Run a support inbox without paying per agent

Support tools charge per seat, so involving the two people who actually know the answer doubles the bill and you end up funnelling everything through one person instead.

Tools worth considering

Chatwoot

Customer messaging platform with a shared inbox across chat, email and social channels, an AI agent, and a self-hostable open source edition.

Cloud plans billed annually: Hacker $0/agent/mo for up to 2 agents; Startups $19/agent/mo; Business $39/agent/mo; Enterprise $99/agent/mo. Captain AI credits included per plan (300, 500, 800), extra credits $20 per 1,000. Data retention runs from 30 days on Hacker to 3 years on Enterprise. A self-hosted option is offered separately

Visit Chatwoot

Freshchat

Live chat and messaging inbox with a free tier for up to ten agents, social and SMS channels on paid plans, and AI agents sold per session.

Free $0 for up to 10 agents with chat widget, shared inbox and email; Growth $19/agent/mo billed annually adding social messengers, SMS and reporting; Pro $49/agent/mo adding multilingual conversations, intelligent routing and multiple SLA policies; Enterprise $79/agent/mo. Freddy AI Agent is $49 per 100 sessions and Freddy AI Copilot $29/agent/mo

Visit Freshchat

Crisp

All-in-one customer messaging inbox with AI agent, at a flat price per workspace instead of per seat.

Free plan (2 seats); paid from $45/mo per workspace, flat

Visit Crisp

The problem

Per-seat pricing quietly shapes how you work. When each extra login costs money every month, you give logins only to the people who do support full time. Everyone else forwards screenshots over internal chat, and the person with the licence retypes the answer.

For a small business that is exactly backwards. The people who can answer the hard questions are the ones who build or deliver the thing, and they are the ones you priced out of the inbox. The result is a queue that waits on one person, and knowledge that never reaches the place where customers ask.

The cost also arrives at the wrong time. You want more hands during a busy month, which is precisely when adding them is most expensive.

What AI can actually do here

Two separate things, and it is worth keeping them apart.

The first is deflection: an AI agent that answers from your help content before a human sees the message. It works well on the questions you answer constantly, which is usually more of the volume than people expect, and it fails on anything requiring a decision or an exception. Watch what it does with edge cases before you let it answer unsupervised, because a confidently wrong answer about a refund costs more than a slow correct one.

The second is assistance: drafting a reply for a human to approve, summarising a long thread so whoever picks it up is not reading from the start, and suggesting the relevant article. This is lower risk and often the better place to start.

Underneath both, the ordinary shared inbox is doing most of the work: one queue across chat, email and social, assignment so two people do not answer the same message, and a record of what was said.

How the tools differ

The pricing model is the point here, so read it first.

Crisp charges per workspace rather than per person: a free plan with 2 seats, then paid from $45 per month for the workspace, flat. If you want six people in the inbox, this is the shape that makes it possible.

Freshchat is free for up to 10 agents with a chat widget, shared inbox and email, which is unusually generous for a small team. Growth is $19 per agent per month billed annually, adding social messengers, SMS and reporting. Its AI is sold separately: the Freddy AI Agent at $49 per 100 sessions and Copilot at $29 per agent per month.

Chatwoot is the open source option, with a self-hostable edition offered separately from the cloud plans. Cloud billed annually runs Hacker at $0 per agent for up to 2 agents, Startups $19, Business $39 and Enterprise $99, with Captain AI credits included per plan and extra credits at $20 per 1,000. Note the data retention ladder, from 30 days on Hacker to 3 years on Enterprise, because that decides whether you can look up what was said last quarter.

If you want the ticketing-heavy end of this market instead, Zendesk vs Freshdesk covers it.

When you don't need a tool

If two people share one inbox and nothing is being missed, a shared mailbox is genuinely fine. The tool earns its place when messages arrive on several channels or when you cannot tell who is handling what.

If you are considering this to reduce headcount, be careful. Deflection reduces volume, it does not remove the need for someone who can decide. Plan for the same people handling fewer, harder conversations.

And if your problem is that the same five questions arrive forever, see answer repeat customer questions first, because writing the answers down well is the cheaper fix.

Get advice for your situation

Your AI consultant

Free · no signup · advice first

An example

I waste hours every week going back and forth to schedule client calls.
You don't need a big system for this. There's an easier tool for this: Calendly. On the free plan you share one link and clients pick a slot that syncs with your calendar. Paid starts at $10/user/mo, but you likely won't need it.

Your turn

So, what can I help you with? Describe your situation in your own words, the way you would explain it to someone at the next table.Not sure where to start? Tell me what your business does and what eats your week, and we'll work out together where a tool would actually help.

Or start with a question people ask about this:

By chatting you agree to our terms.

Frequently asked questions

Is flat per-workspace pricing always cheaper?

Only above a certain team size. One or two people is often cheaper on a per-agent plan or a free tier. Work out your real number of people who should be able to reply, including the occasional ones, then compare at that number rather than at today’s minimum.

How much can an AI agent actually deflect?

It depends entirely on how repetitive your questions are and how good your help content is. Measure it rather than trusting a percentage from a vendor: look at what share of your last hundred conversations were answerable from existing documentation. That share is your realistic ceiling.

What about customer data and GDPR?

Support transcripts contain personal data and often more than you would guess, including addresses and order details. Check where the provider stores it and for how long, and note that retention on cheaper tiers can be short. Self-hosting moves the question rather than removing it.

Can I move later if I pick wrong?

Conversation history is the sticky part. Export options vary and reimporting into a different tool rarely reproduces threads cleanly. If you expect to move, keep your help content in your own system rather than only inside the support tool, because that is the part worth keeping.