The Best Free AI Chatbot Options, With The Actual Limits

"Free" means four different things, and buyers get burned by the gap

Almost every complaint about free chatbots traces back to one of these four models being marketed as another. Read the pricing page for which one you are actually getting before you build anything.

Permanent free tier

A permanent free tier is a plan priced at $0 that never expires, with hard caps on usage that stop the product rather than start a bill.

This is what most people mean by a free chatbot. The account stays open indefinitely, and when you hit the cap the bot either stops responding or you are asked to upgrade. Nothing is charged automatically. Crisp, tawk.to, Tidio, Chatbase, Landbot, Typebot, Chatwoot, Zoho SalesIQ, HubSpot and Manychat all run this model.

The catch is that permanent does not mean stable. Manychat changed its pricing model on 2 March 2026 and its free plan now covers 25 active contacts a month. People who built on the older, far more generous free plan woke up on a different product. Free tiers are a marketing budget, and marketing budgets get cut.

Examples: tawk.to, Crisp Free, Tidio Free, Chatbase Free, Chatwoot Hacker

Time-limited free trial

A free trial gives you the paid product for a fixed number of days, after which you either pay or lose access.

A trial is not a free tier and should never appear on a list of free chatbots without a label. The useful question is whether a card is required up front, because a trial that takes a card converts to a charge by default and a trial that does not cannot.

Trials are the honest way to evaluate a paid product, since you get the real thing at full capability instead of a crippled sample. They are a terrible way to run a chatbot you intend to keep. This is the model matram.ai uses, which is exactly why we are not in the table below.

Examples: matram.ai (7 days, no card), Zoho SalesIQ paid tiers (15 days)

Free with vendor branding

A free-with-branding plan is fully functional but displays the vendor's name on your widget, and you pay to remove it.

tawk.to is the clearest example. The core product is free at any volume, with unlimited agents and unlimited chat history, and the company states plainly that removing the "powered by tawk.to" badge is one of the paid add-ons. That is a fair trade, openly described.

Assume branding is present on any free plan unless the pricing page says otherwise. On a hobby project it does not matter. On a B2B site where the chat widget sits next to your enterprise pricing, it is the thing that pushes people to upgrade, and vendors know that.

Examples: tawk.to, Most free tiers of hosted widget products

Free and open source, self-hosted

Open-source chatbot software costs nothing to license, and you pay instead in servers, model API bills and your own engineering time.

Chatwoot and Typebot both publish their code and both also sell hosted plans. Self-hosting removes the vendor's caps entirely. It does not remove cost. You are now paying for a server, a database, backups, TLS certificates, upgrades, and if the bot generates answers, a model provider bill that scales with every conversation.

This is the right choice when you have an engineer who wants the job and a compliance reason to keep data on your own infrastructure. It is the wrong choice when you picked it to save $29 a month, because the first outage will cost more than a year of the subscription.

Examples: Chatwoot Community Edition, Typebot

Ten tools with a genuine permanent free plan

Ordered roughly by how much you can do before the cap bites. Every figure comes from the vendor's own pricing page as it read on 20 July 2026, and every one of them is cited at the bottom of this page. "AI answers" means the plan generates replies from your content with a language model, rather than following a flow somebody drew.

Two tools that often appear on lists like this are missing on purpose. Voiceflow no longer publishes plan names or rates on its pricing page, only a trial signup and a demo request, so there is no free-tier limit to quote. Botpress blocks automated reads of its pricing page and changed its model in May 2026, so no Botpress figure is published here rather than repeating a number from a third-party blog. Both are real products. Check their pricing yourself before you assume a free plan.

Treat every row above as a snapshot. Manychat's free plan is the cautionary example: it was far larger before March 2026. If the free plan is load-bearing for your business, screenshot the pricing page on the day you sign up.

When a chatbot is not the answer at all, free or paid

Free makes it easy to skip this question, because nothing appears to be at stake on a plan that costs nothing. Something is at stake. It is your afternoon first, and then your customers' willingness to try a widget that turned out not to help them.

Answering them yourself is genuinely fine

A handful of questions a week, arriving by email, answered by the person who knows the answer. That is not an automation gap. It is the right amount of effort, and it is more accurate than any retrieval system, because you can see who is asking and why they are asking it. Signing up for a free widget at this stage mostly gives you something to configure on a Sunday. It also costs you the one thing you would need later, which is a written record of how real people phrase a question when nobody has prompted them.

Where the friction starts

The first honest signal is that you're pasting your own replies. Same question, third time this month, and the answer already lives in a note on your desktop. That is a documentation problem before it is a software problem. Write the page. A bot pointed at content you have not written has nothing to retrieve, and the two things it can do instead, refuse or improvise, both look worse on your own domain than they did in somebody's demo.

Where it becomes a liability

It becomes a liability when the delay between question and answer starts costing you something you can name. A buyer leaves the pricing page on a Saturday because nobody could confirm one detail while they were still deciding. A customer acts on a policy that somebody half-remembered, and the argument is now about a sentence rather than about money. That is the stage where the caps in the table above stop reading as generous and start reading as the actual constraint on your week.

The case that breaks it

And there is one pattern that no plan on this page fixes, free or paid. If most of what arrives is about state rather than about facts, where is my order or what is on my invoice, then indexing your content answers none of it. Retrieval finds what somebody wrote down. Your order table is not a page. That needs a live lookup into the system holding the record, or a person with access to it, and no amount of crawling gets you there.

None of that is an argument against free plans. It is an argument for signing up on the day you have something worth serving, rather than in the hope that the content shows up afterwards.

What free plans restrict, in the order it will hurt

Free tiers are designed by people who know exactly which limit makes you upgrade. Knowing the pattern lets you predict the ceiling before you build.

1. AI conversations, capped far below chat volume

This is the sharpest limit in the category, because generated answers cost the vendor real money on every reply. Chatbase gives 50 message credits a month. Tidio gives 50 Lyro conversations once, not monthly, and its ordinary free live chat is capped at 50 billable conversations a month on top of that. Crisp gives zero AI credits on Free. A site with meaningful traffic burns any of these inside a week.

2. Conversations or contacts per month

Even where AI is not involved, the volume cap is the meter. Chatwoot's free plan stops at 500 conversations, Zoho SalesIQ at 100 operator chat sessions, Landbot at 100 chats, Typebot at 200. Manychat counts active contacts instead, which is a fairer unit and a much smaller number. tawk.to is the outlier with no cap, and it funds that by selling branding removal and services.

3. Seats

Crisp and Chatwoot both stop at two seats, Chatbase and Landbot and Typebot at one. Tidio is unusually generous here at ten. Seat limits are what kill a free plan for a growing support team, because the third hire triggers a bill that has nothing to do with how many customers you served.

4. Branding on the widget

Expect the vendor's name under the chat window on any free plan. Removing it is one of the most common first paid upgrades. If your buyers are businesses, decide early whether a competitor's logo on your support widget is acceptable.

5. Data sources and integrations

An AI chatbot is only as good as what it read. Free plans typically allow one bot and a small number of pages or files, which is enough to demonstrate the idea and not enough to answer a real customer. Integrations follow the same pattern: the free plan gets the website widget, and the channels your customers actually message you on sit behind a paid tier.

6. Retention, analytics and support

Chatwoot's free plan retains 30 days of data. Analytics are usually described as "basic", meaning conversation counts without the breakdown that tells you which questions the bot failed. Support on a free plan is documentation and a community forum. None of this matters for a test. All of it matters the month you start relying on the thing.

"Best free chatbot for WhatsApp" is the hardest one to answer honestly

Because the software can be free and the messages still cannot. Two separate bills exist, and only one of them is the chatbot vendor's.

Meta rebuilt WhatsApp Business Platform billing on 1 July 2025. The old conversation-based pricing is deprecated. Today you are charged per delivered template message, at a rate that depends on the recipient's country and on whether the template is marketing, utility or authentication. That charge goes to Meta and no chatbot vendor can waive it.

What is free under Meta's current rules is worth knowing precisely, because it shapes what a free WhatsApp bot can sensibly do. Non-template messages sent inside an open customer service window are free. Utility templates sent inside that window are free. And when a person starts the conversation through a Click to WhatsApp ad or a Facebook page button, a free entry point window opens for 72 hours during which any message type costs nothing.

Read together, that means a reactive WhatsApp bot, one that only replies to people who messaged you first and does so inside the service window, can run with very little Meta spend. A proactive one that sends marketing templates cannot be free at any volume, whatever the chatbot platform charges.

On the software side, the honest position is that free WhatsApp options are thin. Landbot's free Sandbox plan explicitly excludes WhatsApp; you need the WhatsApp Pro tier. Manychat supports WhatsApp on paid plans, and its free plan now allows 25 active contacts a month, which is a demo rather than a deployment. If you want a real number for what a WhatsApp deployment costs once the free part ends, the WhatsApp chatbot pricing breakdown covers both bills. matram.ai does not support WhatsApp today. It is on our roadmap and it is not shipped, so it is not an answer to this question either.

Free AI assistants are not free website chatbots

A large share of "best free AI chatbot" searches are actually looking for ChatGPT, Claude or Gemini. Those are a different product category, and the confusion causes real wasted effort.

ChatGPT, Claude and Gemini all have permanent free tiers with rate limits, and all three are excellent at what they do, which is answer your questions in their own app. None of them is a website chatbot. They do not sit on your site, they have not read your pricing page, they do not capture a lead, they do not hand a conversation to your support team, and they do not know your refund policy.

The thing people try next is pasting an API key into something and calling it a chatbot. That works, and it is a real project rather than a free tool. You now own retrieval over your content, prompt design, a widget, escalation, logging and the model bill. If that sounds like the right trade, it might be. If you thought you were getting a free chatbot, it is not. The chatbot vs ChatGPT comparison goes through the difference properly.

The other thing to keep straight is what sits above free in the AI support market, because it sets your expectations for what these tools cost when they work. Fin, the support agent formerly sold under the Intercom name, charges $0.99 per resolved outcome with a 50-outcome monthly minimum. Salesforce Agentforce charges $2 per conversation, or $0.10 per agent action through Flex Credits sold at $500 per 100,000. Neither has a free tier. That is the price band a free plan is sampling from.

Why matram.ai is not on this list

Because we do not have a free plan, and adding ourselves to a list of free chatbots would be the exact thing this page exists to warn you about.

matram.ai has a 7-day free trial that does not ask for a card. That is a trial, not a free tier. On day eight it stops. There is no free forever plan, no perpetual hobby tier, no thousand-messages-a-month starter that costs nothing, and we would rather say that here than have you find out after you have crawled your site and connected Slack.

The reason is not principle, it is arithmetic. Every answer matram.ai gives is a fresh model call against your content, and cited answers with strict mode cost more per reply than a keyword match against a script. A free tier large enough to be useful would cost us more than it earns, and a free tier small enough to be safe would be a demo pretending to be a product. The tools above that do give away AI answers cap them at 50 for exactly this reason.

What we sell instead is flat: Basic at $29 a month for one chatbot, 2,000 messages and 3 data sources, Standard at $69 for five chatbots, 5,000 messages, 25 sources, every channel and the public API, and Pro at $199 for twenty chatbots, 20,000 messages and 100 sources. Above that there is an Enterprise plan for higher volume, quoted rather than published and billed by invoice. Team seats are unlimited on every plan, so the third support hire does not change the bill.

Where matram.ai is a poor fit, and this belongs on an honest page as much as the pricing does: we are a newer platform with a smaller integration library than Intercom or Zendesk, we hold no SOC 2 or ISO 27001 certification and cannot sign a BAA, we are text only with no voice, and there is no free plan to fall back to if the trial does not convince you. If any of those is disqualifying, one of the tools in the table above is a better answer than we are.

When a free plan is genuinely enough, and when it stops being

Free is the correct answer more often than vendors admit. The test is whether the cap sits above your actual volume, not whether the feature list looks complete.

If you are close to that line, price both sides before deciding. The chatbot ROI calculator compares the cost of a paid plan against the support hours you currently spend, and the best AI chatbot for business roundup covers what the paid tier of each of these tools actually buys you.

What a free plan costs you that never appears on an invoice

Zero is a real price and it is not the whole price. The expensive part of a free chatbot is everything you build around it before you find out where the wall is.

The first cost is the workflow. You wire the widget into your day without ever deciding to: the tab that stays open, the way your one support person triages, the little automation that drops a name into a spreadsheet. None of that was a choice, it just accumulated. Then the cap bites, or the plan changes underneath you the way Manychat's did, and you are not swapping a tool. You're unpicking a habit that three people now depend on, in a week when you also have customers.

The second is what you quietly learn to stop asking. A free plan with basic analytics tells you how many conversations happened and not which ones failed, so you stop looking for failures, because the report has no column for them. After a few months you genuinely believe the bot is doing well. It might be. You have no way to know, and the questions it has been declining every day were the list of pages you should have written.

The third is the rebuild nobody budgets for. Traffic split across two accounts to stay under a monthly cap. A login shared because the next seat costs money. Each of those is twenty minutes, none of them appears on an invoice, and all of them are paid by the person you can least afford to have doing admin. So before you take a row from the table above, ask a different question. If this plan changed its limits next quarter, what exactly would you have to unbuild, and how long would that take you?

Frequently asked questions

Sources

• tawk.to pricing page - tawk.to is free with unlimited agents and chat history; removing the "powered by tawk.to" badge and AI Assist are paid add-ons.
• Crisp pricing page - Crisp Free: 2 seats, unlimited conversations, 100 customer profiles, zero AI credits. Paid tiers Mini $45, Essentials $95, Plus $295 per month.
• Tidio pricing page - Tidio Free: 50 billable conversations a month, 10 seats, 100 flow visitors, and 50 Lyro AI Agent conversations described as a one-off allowance.
• Chatbase pricing page - Chatbase Free: 50 message credits a month, 1 agent, 1 workspace seat, limited model access, and deletion of agents after 14 days of inactivity.
• Chatwoot pricing page - Chatwoot Hacker plan is free for up to 2 agents, limited to 500 conversations a month with 30 days of data retention.
• Landbot pricing page - Landbot Sandbox is free forever with 100 chats a month and 1 seat, and excludes WhatsApp, which requires the WhatsApp Pro tier.
• Typebot pricing page - Typebot Personal is free with 200 chats a month, unlimited typebots and a single user.
• Zoho SalesIQ pricing page - Zoho SalesIQ Free: 3 operator licences, 100 operator chat sessions a month, 10,000 visitors tracked, 1 brand.
• HubSpot chatbot builder product page - HubSpot's chatbot builder is "available for free with no credit card required", with additional functionality in the premium Hub editions.
• Manychat Help Center, Free plan - The Manychat Free plan covers up to 25 Active Contacts a month and up to 4 live automations.
• Manychat blog, "Manychat Pricing Is Changing" - Manychat moved to the Active Contacts pricing model on 2 March 2026, which is the example used for free tiers changing under existing users.
• Meta for Developers, WhatsApp Business Platform pricing - Per-message billing since 1 July 2025, conversation-based pricing deprecated, non-template messages free inside the customer service window, and a 72-hour free entry point window.
• Fin pricing page - Fin, the AI support agent formerly sold under the Intercom name, charges $0.99 per resolved outcome with a 50-outcome monthly minimum and has no free tier.
• Salesforce Agentforce pricing - Agentforce is $2 per conversation, or $0.10 per agent action via Flex Credits sold at $500 per 100,000 credits, with no free tier.
• Voiceflow pricing page - Voiceflow no longer publishes plan names or rates, showing only a free trial signup and a demo request, which is why no Voiceflow free-tier limit is quoted here.