All four tools put an AI assistant on your website that answers from your own content. They differ in the thing that decides your real bill: what they meter. One charges by model credits, one by pooled queries, one by AI conversations on top of a chat plan, and one folds AI into a flat workspace fee. This guide compares those billing units — not just the headline prices.
Hobby, billed annually ($40/mo monthly). No free plan on the current pricing page — 7-day trial.
Standard, billed annually ($99/mo monthly). No free plan — 7-day trial, card required.
Starter, billed annually ($29/mo monthly). The Lyro AI agent is billed separately — see below.
Mini, per workspace. Human conversations are unlimited on every plan; AI runs on included credits.
| Chatbase | CustomGPT.ai | Tidio | Crisp | |
|---|---|---|---|---|
| Best for | An AI agent trained on your content, embedded anywhere | Answering from a large document / help-center library, with citations | Shops and SMBs that want live chat and AI in one inbox | Small teams that want one flat workspace bill |
| What you are billed on | Message credits — and one reply costs 1–5 credits depending on the AI model you pick | Pooled queries per month, shared across all agents on the account | Three separate meters: human (billable) conversations, Lyro AI conversations, and Flows visitors | Flat fee per workspace, with a dollar amount of AI credit included in each plan |
| Free option | No free plan on the current pricing page — 7-day trial of paid plans | No free plan — 7-day trial, credit card required, auto-charges at the end | Yes — 50 billable conversations/mo; Lyro AI limited to 50 conversations once, not monthly | Yes — 2 seats and unlimited conversations, but no AI credits, so no AI agent on Free |
| Paid plans from | Hobby $32/mo billed annually ($40 monthly), 700 credits | Standard $89/mo billed annually ($99 monthly), 1,000 queries/mo | Starter $24.17/mo billed annually ($29 monthly) for chat — plus Lyro from $32.50/mo annual for 50 AI conversations | Mini $45/mo per workspace, incl. $5 of AI credit (about 90 automated conversations) |
| The working tier most buyers land on | Basic $120/mo annual ($150 monthly): 4,000 credits, helpdesk inbox, API | Standard for most; Premium $449/mo annual ($499 monthly) is a 5x jump with nothing in between | Growth from $49.17/mo annual plus a Lyro package sized to your AI volume | Essentials $95/mo: 10 seats, $25 of AI credit (about 450 conversations), knowledge base |
| Training limits to read | 40 MB of content per agent; 1 agent included, extra agents are an add-on | 5,000 documents per agent and 60M words stored on Standard; queries are pooled account-wide | Lyro learns from your help content and site; the binding limit is conversation volume, not data size | Knowledge-base training for the AI starts at Essentials — Mini's AI credit exists before the full training setup does |
| Human handoff | Own helpdesk inbox from Basic up; on Hobby you lean on external helpdesk integrations | An answer engine, not an inbox product — human follow-up needs a connected helpdesk or your own contact flow | Native: AI and humans share one inbox, so a person takes over in the same thread | Native: Hugo works inside the Crisp inbox your team already answers from |
| Languages | Vendor states 95+ languages for replies | Vendor states 90+ languages, Japanese explicitly included | Multilingual widget and AI; the published language coverage for Lyro is narrower than the other three — verify Japanese coverage for your use case | Multilingual inbox and AI replies; admin experience is primarily in English and other major languages |
| Japanese-language note | All four can reply to visitors in Japanese through their underlying language models (CustomGPT.ai lists Japanese explicitly). None of the four should be assumed to offer a Japanese-language admin console or Japanese support desk — plan on operating the tool in English, and test Japanese replies on your own content during the trial. | |||
| Trade-off to know | The same question can cost 1 credit or 5 depending on the model — your bill moves with a settings toggle | Removing the "Powered by" branding and auto-syncing your site both wait for the $499 Premium tier | The $29 headline is only the chat meter. AI and automation are separate sliders on the same bill | The per-conversation rate for AI credit beyond the included amount is not published on the pricing page |
Prices are USD, before tax, as published by the vendors in November 2026 (Prices checked November 2026). Chatbase plan names changed recently (Standard → Basic, Pro → Business); older articles may still use the old names. Plans change — always confirm the current price, limits and what's included on the vendor's site before you buy.
Headline prices can't be compared directly because the meters differ. So here is the same workload priced four ways, using only published prices and published billing rules. Assumptions, stated plainly: 300 conversations a month, averaging 4 AI replies per conversation (1,200 AI replies). This is arithmetic on public pricing, not a measurement — your reply count per conversation will differ, and that difference is exactly what moves the credit-based bills.
| Tool | How 300 conversations consume the plan | Estimated monthly cost (annual billing) |
|---|---|---|
| Chatbase | 1,200 replies × 1 credit (standard model) = 1,200 credits — over Hobby's 700, inside Basic's 4,000. At 3 credits per reply (higher-end model) = 3,600 credits, still inside Basic. At 5 credits = 6,000, over Basic. | About $120 (Basic) on a standard model; up to about $400 (Business) if you run a top-tier model for every reply |
| CustomGPT.ai | Each visitor question is roughly one query: 300 × 4 = about 1,200 queries against Standard's pooled 1,000. Over the cap, agents stop answering; extra-query add-ons are priced in large blocks. | $89 (Standard) only if your real volume stays under 1,000 queries — at a steady 1,200, budget for the next step rather than the entry price |
| Tidio | Lyro is billed per AI conversation, at roughly $0.70–0.78 per conversation on published self-serve packages: 300 × ~$0.70–0.78 ≈ $210–234 for the AI meter alone, on top of the chat plan. | Roughly $235–260 total with the Starter chat plan — about ten times the $29 headline once AI does the work |
| Crisp | Essentials includes $25 of AI credit, which Crisp itself converts to about 450 automated conversations. 300 conversations fit inside the included amount; human conversations are unlimited either way. | $95 flat (Essentials) — the only one of the four where the sticker price is the whole bill at this volume, as long as you stay within the included AI credit |
Estimate basis: Chatbase credit-per-reply rates (1–5 by model) and plan credit pools; CustomGPT.ai's published definition of a query and its 1,000-query Standard pool; Tidio's published Lyro package pricing (from $39/mo for 50 conversations, monthly billing); Crisp's published conversion of included AI credit to approximate conversation counts. All figures checked November 2026. Treat this as a budgeting model, not a quote.
Chatbase builds an AI agent from your website, files and text, then lets you embed it on your site or connect it to channels such as WhatsApp and Slack. Hobby ($32/mo billed annually, $40/mo monthly) includes 700 message credits, 2 seats and one agent with 40 MB of training content. Basic ($120/mo annual, $150/mo monthly) raises that to 4,000 credits and 3 seats and adds the helpdesk inbox, voice and API access. Business ($400/mo annual, $500/mo monthly) carries 15,000 credits and 5 seats. The detail that decides your bill: a single reply costs between 1 and 5 credits depending on which AI model the agent runs, so the same traffic can drain the pool at very different speeds. Removing Chatbase branding is a paid add-on ($99/mo on published pricing), and extra credits auto-recharge at a published per-1,000 rate.
CustomGPT.ai is a retrieval-first agent platform: you index your documents, help center and site, and it answers from that library with source citations attached. Standard ($89/mo billed annually, $99/mo monthly) covers 10 agents, 5,000 documents per agent, 60 million words of storage and 3 team members — the most generous training limits in this comparison. The catch sits in the usage pool: 1,000 queries a month, shared across every agent on the account, and when the pool is empty the agents stop responding. Premium ($449/mo annual, $499/mo monthly) lifts the pool to 5,000 queries and 20,000 documents per agent, and is also where branding removal and automatic website syncing live. There is nothing priced between $99 and $499. The vendor publishes anti-hallucination claims for its retrieval approach; those are the vendor's claims, presented here as positioning, not as something this guide has verified.
Tidio is a customer-service panel — live chat widget, shared inbox, ticketing — with an AI agent called Lyro that learns from your help content and store data and answers inside the same inbox your team works in. That shared inbox is its handoff story: when Lyro can't finish a conversation, a human continues the same thread. The pricing needs care. The customer-service plan runs from a free tier (50 billable, human-handled conversations a month) through Starter ($24.17/mo billed annually, $29/mo monthly) to Growth (from $49.17/mo annual), which scales with conversation volume. Lyro is a second subscription on top of that — from $39/mo for 50 AI conversations on monthly billing — and automation Flows are a third meter, priced by visitors reached. On the free and entry plans, Lyro's included allowance is a one-time 50 conversations, not a monthly budget.
Crisp prices the workspace, not the conversation: every plan includes unlimited human conversations, and its AI agent, Hugo, runs on a dollar amount of AI credit bundled into the plan. Free ($0) gives 2 seats and the chat/inbox basics but no AI credit. Mini ($45/mo per workspace) includes 4 seats and $5 of AI credit — about 90 automated conversations by Crisp's own conversion. Essentials ($95/mo) is where the product becomes an AI setup: 10 seats, $25 of credit (about 450 conversations), the knowledge base Hugo trains on, the workflow builder and the full omnichannel inbox (website, email, WhatsApp, Instagram, LINE and more). Plus ($295/mo) scales to 20+ seats and about 1,350 automated conversations. Extra seats are a flat $10. One gap to know: the pricing page does not publish what AI credit costs once you pass the included amount, so very high AI volume reintroduces an unknown the flat fee seemed to remove.
If a person must pick up the thread, Tidio and Crisp keep AI and humans in one inbox by design, and Chatbase does from its Basic plan. CustomGPT.ai answers from your library but leaves the human side to a helpdesk you connect yourself. Choose the handoff model first — it decides what else you have to buy.
Estimate last month's inquiries and replies per conversation, then run the worked example above with your numbers. Per-conversation pricing (Tidio) rewards short chats; pooled queries (CustomGPT.ai) punish long ones; model credits (Chatbase) move with a settings choice; flat fees (Crisp) don't move at all until the included AI credit runs out.
A single FAQ page fits anywhere. A real help center or document library points to CustomGPT.ai's document limits or Chatbase's 40 MB per agent — and check which plan actually unlocks knowledge training before you pay for AI credit you can't feed (Crisp's knowledge base starts at Essentials).
All four offer a trial or free tier. Load your actual help content, ask your twenty most common — and five most awkward — customer questions, including in Japanese if your customers write in Japanese, and watch both the answers and the usage meter. The meter's behaviour during the trial is the most honest pricing page there is.
For the specific job of answering website inquiries from your own help content, yes — that is the comparison this page makes. Intercom combines per-seat pricing with per-outcome AI pricing, which suits larger support organisations with a full helpdesk operation. The four tools here are narrower and, for a small team, easier to budget. What they don't replicate is Intercom's depth of ticketing, CRM and outbound features — if you rely on those, you're not shopping in this category anymore.
It depends on volume, which is the point of the worked estimate above. At very low volume, Tidio's free chat plan or Crisp Mini ($45) is the cheapest way in. At around 300 AI-handled conversations a month, Crisp Essentials ($95) is the only flat total on published pricing, while Tidio's AI meter alone runs past $200 and credit-based tools depend on your model and reply counts. "Chatbot pricing" searches usually surface the $29–45 headlines; the meter is where the money is.
Per the vendors' published language support, yes for replies: CustomGPT.ai lists 90+ languages including Japanese, Chatbase states 95+ languages, and Tidio and Crisp provide multilingual AI and widgets, with Tidio's published coverage for its AI agent narrower than the others. What you should not assume is a Japanese admin console or Japanese-language vendor support — expect to configure and operate these tools in English, and verify Japanese answer quality on your own content during the trial.
That depends on the handoff design, and it's the biggest practical difference here. In Tidio and Crisp, the conversation sits in a shared inbox, so a human agent continues the same thread. Chatbase offers its own helpdesk inbox from the Basic plan up. CustomGPT.ai is an answer engine rather than an inbox, so the fallback — a connected helpdesk, a form, an email route — is something you set up around it. None of these behaviours were tested by this guide; they are described from the vendors' published product documentation.
Because this guide is compiled from the vendors' published specifications and pricing, not from hands-on testing. Vendors publish their own resolution-rate and accuracy claims, and those claims use each vendor's own definitions, so they can't be lined up honestly. What can be compared without testing is what this page compares: billing units, published limits, training caps and handoff design — with the arithmetic shown.