For a standalone bot trained on your website and documents, shortlist Agentkit and Chatbase. If your team needs live chat and AI in one inbox, look at Tidio or Crisp. If you already run support through Intercom, test Fin there; Intercom also offers Fin for supported external help desks. The best choice depends on the handoff you need and the unit you will pay for, not on a feature count.
We publish Agentkit, so treat its entry here as an affiliated recommendation. This is a desk comparison of official product and pricing pages checked on October 26, 2026. We did not run the same questions through all five products or measure their answer quality. Prices and plan limits can change; open each linked pricing page before buying.
Five website chatbot options worth comparing
| Product | Good fit when | Website and human path | Cost question to ask |
|---|---|---|---|
| Agentkit | You want a bot trained on website pages and documents | Copy-paste website widget; paid plans can create tickets in a connected support tool | How many bot messages will you use each month? |
| Chatbase | You want a standalone AI agent and may later need its helpdesk or channel integrations | Integrations and API access vary by plan | How many message credits and paid add-ons will your agent use? |
| Tidio | Your team wants live chat, tickets and Lyro AI in the same service | Website chat with a human team able to take over | How many billable human conversations and Lyro AI conversations do you need? |
| Fin with Intercom | Your support team already uses or wants Intercom's inbox and help center | Intercom Messenger and configured handoff to teammates | What will seats and each billable Fin outcome cost together? |
| Crisp with Hugo | You want a shared inbox and website chat widget, then AI on a paid workspace | Crisp widget and inbox, with Hugo AI on eligible paid plans | What is the workspace price, included Hugo credit and likely extra usage? |
These are different products, not equivalent bundles. Crisp's free plan has a website widget but no included AI credits. Tidio's free Lyro allowance is one-time, not a recurring monthly AI quota. Intercom's seat-plus-outcome plans start at $0.99 per Fin resolution, while a qualification outcome costs $9.99; its outcome definitions include more than answered support questions. Chatbase sells message credits and separate add-ons, which are not the same unit as Agentkit messages. Check current plan details before comparing totals.
Agentkit: a content-trained bot for your site
Agentkit's current plans include one chatbot and 50 monthly messages on Free. Hobby includes 2,000 monthly messages, one chatbot and one team seat at $29.99 per month on monthly billing. Website crawling, uploaded documents, Q&A pairs and text snippets are available across plans. Hobby lists API access and integrations; Standard adds scheduled website-source retraining and branding removal. The paid Escalate to Human action creates a ticket, case or conversation in a connected support tool. If you need agents working tickets in one shared inbox, compare the other products here rather than assuming the widget replaces one.
To test the actual website setup, open your chatbot's Channels → Website embed, select Chat widget, enable it and copy the complete generated code. It contains an inline bootstrap and a separate widget loader. Place both script tags before </body>, then send a question on the published page. The widget instructions cover allowed domains and checks when the bubble does not appear.
Chatbase: a standalone agent with plan-dependent extras
Chatbase's plan comparison matters. Hobby currently lists 700 message credits per month and basic integrations; API access and advanced integrations appear on higher plans. Auto-recharge, additional agents and removing Chatbase branding are separate paid options. Chatbase is worth testing if its channel mix fits you, while the purchase decision needs the exact credit and add-on bill for your workload.
Tidio: AI beside live chat and tickets
Tidio combines Lyro with its help desk. Its feature list shows website knowledge sources, a test playground and human takeover. The pricing configurator separates billable conversations for the team from Lyro AI conversations. If a person must step in regularly, test the handoff and price both types of work. Do not treat a free live-chat allowance as a recurring free AI allowance.
Fin: AI inside an Intercom support operation
Fin can answer from Intercom articles, documents and webpages, and Intercom lets teams configure handoff to a teammate. Its deployment guide calls for an installed Messenger and at least ten public Help Center articles before setting Fin live over chat. The combined Intercom pricing includes helpdesk seats plus Fin outcomes. A resolution may be confirmed or assumed when the customer leaves without asking for more help, so audit the outcome definition against your own support goals before using billed outcomes as a success metric.
Crisp: shared inbox first, Hugo on paid plans
Crisp's free plan includes a website widget and two seats, but no included Hugo AI credit. Its paid Mini plan lists $45 per workspace per month and $5 of monthly Hugo credit. Crisp says Hugo credit usage varies with context length, replies and model choice. Crisp is a sensible shortlist item if your team values its inbox and wants to add AI there. Test a realistic conversation before estimating how far the included credit goes.
Run the same trial on each shortlist option
Use one approved FAQ or help page, then write down ten real visitor questions before configuring any bot. Include two questions the source cannot answer and one request for a person. Run the same set on each trial. Record the answer, cited source if shown, whether the bot admitted uncertainty and what happened to the handoff. Test the widget on a phone and on the published page, not only in a vendor playground.
Copy this small record for each product:
Product and plan tested: Source page and last update date: Correct answers / 10 (using our approved answer key): Unsupported questions handled safely / 2: Human request reached the right destination: yes / no / untested Published mobile page opened and accepted a message: yes / no Normal month: bot messages, AI conversations or outcomes, human seats Busy month: the same units plus add-ons and overage rules
This is a test plan, not a benchmark or proof that ten questions represent all visitor needs. A passed chat answer does not prove that a lead reached your CRM or that a support issue was resolved. Check the destination record and assign someone to review failures after launch.
Use normal and busy month counts to compare the whole bill in each vendor's own units. For example, 300 visitor conversations with four bot replies each would be 1,200 replies in a simple illustration. That is not an Agentkit message count or a vendor bill: it does not tell you Chatbase credits, Tidio Lyro conversations, Fin outcomes or Hugo token use. Ask each vendor or trial report to map your observed workload to its billing rules. The pricing-model worksheet gives a fuller comparison method.
If Agentkit matches your source and website requirements, create a free chatbot and run the question set before moving it onto public pages. If your team needs an established human inbox, start the same trial with Tidio, Intercom or Crisp and make the handoff your first acceptance check.
