not a script that fakes that address.

Serlion reads the public pages of a verified website, up to 12 of them, and proposes the chat colour, the voice, the kind of business, facts the assistant can answer with and pages for its knowledge. It skips sign-in and basket pages and follows your robots.txt. Each suggestion comes with its reason and the page it came from, you preview and apply what you like, and you can undo it for 30 days. Nothing is applied by reading.

The install code is two small blocks you paste before the closing body tag. Guides cover plain HTML, WordPress, Shopify, Webflow and Squarespace. The widget is plain JavaScript with no dependencies. A key can have an expiry date, its own monthly conversation limit and its own list of allowed domains. You get 3 keys on Free, 10 on Starter, 40 on Pro and 100 on Scale.

The Free plan answers with your own key. On the paid plans you choose between your own key and Serlion’s AI, where you need no key at all. Serlion’s AI is spread across seven providers, so if one is rate limited or down the request moves to the next. With your own key you can use Claude, GPT, Gemini, Groq, OpenRouter, Together AI or Fireworks AI. Keys are stored encrypted and checked with the provider before they are saved.

Free, Starter, Pro, Scale and Custom differ in how many credits a month they include and in what they connect to. One chatbot conversation uses one credit, and the Serlion Agent draws on the same credits. Pricing lists every number.

There are four roles. The owner can do everything. A co-owner can do everything except billing, legal matters and making co-owners, and needs two-step sign-in. An admin manages settings and the team but not keys, connections or billing. An agent can only work conversations: read, reply, resolve, assign, take notes, use saved replies, see leads and feedback, and export. Seats besides the owner are 3 on Pro, 10 on Scale and 3 to 100 on Custom. An invitation is valid for 72 hours.

Sign in with a password or with Google. Each person can switch on two-step sign-in with an authenticator app and gets 10 single-use recovery codes. Password reset links work for 60 minutes. Only the owner can delete the account, by typing DELETE and a code sent by email. Security and privacy has the details.

Read your conversations, leads, analytics and knowledge base, create leads, and manage knowledge entries and webhooks, from your own server. A key that only starts chats works on every plan. Keys are shown once and stored only as a hash.

Seven events, from conversation started to conversation deleted, are sent to your address, signed so your server can verify them. Failed deliveries are tried again over about 20 hours, and every attempt is logged for 30 days.

The assistant can call your own API in the middle of a conversation, for example to look up an order. Actions only read. You can have 10, and the assistant fills in only the parameters you declare. It never chooses the address or the credentials.

Tell the widget who is signed in on your site with a signature your server creates, so an action can use that person’s id and no one else’s. The widget also offers open, close, toggle, identify and events for messages, leads and ratings.