Contents
- Verdict
- What is Chatbase?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with it
- What Chatbase actually costs at small-business volume
- The gotcha: one agent per plan, at every self-serve price
- What breaks in production, and how you find out
- What it connects to that you already run
- Chatbase alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
Chatbase is the fastest way in this batch to get a documentation-grounded support agent live, and its pricing is more slippery than it looks. As of August 2026 the displayed prices are annual-billed: Hobby shows $32 a month but costs $40 monthly, Standard shows $120 and costs $150, Pro shows $400 and costs $500. The unit is a message credit consumed per agent response, and one response costs between one and six credits depending on the model, so your effective allowance can vary sixfold on the same plan. Every self-serve plan includes exactly one agent; extras are $300 a year each. Removing the branding is $1,188 a year. It is genuinely good at the job it does.
What is Chatbase?
Chatbase is a platform for building AI support agents trained on your website, documents, and knowledge base, then deployed as a chat widget or through channels and APIs. You upload or crawl your content, tune the agent's instructions, and it answers customer questions grounded in that material, with actions available for tasks beyond Q&A. It is one of the most popular entry points for businesses adding an AI chatbot to their site, prized for setup speed over deep customization.
Best for
Small and mid-sized businesses that want a documentation-grounded support chatbot live on their site quickly.
Not for
Support that needs deterministic routing or several distinct agents. One agent per plan and no conditional flow builder are hard structural limits, not settings.
Strengths
- Fastest setup of the five: point it at your site, let it crawl, and you have a working agent in minutes
- Credit cost per model is published exactly, so you can halve your effective cost by choosing a cheaper model and know precisely by how much
- Auto-recharge credits do not expire, unlike subscription credits, which makes them a reasonable buffer rather than a penalty
- Standard at $120 a month billed annually includes helpdesk, voice, telephony with 10 concurrent calls, outbound campaigns and API access, which is a lot of surface for the money
- SIP trunk support added on 3 August 2026 lets you point existing phone numbers at the agent rather than buying new ones
- Full agent lifecycle management through the API since 5 August 2026: create, train, clone, configure and delete agents programmatically
- Very broad integration list for the price, including Zendesk, Salesforce, HubSpot, Freshdesk, Gorgias, Shopify, Slack and WhatsApp
- Active development: three changelog entries in the first week of August 2026 alone
Limitations
- Displayed prices are annual-billed; monthly billing is $40, $150 and $500, which is 25% more
- Every self-serve plan includes exactly one agent, including Pro at $400 a month, with extras at $300 per agent per year
- Credit cost varies from 1 to 6 per response by model, so the headline allowance can mean sixfold different volumes
- Running out of credits is a hard stop: the agent tells your customers it is unavailable unless auto-recharge is on
- Credits renew on the 1st of the month regardless of when you subscribed, so a mid-month signup gets a short first cycle
- Removing 'Powered by Chatbase' costs $1,188 a year, which is more than a full year of Hobby
- Free tier agents are deleted after 14 days of inactivity, so it is a demo rather than a free plan
- Training content is capped by plan at 1MB, 10MB, 20MB and 40MB, which binds sooner than people expect on documentation-heavy sites
- No conditional logic builder, so every decision rides on model interpretation rather than a deterministic branch
Chatbase pricing
Chatbase sells monthly message credits, consumed each time your agent responds. Credit cost per response depends on the model: 1 credit for most models, 2 for the fast frontier tier including GPT-5.2 and Gemini 3.5 Flash, 3 for Claude 4.5 and 4.6 Sonnet and GPT 5.6 Terra, 4 for Grok 4 and GPT-5.5, 5 for Claude 4.5 and 4.6 Opus, and 6 for Claude 4.7 and 4.8 Opus. Credits renew on the 1st of each month regardless of your subscription date, and do not roll over. When they are exhausted the agent stops and displays a message saying it is unavailable, unless you have bought the auto-recharge add-on at $40 per 1,000 credits, whose credits do not expire. Plan prices are displayed annual-billed by default; the monthly-billed prices are $40, $150 and $500. Every self-serve plan includes one agent, with extras at $300 per agent per year and branding removal at $1,188 a year.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | 50 message credits/month, 1 agent, 1 member, 1MB training content, 0 AI actions, limited models. Agents are deleted after 14 days of inactivity (as of Aug 2026) |
| Hobby | $32/mo billed annually | $40 billed monthly, $384/yr. 700 credits, 1 agent, 2 members, 10MB training content, 5 AI actions, advanced models, integrations, basic analytics, attachments |
| Standard | $120/mo billed annually | $150 billed monthly, $1,440/yr. 4,000 credits, 1 agent, 3 members, 20MB, 8 AI actions, helpdesk, voice, telephony with 10 concurrent calls, outbound campaigns, API access, auto retrain |
| Pro | $400/mo billed annually | $500 billed monthly, $4,800/yr. 15,000 credits, 1 agent, 5 members, 40MB, 12 AI actions, 20 concurrent calls, advanced analytics, source suggestions, tickets as a source |
| Enterprise | Let's Talk | Higher limits, flexible billing, custom roles, SSO, white-labeling, audit logs, SLAs, HIPAA-eligible, zero data retention, success manager |
| Add-ons | $40 / $300 / $1,188 | Auto-recharge $40 per 1,000 credits that never expire; extra agent $300 per agent per year; remove 'Powered by Chatbase' $1,188 per year |
Pricing reflects public plans as of August 6, 2026 and can change. Check Chatbase for the latest.
What people actually build with it
Chatbase does one thing extremely well and several things adequately. The one thing is answering questions from your own content, and almost every successful build is a variation on it.
Documentation-grounded website support
The canonical build: crawl your site, upload your policy PDFs and FAQs, tune the instructions, embed the widget. The agent answers delivery, pricing, policy and how-to questions from that corpus, and the retrieval-augmented generation behind it means the answers cite your content rather than the model's memory. For a business whose support queue is 70% repeat questions, this is the highest-return automation available and it takes an afternoon.
The constraint to design around is the training content cap: 1MB on Free, 10MB on Hobby, 20MB on Standard and 40MB on Pro. Documentation-heavy sites hit that faster than expected, and the fix is curation rather than upgrade - a smaller, cleaner corpus usually answers better than a large messy one.
Lead capture inside the conversation
The Collect Leads action gathers contact details naturally during a chat, and since 30 July 2026 it can email you the moment a lead is captured rather than waiting for the daily summary. For a cleaning company or a salon, the workflow is: visitor asks about pricing, agent answers from the price list, agent asks for a name and number to send a quote, you get an email within seconds. Speed to lead is the entire value there, which is why the notification change matters more than it sounds.
Voice and telephony on a chat platform
Standard and above include voice and telephony with 10 concurrent calls on Standard and 20 on Pro, and since 3 August 2026 you can connect existing numbers through a SIP trunk by pointing your PBX or provider at a SIP URI Chatbase generates. That is a genuinely useful path for a business with an existing phone system that wants after-hours answering without moving its numbers.
Where it falls down
Anything needing deterministic branching. Chatbase has no conditional logic builder, so you cannot guarantee that a refund question always follows the refund path; you can only instruct and hope. Anything needing several distinct agents also struggles, because every self-serve plan includes one agent and extras are $300 a year. A dental practice wanting one agent for patients and another for referring dentists is buying an add-on, not changing a setting. For either requirement, Botpress or Voiceflow is the better tool.
Bottom line: Chatbase is the right tool for documentation-grounded Q&A with lead capture, and the wrong one whenever an answer must follow a guaranteed path.
What Chatbase actually costs at small-business volume
On the reference workload used across this batch - 1,000 support conversations a month averaging six turns, so 6,000 agent responses - Chatbase's bill depends entirely on which model you choose, because credits are consumed per response at a published rate per model.
On a 1-credit model, 6,000 responses cost 6,000 credits. Standard includes 4,000, so you need 2,000 more, which at $40 per 1,000 through auto-recharge is $80. Total: $200 a month at the annual-billed price of $120, or $230 if you pay monthly. On a 2-credit model such as GPT-5.2 or Gemini 3.5 Flash, the same traffic costs 12,000 credits, which puts you on Pro at $400 billed annually or $500 monthly. On a 6-credit model such as Claude 4.8 Opus, 6,000 responses would be 36,000 credits, which is more than double Pro's entire allowance.
That sixfold spread on identical traffic is the most important thing to understand about Chatbase pricing, and it is entirely under your control.
| Configuration | Credits per response | Credits needed | Monthly cost |
|---|---|---|---|
| Chatbase, economical model | 1 | 6,000 | $200 on Standard plus $80 top-up ($230 monthly-billed) |
| Chatbase, fast frontier model | 2 | 12,000 | $400 on Pro ($500 monthly-billed) |
| Chatbase, Sonnet class | 3 | 18,000 | $400 on Pro plus $120 top-up |
| Chatbase, Opus class | 5-6 | 30,000-36,000 | Pro plus $600-$840 of top-ups |
| Botpress for comparison | n/a, per conversation | 1,000 conversations | $670-$750 billed annually |
| MindStudio for comparison | n/a, tokens at cost | ~16M tokens | ~$26 |
The add-ons that change the answer
Three add-ons move the total and none appear in the plan price. Auto-recharge is $40 per 1,000 credits, and those credits do not expire, which makes it the sanest way to handle a variable month. Extra agents are $300 per agent per year, which matters because every self-serve plan includes exactly one. And removing the 'Powered by Chatbase' branding is $1,188 a year, which is $99 a month and more than a full year of Hobby at the annual rate.
For comparison, Botpress includes whitelabel webchat on its $150-a-month Plus plan. If branding removal is non-negotiable for you, that single line item narrows the gap between the two platforms considerably.
Bottom line: The same 1,000 six-turn conversations cost $200 a month on Chatbase with an economical model and over $1,000 with an Opus-class one, because credits per response range from 1 to 6.
The gotcha: one agent per plan, at every self-serve price
Read the comparison table on Chatbase's own pricing page carefully and one row does not scale with the others. Message credits go 50, 700, 4,000, 15,000. Seats go 1, 2, 3, 5. Training content goes 1MB, 10MB, 20MB, 40MB. Agents go 1, 1, 1, 1. Pro at $400 a month billed annually includes exactly as many agents as Free.
Extra agents are an add-on at $300 per agent per year. That is not an outrageous price on its own, but it is a structural surprise, because most people assume a higher tier buys more agents and design around that assumption. A business wanting a customer-facing agent and an internal staff agent needs the add-on regardless of which plan it is on. An agency wanting one agent per client is buying an add-on per client, and should be evaluating platforms that include unlimited agents, which both Botpress and Voiceflow do on paid plans.
The second gotcha is the annual display. Chatbase's pricing page opens on the yearly toggle with a 20% discount, showing $32, $120 and $400. Its own checkout prices for monthly billing are $40, $150 and $500. Anyone quoting Chatbase at $32 a month is quoting the annual commitment, and the difference over a year on Standard is $360.
The third is the hard stop. When subscription credits run out, your agent displays a message telling customers it is unavailable. Not a queue, not a degraded model - it stops answering. Auto-recharge at $40 per 1,000 credits is the only protection, and unlike subscription credits those bought credits do not expire, so a modest standing buffer is cheap insurance against a busy week.
Bottom line: Every self-serve Chatbase plan includes exactly one agent, from Free to $400-a-month Pro, with extras at $300 per agent per year.
What breaks in production, and how you find out
Chatbase's simplicity is its main production virtue and its main production risk: there are fewer moving parts to break, and fewer controls when one does.
The agent tells your customers it is unavailable
This is the failure everyone should plan for. Credits are exhausted, and the widget on your homepage tells visitors that the agent is currently unavailable and the owner should check their account. It is a customer-facing failure caused by a billing event, which is the worst combination. Credits also renew on the 1st of the month regardless of when you subscribed, so your first cycle can be much shorter than a month.
The fix is auto-recharge with a threshold you set, plus a habit of looking at the Usage page. Because top-up credits never expire, over-buying costs you nothing but timing.
Answers drift when the source drifts
As with every retrieval-grounded agent, the failure that generates complaints is a confident answer drawn from content you changed months ago. Auto-retraining on a seven-day cycle is available on Standard and above and can be toggled through the API since August 2026, which reduces but does not eliminate the problem: a crawl re-reads what is published, so a policy that changed only in someone's inbox never reaches the agent.
The related risk is the training content cap. At 20MB on Standard, a documentation-heavy business will eventually be choosing what to leave out, and answer quality follows that choice more than it follows the model.
No deterministic path, so no guarantees
Without a conditional logic builder, every routing decision is model interpretation. There is no branch you can point at to prove that a question about cancellations always produces the cancellation policy. In practice this shows up as inconsistency: the same question phrased two ways gets two different levels of accuracy. Prompt instructions and Q&A pairs mitigate it, and a guardrail is an instruction rather than a control.
For anything where a wrong answer has a financial or regulatory consequence, keep a human in the loop or use a platform with real branching. This is the clearest structural reason to choose Botpress or Voiceflow over Chatbase, and it has nothing to do with price.
Bottom line: The Chatbase failure to plan for is a customer-facing outage caused by exhausted credits, and the structural one is that nothing in the product can guarantee an answer follows a required path.
What it connects to that you already run
For its price, Chatbase's integration list is the broadest in this batch. The helpdesk side covers Zendesk, Freshdesk, Help Scout, Gorgias, Zoho Desk, Intercom and Sunshine. The business side covers Salesforce, HubSpot, Stripe, Shopify, Calendly and WordPress. The channel side covers Slack, WhatsApp, Messenger and Instagram. Zapier and Twilio are both listed, which between them reach most of what is missing.
Most of that catalogue sits on Standard and above rather than Hobby, along with API access, which is worth knowing when you price the entry tier. The API itself became substantially more useful on 5 August 2026, when Chatbase added full agent lifecycle management: create agents, add sources or point it at a URL to crawl and train automatically, update model, instructions, temperature, domains, rate limiting and widget appearance, trigger training runs, control the seven-day auto-retrain cycle, clone agents and delete them. For anyone managing agents on behalf of clients, that turns a dashboard chore into a script.
Telephony is the newest surface. Voice and telephony are included from Standard, and the SIP trunk support added on 3 August 2026 means an existing phone system can route calls to the agent without moving numbers, which sidesteps the usual porting problem entirely.
Bottom line: Chatbase connects to more of the helpdesk and commerce stack than anything else at its price, though most of the catalogue and the API start at Standard rather than Hobby.
Chatbase alternatives: what to pick instead, and when
Chatbase's competition splits by what is limiting you. If credits are limiting, the answer is a cheaper unit. If the single agent or the absence of branching is limiting, the answer is a different class of product entirely.
| Alternative | Billing unit | Cost at the reference workload | Pick it instead when |
|---|---|---|---|
| Botpress | Conversation, AI bundled | $670-$750 billed annually | You need routing, teams, RBAC, WhatsApp or unlimited agents |
| Voiceflow | Credits, 200 to the dollar | ~$56 usage plus unpublished plan fee | Answers must follow designed, deterministic paths |
| Intercom Fin | Per resolved outcome | Outcome price plus helpdesk seats | Your support already lives in Intercom |
| MindStudio | Flat plan plus tokens at cost | ~$26 | You want raw model economics and will build the surface yourself |
| Autonoly | Managed workflows | Quoted per build | You want it built, run and reported on rather than maintained in-house |
Chatbase vs Intercom Fin
This is the comparison most support teams actually face. Fin charges per resolved outcome, so you pay when a conversation ends successfully rather than when it happens, and it lives inside a helpdesk you may already be paying for in seats. Chatbase charges for every response whether or not it helped, and lives as a widget you own.
The honest split: if Intercom is already your helpdesk, Fin is a shorter path and its pricing shifts risk away from you. If you have no helpdesk, or your helpdesk is Zendesk or Gorgias, Chatbase's flat credit model is cheaper and simpler and its integration list already covers you.
When to move up to Botpress or Voiceflow
Three triggers should make you switch rather than upgrade. You need more than one agent and the $300-a-year-per-agent add-on has stopped feeling incidental. You need a guaranteed path for a category of question, which no amount of prompting will give you. Or you need routing, teams and RBAC because humans are now part of the workflow. Botpress covers all three on Team; Voiceflow covers the second best of anyone in this batch.
Bottom line: Stay on Chatbase while the job is Q&A with one agent; move to Botpress when humans and routing enter the picture, and to Voiceflow when answers must follow a designed path.
Migration and lock-in: can you get your work out?
Chatbase has the lightest lock-in of the five, which is a direct consequence of its simplicity. There is no flow graph to rebuild, because there is no flow builder. What exists is a corpus, a set of instructions, some Q&A pairs and a widget configuration.
The corpus is your own website content, PDFs and text, all of which you already have. Chatbase states that your data is never used to train other models, that it is isolated per account, and that you can delete it at any time with full data portability. The instructions and Q&A pairs are text you can copy. As of August 2026 the API can read the complete configuration for any agent, which makes taking a full snapshot a scripted job rather than a manual one.
What does not move is anything Chatbase-specific in how the answers behave: the tuned temperature, the prompt phrasing that finally stopped it over-promising, the accumulated Q&A pairs that patch its weak spots. That is real work and it is portable as text even though it is not portable as a file. Keep it in a document outside the platform, alongside the list of things the agent must never say.
Commercially the exits are clean but timed. Annual plans are where the advertised prices live, so if you took Standard at $120 rather than $150 you bought a year. Add-ons follow the same shape: extra agents are billed per year at $300 and branding removal at $1,188 a year. And on the Free tier there is a hard trap worth naming: agents are deleted after 14 days of inactivity, so an evaluation you paused over a holiday may not be there when you return.
Bottom line: Chatbase holds nothing you cannot export, and the API can snapshot a full agent configuration, so the only real switching cost is the tuning you did to the instructions.
Chatbase FAQ
How much does Chatbase cost?
As of August 2026 the pricing page displays annual-billed prices: Hobby $32 a month, Standard $120 and Pro $400. Billed monthly those same plans are $40, $150 and $500. Free gives 50 message credits a month. Allowances are 700, 4,000 and 15,000 message credits respectively, with 2, 3 and 5 workspace seats. Add-ons sit on top: auto-recharge credits at $40 per 1,000, extra agents at $300 per agent per year, and removing the Chatbase branding at $1,188 per year.
How do Chatbase message credits work?
Each response from your agent consumes credits, and the number depends on the model. Most models cost 1 credit. The fast frontier tier including GPT-5.2 and Gemini 3.5 Flash costs 2. Claude 4.5 and 4.6 Sonnet and GPT 5.6 Terra cost 3. Grok 4 and GPT-5.5 cost 4. Claude 4.5 and 4.6 Opus cost 5, and Claude 4.7 and 4.8 Opus cost 6. Credits renew on the 1st of each month regardless of your subscription date and do not roll over.
What happens when Chatbase credits run out?
The agent stops and displays a message to your visitors saying it is currently unavailable and that the owner should check their account. That is a customer-facing failure triggered by a billing event, which is why the auto-recharge add-on is worth buying before you need it: at $40 per 1,000 credits it tops you up automatically below a threshold you set, and unlike subscription credits, auto-recharge credits never expire. You can monitor consumption on the Usage page in the dashboard.
Does Chatbase have a free plan?
Yes, with 50 message credits a month, 1 agent, 1 member, 1MB of training content, no AI actions and limited model access. The important caveat is on the pricing page itself: agents are deleted after 14 days of inactivity on the free plan. That makes it a demo for validating answer quality rather than a plan a small site could run on. The Chatbase branding also cannot be removed on any self-serve plan without the $1,188-a-year add-on.
How many agents do you get with Chatbase?
One, on every self-serve plan. Free, Hobby at $32, Standard at $120 and Pro at $400 billed annually all include exactly one AI agent; what scales between them is credits, seats, training content size and features. Additional agents are an add-on at $300 per agent per year. If you need a customer-facing agent and an internal one, or one agent per client, price that in from the start, because Botpress and Voiceflow both include unlimited agents on paid plans.
Chatbase vs Intercom Fin: which is better value?
It depends on where your support already lives. Fin charges per resolved outcome plus Intercom helpdesk seats, so you pay for success rather than for volume, which is the friendlier shape when your deflection rate is unproven. Chatbase charges per response regardless of outcome but needs no helpdesk subscription underneath it and integrates with Zendesk, Freshdesk and Gorgias instead. If Intercom is already your helpdesk, Fin is the shorter path; if not, Chatbase is usually cheaper and simpler.
Can Chatbase handle phone calls?
Yes, from the Standard plan. Voice and telephony are included with 10 concurrent calls on Standard and 20 on Pro. Since 3 August 2026 you can also connect existing phone numbers using a SIP trunk: add the number under Deploy then Phone and point your trunk, PBX or GSM gateway at the SIP URI Chatbase generates. Incoming calls are then answered by your agent using its configured voice, personality and knowledge base, which avoids porting numbers away from your current provider.
Is Chatbase worth it?
For documentation-grounded question answering with lead capture, yes, and it is the fastest route to a working agent in this batch. It is not worth it if you need several agents, deterministic routing, or a guarantee that a particular question always follows a particular path, because there is no conditional logic builder and every plan carries one agent. Be honest about which situation you are in before you subscribe annually, because the annual price is the one advertised.
Can you set up a support agent for us?
Yes, and the useful work is rarely the setup. It is deciding what the agent must never answer alone, how it escalates, who keeps the content current and how you will know when its answers start drifting. We scope that on a free automation audit, pick the platform on your real volume and requirements, and hand it over with documentation you own plus a recurring plain-language report covering what it handled, what it escalated and what needs attention.
Looking at alternatives? Chatbase genuinely beats us on time to first working agent. If your requirement is a chat widget trained on your documentation and you want it live this afternoon, nothing in this batch is faster and the entry price is fair. Two criticisms we would put to any buyer: the advertised prices are the annual ones, so budget $40, $150 or $500 rather than $32, $120 or $400 if you pay monthly, and one agent per plan at every self-serve tier is a structural limit people design around too late. Autonoly is aimed at the operational work behind the chat window rather than the window itself, and for a business whose real problem is what happens after the customer asks, that is the difference that matters. See the Autonoly review.
