Onsites AI

Most support teams never review their queue. Not because it is hard, but because the review has no shape. Here is one that fits in five minutes and produces exactly three decisions. The five minutes Minute 1–2: what is unassigned? Anything older than a day gets an owner now, or a date. Minute 2–3: what repeated? Three or more conversations on the same topic become a document or a product change. Minute 3–4: what is stuck waiting on us? Not on the customer — on us. That is the real backlog. Minute 4–5: what did we promise and not deliver? One list, one owner each. What not to review Total volume, average resolution time, and CSAT with nine responses. None of them change what anyone does on Monday. Why this works Because it ends with decisions rather than a report. A review that produces a document is a cost; a review that produces three assignments is a control. States, tags and views · the complete guide Free for the first 3 seats. onsitesai.com

Two messages from the same person on the same day, in the same channel, often looks like two tickets. It is one problem — and treating it as two is how customers end up explaining themselves twice. What should happen automatically Merge on identity , not on subject. Same email or same phone means same conversation. Keep both messages visible in order, so the second one has context without a search. Keep one owner. Reopening a thread should notify the person already on it, not create a race between two agents. When merging is wrong If the second message is a different request — a new order, a separate invoice — merging buries it. The rule that works: merge by default, split explicitly, and make splitting a one-click action so nobody avoids it out of friction. Why it matters at three people Duplicate threads are cheap to ignore when the team is small, and expensive later: two answers, one customer, and no way to tell which one they believed. How a shared inbox works · the c...

A handover that lives in a chat message is lost by Tuesday. The fix is not discipline; it is putting the note where the next person will already be looking. The three-part note State: what is true right now, in one line. Next action: the single thing that must happen next, and who owns it. Traps: what was already tried and failed, so nobody repeats it. Written inside the conversation, these three lines mean the customer never has to repeat themselves, and the person taking over does not have to read forty messages. What to avoid Long narratives. A handover is an index, not a story. If it takes more than a minute to read, it will be skimmed. States, tags and views · the complete guide Free for the first 3 seats. onsitesai.com

Most support policies are written for auditors, not for customers. A one-page version that customers can actually read does more for trust than ten pages nobody opens. Seven lines Channels we answer. Name them, including the ones you do not. Hours, in the customer's time zone where you know it. First response target for each channel — only promise what you hit. What counts as urgent , in one sentence. What we cannot help with , with a pointer to who can. Escalation path when the first answer does not work. Where the record lives , so a customer can ask "what did we agree". Why one page Because the policy has to be quotable in a reply. If an agent cannot paste two lines from it into a conversation, it will not be used, and then it is decoration. Saved replies that still sound human · the complete guide Free for the first 3 seats. onsitesai.com

Every shared inbox drifts toward noise: newsletters, vendor pitches, "quick question" emails that belong to sales, and notification mail from your own tools. At three people, that noise is not annoying — it is the reason real messages get answered late. Four filters that do most of the work Route, don't delete. Vendor pitches and partnership requests belong in a separate view, not the bin. Some of them become customers. Match on the sender, not the subject. Subject lines vary; senders repeat. Kill your own notifications at the source. Most of the "system mail" is a setting in another tool. One rule per week, not a batch of twenty. Rules nobody reviews become a second inbox problem. What to measure Not "spam filtered". Measure how many conversations a person had to open before finding one that needed them. That number is the cost of noise, and it goes down when the rules are right. How a shared inbox works · the complete guide Free for t...

Export-oriented small companies end up answering in three or four languages, usually by hoping one person in the team speaks each of them. That works until that person takes a holiday. What actually needs translating The first reply. A customer writing in their own language should not receive an English template. Commitments. Dates, prices and exceptions must be confirmed in the language of the deal, or they are not commitments. Internal notes. These can stay in the team's working language — they are for the team. Structuring for it Tag conversations by language so the queue can be routed to whoever can answer, instead of by channel. Saved replies should exist per language, not as one translated set that drifts. Onsites AI keeps every channel in one shared inbox, so a customer who writes on WhatsApp in Spanish and follows up by email in English stays one thread with one history. Multi-channel inbox · saved replies · the complete guide Free for the first 3 seats. ons...

Most writing about AI support is about capability. The more useful question for a small team is the boundary: what should the assistant never do without a human? Four things to keep human Deciding that a ticket is closed. An assistant can propose it; a person should confirm, or the desk teaches customers that raising a problem gets them ignored. Promising dates, discounts or exceptions. Anything that binds the company should come from someone who can be held to it. Handling an angry customer. Tone is the whole message in a complaint; a generated reply reads as indifference. Anything involving personal data it was not given for this purpose. The safe default is to escalate. What AI should do Draft answers to the questions that repeat, summarise a long thread before handover, and tag incoming conversations so triage is faster. All three make people faster without changing what the company promises. That is the design constraint behind Onsites AI: replies live inside the con...

The usual advice is to build a knowledge base first. For a three-person team that order is usually wrong: you end up writing documentation for questions nobody has asked yet, and none for the ones that arrive daily. The order that works Run the inbox for two weeks. Let the questions accumulate. Tag the repeats. You will find three or four topics doing most of the work. Write only those. Three genuinely useful pages beat thirty speculative ones. Attach them to saved replies , so answering gets faster without sounding canned. A knowledge base is not a documentation project. It is the record of questions you have already answered. Saved replies that still sound human · setting up the desk · the complete guide Free for the first 3 seats. onsitesai.com

A to-do list asks one question: is it done? A support record has to answer a second one: what did we tell this customer last time? Three-person teams often collapse the two. The conversation is answered, archived, and effectively forgotten — so when the same customer writes again in a month, the answer starts from zero. What a queue has that a list does not State: new, waiting on us, waiting on customer, closed. Not just "done". Owner: one person accountable, even if anyone can reply. Memory: the thread stays attached to the customer, not to the inbox view. That is the whole difference between answering messages and running support. The tool matters less than whether those three properties exist. How a shared inbox works · states, tags and views · the complete guide Free for the first 3 seats. onsitesai.com

A launch produces the worst support conditions a small team will see: a spike, unfamiliar questions, and everyone who could answer them is busy shipping. Planning for it takes an hour and saves the week. Before the launch Write the five answers you will need — the questions buyers always ask in the first 48 hours. Structure, not scripts. Publish the support hours on the channel itself, plus what happens outside them. Decide the escalation path for anything that smells like a bug, and who is allowed to call it one. Tag the launch so every conversation can be reviewed afterwards as a set. During the launch One person owns the queue at any moment, even if everyone is answering. Ownership is not about doing all the work; it is about noticing what is unassigned. After Read the tagged set once. The repeat questions become documentation or product changes; the one-off bugs become a list with dates. That review is the only part of launch support that compounds. Tags, priorities...

Once a product sells through resellers, support splits in a way nobody designed: the reseller answers the customer, the vendor answers the reseller, and the end customer's history exists in two places or neither. The three questions that decide it Who is the customer of record? If the reseller owns the relationship, the vendor's desk should treat the reseller as the contact — and say so in the handover rules. What does the vendor see? Enough context to solve the problem, without taking over the relationship. What happens when the reseller changes? Conversations should belong to the workspace, not to the individual account that handled them. Why one shared queue helps here too If reseller email, end-customer chat and internal escalation all land in the same queue with tags, the split of responsibility is visible instead of implied. The tag says whether an answer goes to the reseller or to the customer — a rule that is easy to follow and easy to audit. Partners and ag...

The refund request arrives in a support conversation. Then someone opens the invoice tool, finds the order, calculates the amount, and replies. Three systems, one customer, and a real chance of getting the arithmetic wrong. Why it breaks The conversation does not know which invoice is being discussed. The invoice does not know what was promised in the conversation. Nobody can answer "how much have we refunded this customer this year" without a spreadsheet. Keeping it in one place The conversation should already be attached to the customer record, and the record to the invoice. A refund is then an adjustment on a known order, visible to whoever reads the thread next. That is the reason invoicing sits inside Onsites AI rather than beside it: the conversation, the quote and the invoice share one record, so a refund is not a research project. CRM and invoicing · free CRM setup Free for the first 3 seats. onsitesai.com Related Support metrics that actually change a dec...

Most support dashboards measure things nobody acts on. A metric earns its place only if a number moving changes what somebody does on Monday. Metrics that pass the test Unassigned, older than a day. A number you can act on this morning, by assigning someone. Waiting on customer vs waiting on us. Separating the two tells you whether the queue is a backlog or a bottleneck. Repeat questions by topic. If the same three questions keep arriving, the fix is a document or a product change, not faster replies. Time to first human reply, by channel. Different channels have different expectations; one blended average hides the channel that is failing. Metrics that usually do not CSAT with a handful of responses. Average resolution time before "resolved" is defined consistently. Total messages handled, which rewards volume rather than outcomes. The test is simple: name the decision the metric would change. If there is no decision, it is decoration. Tags, priorities and...

Most small companies answer this by buying both and complaining about the integration. The question is really about which record is the source of truth. If the help desk is the source Every conversation, quote and invoice has to be copied into the CRM. The copy is always slightly behind, so the CRM becomes a reporting tool that nobody trusts. If the CRM is the source Then the conversation has to be attached to the customer record from the first message — otherwise support is working in a system that does not know who it is talking to. The version that works for three people One record, created by the conversation. The customer exists once; conversations, deals, quotes and invoices hang off it. Nobody re-keys anything, and "what did we promise this customer" is answerable from a single place. That is why CRM and invoicing sit inside Onsites AI rather than beside it. CRM and invoicing · free CRM setup Free for the first 3 seats. onsitesai.com Related Running support a...

Three people covering two time zones cannot be online at once, so the desk has to work while nobody is watching. That is a design problem, not an effort problem. What makes it work A queue, not an inbox. Whoever starts the day opens the queue and sees what is unassigned and urgent, without reading every thread. Handover notes that live in the conversation , not in a chat message that scrolls away. Expectations set on the channel itself: publish the hours, and say what happens outside them. One agreed definition of urgent , or everything becomes urgent at 6pm. What to avoid Do not split channels between people ("you take WhatsApp, I take email"). It recreates the silos the shared inbox was meant to remove, and it breaks on the first sick day. Tags, priorities and views · the complete guide Free for the first 3 seats. onsitesai.com

It happens daily: the same question arrives by email, then WhatsApp, then the website chat. Three agents pick it up, and the customer receives three different answers. The root cause is not careless staff. It is that each channel had its own queue, so nobody could see the other two. What to do about it Merge the channels into one queue so the duplicate is visible as it arrives. Attach identity, not just messages — the same person writing on two channels should resolve to one contact. Make ownership explicit , so the second agent sees a thread is already being handled. None of this requires the customer to change behaviour. It requires the desk to stop treating channels as separate products. Onsites AI puts website chat, email, Telegram, WeChat, LINE, WhatsApp and Messenger in one shared inbox. Shared inbox basics · the complete guide Free for the first 3 seats. onsitesai.com

Most support advice is written for teams with a support manager, a knowledge base owner and a QA process. If your team is three people and one of them is the founder, that advice is a list of things you will not do. This guide is the other list: what a small team actually needs, in the order the decisions arrive, with the trade-off stated rather than hidden. Every section links to a longer piece if you want the detail. 1. Start with one channel, not seven The temptation is to connect everything on day one. The result is an inbox nobody trusts, because nobody has proved that one channel works end to end. Connect the channel your customers already use. Answer one real conversation through it. Only then add the next. Each step should end with a working desk rather than a half-configured one. Free help desk setup, step by step 2. One inbox, or the team answers twice A shared Gmail account works for one person and breaks at three. Everyone sees everything, nobody owns anything, an...

Most website chat widgets are good at starting a conversation and bad at finishing one. The visitor types, a bot replies, and then nothing happens until someone notices the tab. What makes chat work is less about the widget than the queue behind it. It becomes a conversation with an owner , not an anonymous session that expires. Anyone on the team can answer , so it does not depend on one person being at a desk. The history survives — if the visitor returns by email, the team can see what was already discussed. Onsites AI puts website chat in the same shared inbox as email, WhatsApp, Telegram, WeChat, LINE and Messenger, so the widget is a front door rather than a separate product. Website chat · adding a chat widget Free for the first 3 seats. onsitesai.com Related Five questions prospects ask before buying a support desk Comparison pages are only useful if they admit the trade-offs AI support without a per-seat price

Across small-team sales conversations, the same five questions come up, and they are rarely about features. What happens after three seats? A price, not a call. Can we keep our email address? Yes — the desk is a queue in front of the mailbox, not a replacement for the address. Where does the data live? Cloud or self-hosted, and it should be a choice. Can one person leave without losing the history? Conversations belong to the workspace, not to the user who answered. What does the AI cost, and can we turn it off? Per action from about $0.03, and yes. If a vendor cannot answer all five in one page, the evaluation will take weeks instead of an afternoon. FAQ · pricing Free for the first 3 seats. onsitesai.com Related Website chat that a customer can actually reach a human on Comparison pages are only useful if they admit the trade-offs AI support without a per-seat price

Every software comparison page claims the same thing: we are better. The reader learns nothing, because a page that admits no weakness is a sales page wearing a table. A useful comparison answers three questions. Where does the other tool genuinely win? An honest answer builds the credibility that makes the rest believable. What does the difference cost in practice? Not the sticker price — the migration, the training, the rework. Who should not choose us? Teams that need a call centre suite, or a 24/7 managed service, are not the customer for a small-team desk. Onsites AI publishes a comparison page built on that structure, including where the alternatives are the better fit. Comparisons Free for the first 3 seats. onsitesai.com Related Website chat that a customer can actually reach a human on Five questions prospects ask before buying a support desk AI support without a per-seat price

Most AI support pricing is per seat, which produces a strange incentive: the more work you hand to the assistant, the more you pay for people. A team that automates well ends up paying for the seats it no longer needs. Per-action pricing changes that. The cost tracks the work done, so the number moves with volume rather than headcount. What to check either way What counts as an action? A draft, a sent reply, a classification? The definition decides the bill. Can a human take over mid-thread? Without that, the assistant becomes a wall between you and the customer. Is the reply visible in the conversation? An AI log nobody reads is not an audit trail. Onsites AI bills AI per action from about $0.03, keeps every generated reply inside the conversation, and lets any agent step in and finish it. AI inside the inbox · what AI support costs per month Free for the first 3 seats. onsitesai.com Related An on-premise support desk does not have to be a data-centre project Free help d...

The phrase "self-hosted" makes small teams picture racks, VLANs and a procurement cycle. In practice a support desk is one of the lighter things to run: a container, a database, a backup, and a TLS certificate. The smallest viable footprint One VM, sized for attachments and history rather than traffic. A database with a tested restore, not just a nightly dump. TLS on the domain your customers already write to. A documented upgrade step you can run without reading source code. What makes self-hosting painful is not the software; it is discovering the backup was never restorable, or that nobody knows who applies updates. Onsites AI ships the same product for cloud and self-hosted use, with deployment documentation for both — so the decision is about where the data lives, not about giving up features. Self-hosted deployment · deployment docs Free for the first 3 seats. onsitesai.com Related AI support without a per-seat price Free help desk software for a small team:...

"Free" help desks differ mostly in what they take away. Some cap seats, some cap channels, some cap history. The useful question is not whether it is free, but which of the four things below survive the free tier. What a free tier may reasonably limit Automation rules and workflow builders. Advanced reporting. Premium support. What it should not take away The channels you actually use. A free plan that covers email but not chat is a trial of half a product. Seats. A cap at two seats means the plan expires the moment you hire. History. If old conversations vanish, the desk cannot answer "what did we promise last time". Export. Data you cannot take with you is not really yours. Onsites AI is free for the first three seats permanently, includes the shared inbox across channels, and publishes the pricing for what comes after. Free help desk setup · pricing Free for the first 3 seats. onsitesai.com Related AI support without a per-seat price An on-pr...

A new support desk produces a rush of numbers in week one, and almost all of them are too small to mean anything. First response time on eleven conversations is noise, not a baseline. What is worth watching in week one Are conversations reaching the inbox at all? Coverage first: a channel that silently drops messages is worse than no channel. Does every conversation have an owner? Unassigned threads are the first thing a shared inbox loses. Which three questions repeat? Those become saved replies, not metrics. What to ignore CSAT with a handful of responses. Average resolution time before the team has agreed what "resolved" means. Any dashboard nobody has opened twice. Week one is for proving the pipe works. Numbers become useful once there are enough conversations for them to be a rate rather than an anecdote. Getting started · cloud setup docs Free for the first 3 seats. onsitesai.com Related When an agency runs your support inbox, who owns the account? S...

Outsourced support is common for small companies, and the arrangement usually starts informally: a freelancer or agency gets access to the shared inbox and answers customers on your behalf. Then the contract ends, and the questions arrive together. Who owns the conversations that were answered? Can the agency export the customer records they created? What happens to the email address customers have been writing to? The arrangement that survives a change of agency The account belongs to the company; the agency is a user inside it, with a role that can be revoked. Conversations and contacts stay with the business. The agency gets what it needs to do the work, and nothing that keeps the customer relationship hostage. Onsites AI supports that structure: the workspace is the company's, and access is granted per user. Partners and agencies Free for the first 3 seats. onsitesai.com Related The first week of a new support desk: what to measure, what to ignore Self-hosting: what ...

Self-hosted software always comes with a documentation page, and that page always assumes something. Knowing what is assumed is the difference between an afternoon and a week. Usually assumed A host with a public IP or an internal DNS name, and TLS you can terminate. A database you are allowed to back up and restore. Someone who can read a compose file without guessing. A way to receive email or webhooks from outside the network. Often not assumed, but needed A restore test. A backup you have never restored is a hope. An upgrade path. Know how you will apply the next release before it is urgent. An owner. Self-hosted means someone is on call when it breaks. Onsites AI documents both deployment paths, cloud and self-hosted, so the choice is about where the data lives rather than which features you lose. Self-hosted deployment docs · self-host overview Free for the first 3 seats. onsitesai.com Related The first week of a new support desk: what to measure, what to ignore ...

A support desk holds something more sensitive than most SaaS tools: the conversations customers had when something was wrong, plus their contact details and often their order history. The security review should happen before the first message arrives, not after. Four questions cover most of it. Where does the data live? Which region, which provider, and can you choose to keep it in your own infrastructure. Who can read it? Support staff, and anyone else? Access should be by role, not by "the whole company". What happens on exit? Can you export conversations and contacts, and what is deleted when you leave. Is the AI trained on your data? A reply engine that learns from customer conversations is a different risk from one that does not. Onsites AI is offered cloud or self-hosted, so the first question has a real answer rather than a policy document. Security and data handling · FAQ Free for the first 3 seats. onsitesai.com Related The first week of a new support ...

A shared Gmail inbox works surprisingly well for one person and badly for three. The account is a mailbox, not a queue: everyone sees everything, nobody owns anything, and "who replied?" is answered by scrolling. The symptoms are predictable: Two people answer the same email with different answers. A customer replies and lands in the inbox of whoever is not working today. Nothing can be counted afterwards, because there is no ticket, only mail. What replaces it Not necessarily a big help desk. What is needed is assignment, status and history: each conversation has an owner, a state, and a record attached to the customer. Onsites AI keeps the email address you already give out, but the messages arrive in a shared inbox where they can be assigned, answered by anyone, and counted afterwards — alongside chat and the messaging channels. Email in Onsites AI · how a shared inbox works Free for the first 3 seats. onsitesai.com Related The support stack three people actual...

Software catalogues are written for teams that already have a support organisation. A three-person company needs something much smaller, and buying the bigger thing creates work rather than removing it. What earns its place One inbox covering the channels you actually receive messages on. Customer records attached to conversations, so history does not live in someone's head. A way to invoice what was agreed, without retyping it into another tool. What to skip until it hurts Workflow builders with dozens of branches nobody maintains. Separate tools for chat, email, CRM and invoicing — four subscriptions and three integrations. Anything priced per seat when your headcount is the thing you are trying to grow. Onsites AI puts the first three in one workspace and is free for the first three seats, so the stack does not grow faster than the team. What is in the product · pricing Free for the first 3 seats. onsitesai.com Related Why a shared Gmail account stops working ...

Setting up support software usually fails the same way: someone tries to configure everything before answering the first customer. The configuration becomes the project, and the project never finishes. An afternoon is enough if the order is right. Connect one channel. The one your customers actually use today. Not all seven. Import the contacts you already have. A desk with no history starts every conversation from zero. Write three saved replies. The three questions you answer every week. Not a script library. Answer one real conversation end to end. Before adding a second channel, prove the first one works. Only then add the next channel. The point is not speed. It is that every step ends with a working desk rather than a half-configured one. Free help desk setup, step by step Free for the first 3 seats. onsitesai.com Related Why a shared Gmail account stops working at three people The support stack three people actually need (and what to skip) Adding a live chat widge...

Most teams add website chat the same way: install a widget, then discover it has its own dashboard, its own login and its own unread counter. Now there is a chat inbox and an email inbox, and nobody is sure which one the customer used. The widget is not the problem. The second inbox is. What to check before installing one Does it land in the same queue as email? If not, someone has to watch two screens. Does the visitor become a contact? A chat that disappears at the end of the session is a conversation you cannot follow up. Can anyone answer, or only the person who installed it? Chat that only one person sees is a personal inbox with a company logo. Onsites AI puts website chat in the same shared inbox as email, WhatsApp, Telegram, WeChat, LINE and Messenger — the visitor becomes a contact, and the conversation keeps its history. Adding a live chat widget Free for the first 3 seats. onsitesai.com Related Why a shared Gmail account stops working at three people The support...

Start here: The small-team support desk: a complete guide — the nine decisions in order, with the trade-offs stated. This is the index. Each link goes to a guide on onsitesai.com that solves one specific problem, in the order a small team usually hits them. Getting started Free help desk setup What a free ticketing system should include Adding a live chat widget One inbox across channels Free shared inbox WhatsApp shared inbox LINE shared inbox WeChat shared inbox Running the desk Tags, priorities and views Canned responses CRM, cost and hosting Free CRM setup What AI support costs per month Cloud vs self-hosted checklist Self-hosted help desk with Docker Migrating a help desk The product Product overview · chat · email · CRM and invoicing · AI · pricing · self-host · comparisons Free for the first 3 seats. onsitesai.com Related Seven customer channels, one shared inbox — and why support breaks at the seams WhatsApp for customer support: the probl...

Reading this first? The small-team support desk: a complete guide puts every decision in order. "Free" in support software usually means one of three things: a trial with a deadline, a tier with a seat cap that ends before your team does, or a real free plan with the features you need to run support. Four questions separate them: How many seats, permanently? Not for 14 days — for as long as you stay small. Which channels are included? A free plan that covers email but not chat is a trial of half a product. What does the paid part cost per unit? Per seat is predictable; per action needs a number you can check. Is the price published? If the answer to "what will this cost" is a call, the number is not for you. Onsites AI is free for the first three seats, permanently. Beyond that, AI is billed per action from about $0.03, and the pricing is on the site — no call required. What a free ticketing system should include · published pricing Free for the firs...

Small teams do not lack good intentions. They lack a way to agree on what to do next, quickly, without a meeting. Three primitives are usually enough: Tags say what the conversation is about. Keep the list short — five or six, not thirty. Priority says how fast it needs an answer. If everything is high, nothing is. Views turn those two into a worklist: "unassigned and high", "mine and waiting on customer", "everything tagged billing this week". The point of a view is that nobody has to ask "what should I pick up?" — the queue answers it. A useful test after two weeks: if the team still scrolls the whole inbox to decide, the taxonomy is wrong, not the team. Tags, priorities and views in Onsites AI Free for the first 3 seats. onsitesai.com Related What a support desk should cost when you are three people What “free” should mean in a ticketing system Canned responses that do not read like canned responses

A saved reply is supposed to save time. Too often it saves time for the agent and costs it back in customer trust, because the customer can tell they received paragraph four of a script. The fix is not to write better scripts. It is to narrow what a saved reply is allowed to do. Three rules Save the structure, not the sentence. "Here is what happens next, and here is when you will hear from us" survives editing; a paragraph of apology does not. Leave a slot for the specific. If the reply has no place for the customer's actual situation, it will read as generic no matter how well it is written. Do not save answers to questions you have not answered. A saved reply that closes a ticket without resolving it teaches customers to distrust the whole channel. Used this way, saved replies keep the team consistent without flattening the conversation — which is also the standard an AI reply should be held to. Canned responses in Onsites AI Free for the first 3 seats. on...

The reason teams stay on a bad support tool is rarely the price. It is the history: years of conversations that nobody wants to lose, and no clear answer about what survives a move. Sort the data into three buckets before you migrate: Must keep: open conversations, customer records, anything under a contractual or legal retention obligation. Nice to have: closed tickets older than a year, internal notes nobody reads. Can go: duplicates, test tickets, the export nobody has opened since 2021. Then migrate in that order, and verify the first bucket by opening ten random conversations in the new tool and comparing them with the old one. If the history is wrong for ten, it is wrong for all of them. The cheapest migration is the one where the new tool holds conversations, customer records and invoices in one place, so you are not migrating three systems into three others. Migrating a help desk, step by step Free for the first 3 seats. onsitesai.com Related Every Onsites AI guide...

Self-hosting sounds like a deployment detail until the day you need it: the customer contract that says data stays in your infrastructure, or the policy that rules out a third-party processor. Before running the first container, four things are worth deciding: The host. A small VM is enough for a team of three; size the disk for attachments and history. Backups. A database backup you have never restored is not a backup. Test one. TLS and the domain. Support traffic includes customer records; plain HTTP is not an option. Upgrades. Know how you will apply the next version before you need to. Onsites AI is offered cloud or self-hosted with the same product behind both, so the choice is about where the data lives rather than which features you give up. Self-hosted help desk with Docker · self-hosted deployment Free for the first 3 seats, either way. onsitesai.com Related Migrating from an old help desk without losing the history that matters Every Onsites AI guide in one pla...

WeChat is the hardest channel to turn into a support process, because so much of it is personal. Customers add an individual, not a company. When that individual leaves, the relationship and the history go with them. Three things worth separating The channel and the person. A customer should belong to the company, not to whoever added them first. Answering and owning. Anyone on the team can reply; one person remains accountable for the thread. The chat and the record. If it only lives in an app, it cannot be reviewed or handed over. None of this requires changing how customers reach you. It requires the inbox behind it to be shared — which is what Onsites AI does, putting WeChat in the same queue as email, website chat, WhatsApp, Telegram, LINE and Messenger. A WeChat shared inbox, in practice Free for the first 3 seats. onsitesai.com Related AI replies you can audit Canned responses that do not read like canned responses LINE support without passing one phone around the of...

In markets where LINE is the default, support runs through it — and that usually means a single account on a single phone. A day off, a lunch break, or a busy afternoon is enough for customers to feel abandoned. Two questions decide whether LINE support scales: Can more than one person answer? If the answer is "only the account holder", the channel is a bottleneck, not a support desk. Does the conversation become a record? A chat that only exists on a handset cannot be reviewed, handed over, or counted. A shared inbox fixes both by treating LINE like any other channel: the message arrives in a queue, gets an owner, and keeps its history attached to the customer. That is how Onsites AI handles LINE — alongside email, website chat, WhatsApp, Telegram, WeChat and Messenger. Setting up a LINE shared inbox Free for the first 3 seats. onsitesai.com Related WeChat at work: answering as a team, not as one person AI replies you can audit WhatsApp for customer support: the p...

WhatsApp is where a lot of customers actually are, so small teams answer there. Then the team grows, and the same number has to be handled by three people — except it sits on one handset, logged into one person's account. The result is predictable: messages wait until that person is free, and nobody else can see what was promised. What has to change The conversation has to land in a queue the whole team can see. Assignment has to be explicit, so two people do not answer the same thread. History has to stay attached to the customer, not to the phone. That is the difference between "we have WhatsApp" and "we run support on WhatsApp". The channel is not the problem; a one-person inbox pretending to be a team inbox is. Onsites AI puts WhatsApp in the same shared inbox as email, website chat, Telegram, WeChat, LINE and Messenger, so the team answers together. Running a WhatsApp shared inbox Free for the first 3 seats. onsitesai.com Related LINE support wit...

Reading this first? The small-team support desk: a complete guide puts every decision in order. The failure mode of AI in support is not bad grammar. It is a confident answer nobody can trace: no record of what the model was told, what it used, or why it said what it said. An auditable setup has three properties: The reply is attached to the conversation , not to a separate AI log nobody reads. A human can take over mid-thread without the customer repeating themselves. You can see what it cost — per action, not as a monthly surprise. That is the design constraint behind Onsites AI: the assistant works inside the same inbox as the team, so every generated reply sits next to the conversation it belongs to, and any agent can step in and finish it. And because pricing is per action from about $0.03, the AI budget is a number you can watch rather than a tier you buy. How the AI works inside the inbox · starting from zero Free for the first 3 seats. onsitesai.com Related Canned...

Reading this first? The small-team support desk: a complete guide puts every decision in order. For most SaaS purchases the hosting model is an implementation detail. For a support desk it is not: the tool holds customer conversations, contact records and invoices. Where that data lives is a decision, not a footnote. Three questions that settle it Who can read the data? A self-hosted deployment keeps conversations and customer records inside infrastructure you control. What happens when the vendor changes terms? Self-hosting makes the exit a migration, not a negotiation. What does your own customer promise say? If you have told clients their data stays in a specific region or system, the support tool has to fit that promise. Cloud is usually the right answer for a team without infrastructure. Self-hosted is the right answer when the data question is the deciding factor — and it should be a supported option, not a special project. Onsites AI offers both, with the same produc...

The support desk knows what the customer asked for. The CRM knows who they are. The invoice knows what they owe. In most small companies these are three tools and a person moving data between them. Every manual transfer is a chance to get it wrong: the price quoted in chat, the discount agreed on a call, the line item that never reached the invoice. The chain should be one record The conversation creates the customer record, not a separate form. A quote is built from the same line items the invoice will use. When it is accepted, the order and the invoice follow — no re-keying, no reconciliation. That is why CRM and invoicing sit inside Onsites AI rather than beside it. For a small team the value is not "more features", it is removing the handoffs where information gets lost. CRM and invoicing in the same workspace · setting it up from scratch Free for the first 3 seats. onsitesai.com Related Cloud or self-hosted: the question to settle before you sign Self-hosting ...

Reading this first? The small-team support desk: a complete guide puts every decision in order. Support software pricing is usually designed for a team of forty. Seat minimums, onboarding fees, and a plan tier for features you will never open. For a three-person company, the arithmetic rarely survives the first invoice. Two numbers are worth checking before anything else: What happens at seat three. A free tier that ends at two seats is a trial. Onsites AI is free for the first three seats, permanently. How the AI is billed. Per-seat AI pricing punishes the team that actually delegates work to it. Onsites AI charges per action, from about $0.03 — so the cost tracks the work, not the headcount. There is a third number nobody publishes: the cost of the first month. Most tools hide it behind a call. Onsites publishes the math and a comparison page, because the decision is easier when you can see it. Working out the monthly cost of AI support · published pricing Free for the fi...