Botpress vs Voiceflow: which should you build on in 2026?

Prices and product details verified against Botpress and Voiceflow’s official documentation and live pricing data, 1 August 2026. Cost figures are estimates, not quotes.

Contents

• The verdict
• Botpress vs Voiceflow at a glance
• Why most comparisons are out of date
• Which should you choose?
• What each platform actually costs: three worked scenarios
• Pricing in detail
• Channels and integrations
• Building experience and collaboration
• Knowledge bases and agent behaviour
• Analytics and testing
• Data protection and hosting: a UK procurement checklist
• Alternatives worth a look
• How to run a trial that tells you something
• Getting it built properly

The verdict

Build on Botpress if your customers message you on WhatsApp, if a developer-led team wants deep integration work, or if you would rather buy conversations at a known rate than meter every token: it now bills by the conversation, with AI usage bundled in. Build on Voiceflow if the agent answers the phone, since voice on Botpress is an Enterprise-only channel, or if a mixed team of designers and developers will shape the conversation together.

The two are priced on quite different terms. Botpress charges a flat fee per bracket of conversations with the AI inside it. Voiceflow sells monthly credit bundles, from $60 a month on its Pro tier, and every message, call minute and model token draws the bundle down. At low text volumes Voiceflow usually comes out cheaper; at the point where predictability, WhatsApp or a bundled helpdesk matter more than the invoice, Botpress starts earning its premium. Neither wins everywhere, and the worked scenarios below show how far apart they can land.

Be careful with anything else you read on this question. Botpress rebuilt its billing in May 2026; Voiceflow moved to credits back in April 2025 and has since stopped promoting its prices. Most articles still ranking for this comparison quote figures that predate one change or the other, describe a metered Botpress AI charge that has been abolished, and recommend a self-hosted Botpress that is no longer sold.

Botpress vs Voiceflow at a glance

Verified 1 August 2026.

Why most comparisons are out of date

Four shifts matter more than any feature list.

Botpress now bills by the conversation, and bundles the AI. In its May 2026 pricing update it dropped the subscription-plus-AI-Spend model: no separate metered AI line, no per-token arithmetic, no $5 monthly credit. LLM inference, embeddings and web search sit inside the conversation allowance. A conversation is any exchange with at least two messages in the billing month, and the definition is broader than you might assume: each anonymous session counts separately, and a conversation straddling month end counts in both months. Two footnotes are worth keeping in view. Botpress’s own Academy lesson still describes the old AI Spend model, so treat the pricing page as the purchasing source of truth; and the announcement notes that existing workspaces keep their previous pricing, so a workspace opened before May 2026 may still be billed the old way.

Botpress has repositioned around customer support. It now sells an AI-native helpdesk, Botpress Desk, alongside the Studio, on the same plans and conversation counts.

Voiceflow stopped promoting its prices. Its public pricing page now offers agencies a trial and businesses a demo button rather than a price list, and its billing documentation points buyers at the dashboard. The prices have not vanished, though, and we set them out below. The billing model itself has not changed since credits took effect in April 2025.

Botpress v12 has been withdrawn from sale. Botpress says the self-hosted version older guides still recommend “has been officially sunset” and is no longer sold or supported for new deployments, with Botpress Cloud the only path for new commercial work. The open-source v12 code and releases remain publicly available on GitHub, so the determined can still download it; running an unsupported stack to save money is a different decision from the one this article covers, and not one we would make for a client.

Which should you choose?

Choose Botpress if customers reach you on WhatsApp; if the work is support-ticket deflection and a bundled helpdesk appeals; if the agent needs to read from and write to several systems; or if you want a bill that moves in known steps with the AI already inside it.

Choose Voiceflow if the agent answers the phone, which Botpress reserves for Enterprise; if your volumes make paying per conversation painful; or if the builders are conversation designers rather than developers.

Choose neither yet if you cannot name the specific enquiries you want deflected and say what share of your inbox they represent. The costly mistake is not picking the wrong platform. It is building a capable agent for a problem nobody had.

Want a cost check before you commit? Send us your expected channels, monthly traffic and team size and we will estimate the likely Botpress and Voiceflow costs, and show you which assumptions could move the number.

What each platform actually costs: three worked scenarios

Estimates prepared 1 August 2026 from each vendor’s published rates, in US dollars, excluding VAT, implementation labour and external channel or telephony charges. One caution applies to every Botpress figure below: the AI bundled into each conversation allowance is proportional rather than unlimited, so unusually heavy model use can trigger a pack purchase before the conversation count runs out. Treat the Botpress totals as floors under ordinary usage, not ceilings.

1. A website chatbot handling 1,000 conversations a month

Assumptions: text only, an average of six visitor messages per conversation (roughly 6,000 inbound messages), knowledge-base answers, one builder.

Botpress. Plus includes 250 conversations; the remaining 750 need eight packs of 100. Billed annually that is $150 plus $520, so $670 a month; month-to-month it is $189 plus $520, so $709. There is no separate model bill to estimate. Two cautions: auto-recharge cannot be switched off, so a traffic spike buys packs without asking, and because every anonymous session counts separately, your billed conversation count may run higher than your analytics suggest.

Voiceflow. Six thousand visitor messages consume roughly 6,000 credits, and model usage adds between 0.0125 and 3.75 credits per thousand tokens depending on which model you point it at. On anything economical the whole month sits comfortably inside the entry Pro bundle, so the likely total is $60 a month, or $648 billed annually. A premium model with long retrieved context can push past the bundle, at which point you move up a tier or let top-ups take over.

At this volume Voiceflow costs less than a tenth of Botpress. What the Botpress money buys is predictability and a bundled helpdesk, not cheaper inference, and for some support teams that trade is exactly right. For a simple website agent it rarely is.

2. A three-person team running a WhatsApp support agent

Assumptions: WhatsApp as the main channel, three people building and reviewing, around 500 conversations a month.

Botpress. WhatsApp is one of its official prebuilt integrations, and Plus carries exactly three seats. At 500 conversations you need Plus and three packs: $345 a month billed annually, or $384 month-to-month, AI included. Move to Team ($750 annually, $939 monthly, 1,500 conversations, packs at $50) when you outgrow three seats or need routing, role-based access, team analytics and real-time collaboration. Meta’s own WhatsApp Business conversation charges are billed by Meta and sit outside every figure here.

Voiceflow. There is no first-party WhatsApp channel, so you are paying for development against the API before any subscription, plus ongoing maintenance, plus billable editor seats. For this brief the build cost dominates the platform cost, and Botpress is the natural fit. An estimate for Voiceflow here would be mostly guesswork about developer time, so we will not pretend otherwise.

3. A voice receptionist handling 100 five-minute calls a month

Assumptions: 500 call minutes, roughly 2,000 to 3,000 characters of synthesised speech per call, short caller turns, one builder.

Voiceflow. Voice is built in and everything draws from the same credit bundle. Call time costs 10 credits a minute, so 5,000 credits. Transcription adds roughly 280 to 2,000 credits and speech synthesis anywhere from about 30 to 10,000 depending on the voice, with a little model usage on top. A typical configuration lands between 6,000 and 11,000 credits a month, which fits the $60 or $90 Pro tier; choose a premium voice and it can swallow the bundle on its own, for a difference most callers will not consciously notice. Carrier charges apply only if you connect your own number through a provider such as Twilio, and those are billed by the carrier.

Botpress. The voice channel is listed on the Enterprise plan only, so there is no self-serve price to quote and no way to trial it on a card. If the agent’s main job is answering the phone, this is where the comparison stops being close.

Pricing in detail

Botpress

Verified on the Botpress pricing page on 31 July 2026. Prices are shown in both cadences below; do not mix them when comparing. AI usage is included on every tier, so there is no separate model bill, though as noted above the bundled allowance is proportional to your conversation allotment rather than bottomless.

• Free, $0. 100 conversations a month, hard-capped with no top-ups or overages, 100MB of vector storage, 100MB of file storage and 1,000 table rows. Because it is hard-capped, treat it as a trial rather than a small production tier.
• Plus, $189 a month, or $150 a month billed annually. 250 conversations, then packs of 100 at $65. Three seats, WhatsApp, white-labelled webchat, 1GB of vector storage, 10GB of file storage, 100,000 table rows.
• Team, $939 a month, or $750 a month billed annually. 1,500 conversations, then packs of 100 at $50. Unlimited seats, teams and role-based access control, routing and assignment, team analytics, real-time collaboration, 5GB of vector storage, 50GB of file storage, 500,000 table rows.
• Enterprise, custom. Custom volume, SLA and security review, and the voice channel, which no lower tier offers.

A storage add-on at $40 a month adds 100,000 table rows, 1GB of vector storage and 10GB of file storage on Plus and Team. And watch the conversation definition before you model costs: at least two messages in the billing month, anonymous sessions counted separately, month-straddling conversations counted twice. Spam is the only stated exclusion.

Voiceflow

Voiceflow no longer advertises its self-serve prices, and its billing documentation tells buyers the dashboard is authoritative. The prices still exist, however, in the live pricing data behind Voiceflow’s own credits pricing table. On 1 August 2026 that data showed Pro at $60, $90 or $120 a month for bundles of 12,000, 18,000 or 24,000 credits, and Business at $150, $250, $500 or $1,000 a month for 30,000 up to 200,000. Annual billing grants the whole year’s credits upfront at a 10% discount, so the entry Pro tier is $648 for 144,000 credits.

One inconsistency deserves a mention. Voiceflow’s 2025 launch announcement lists 10,000, 15,000 and 20,000 credits for the same Pro prices, a fifth less than the live data now shows. We cannot tell you which figure your account will get, so confirm the allowance in your own dashboard before you model costs on it.

The rest of the plan structure, from the documentation:

• Free plan. A one-time credit grant that does not renew. In Voiceflow’s own words it “isn’t suitable for production usage”; an evaluation sandbox, not a small production tier.
• Pro is for individual builders, with access to all models. Business adds fallback models and priority support. Enterprise adds private or custom cloud hosting.
• Credits reset each billing cycle and are shared across the organisation. Editor seats are billable add-ons; viewer seats are free and unlimited, though a viewer given project-level edit access counts as an editor for billing.

The usage tariff is published in unusual detail. Each message a user sends, or request to the Conversations API, costs one credit; phone time costs 10 credits a minute; speech synthesis runs from roughly 0.125 to 40 credits per thousand characters depending on the provider; transcription from roughly 0.56 to 4 credits a minute; and model usage from 0.0125 to 3.75 credits per thousand tokens, tracking each provider’s own prices. The table is populated dynamically from Voiceflow’s systems, so check the live page rather than trusting any static article, including this one.

What happens when you run out

The two platforms have taken opposite positions, and it is worth understanding which risk you are buying.

On Voiceflow, with automatic top-ups disabled, exhausting the balance can stop the agent replying mid-month. Enable automatic top-ups and Voiceflow “automatically purchases additional credits when your balance hits zero”, with nothing charged in advance. Top-ups off risks an interruption in front of customers; top-ups on prevents that but lets spending continue until someone notices.

On Botpress you no longer get that choice on paid plans. Auto-recharge cannot be turned off; instead you get usage warnings as you approach quota and a notification for every pack purchased. Packs recur month to month until you remove them in billing settings, and unused pack conversations expire at month end rather than rolling over. The Free tier is the mirror image: hard-capped, with no overages at all.

So on Voiceflow the failure mode is a silent agent, and on Botpress it is a growing invoice. Whichever you choose, give someone responsibility for watching the number.

Channels and integrations

If your customers message you on WhatsApp, this section decides the question on its own.

Botpress ships webchat from its free tier and offers official prebuilt integrations for around ten conversational channels, including WhatsApp, Instagram, Messenger, Telegram, Slack, Microsoft Teams, Discord, Viber and SMS, with a wider hub covering CRMs, Notion, Google Sheets and Gmail and an SDK for building your own. Counts drift as the hub grows, so check the hub rather than quoting a number, and check the pricing page for which channels sit on which tier.

The exception, and it is a large one, is voice. Botpress lists the voice channel on Enterprise only, so a self-serve customer who wants the agent to answer the phone has no route to it. Voiceflow covers the website and telephony natively on paid tiers; its other channels are reachable through the Conversations API, which means development time and ongoing maintenance rather than a switch.

For a UK retailer where a meaningful share of enquiries arrives through WhatsApp or Instagram, that gap is the whole argument, and it points at Botpress. For a business whose customers ring rather than type, it points just as firmly the other way.

Building experience and collaboration

Both give you a visual builder, branching, variable capture and custom code where the drag-and-drop runs out.

Voiceflow is the easier first day. In our own builds, non-technical people reached a working prototype faster, and independent user reviews tend to score its interface higher. Real-time collaborative editing is part of the product on paid plans, with five workspace roles; editors are billable, viewers free.

Botpress asks more of you up front and gives more back later: finer control over logic, structured data and custom events. In our experience a non-developer needs a few weeks, not an afternoon, to become properly competent; treat that as our observation rather than a measured figure. There is no per-seat charge, which becomes a real advantage as a support team grows, but simultaneous editing, routing and role-based access all arrive together at Team.

The honest framing is not which is easier, but who is building. Give Voiceflow to a conversation designer or a marketing manager. Give Botpress to someone comfortable reading API documentation.

Knowledge bases and agent behaviour

Both platforms train an agent on your own content, use vector storage, accept documents and URLs, and answer from that material.

Botpress goes further with its Autonomous Node, which hands the model the decision about when to search the knowledge base, call an API or escalate, rather than making you draw every branch. It also documents a family of specialised agents, including Summary and Vision Agents for conversation summaries and image understanding. For enquiries you cannot predict, this behaves noticeably better than a flowchart; it also makes testing harder, because the agent’s path is no longer fixed.

Whichever you pick, the knowledge base is where projects quietly fail. An agent trained on a PDF from 2023 will answer confidently and wrongly, and no amount of prompt tuning fixes stale source material. Sort the content before you build the bot.

Analytics and testing

Received wisdom has aged badly here, in both directions.

Analytics. Voiceflow’s dashboard is no longer the afterthought older reviews describe. Its current documentation describes a customisable dashboard whose widgets include resolution and deflection rates, customer satisfaction, evaluations, chats, interactions and unique users, latency, credit usage, playbook, tool and knowledge-base usage, and estimated time and money saved. Transcripts capture full chat and voice history with logs, metadata, custom properties and evaluation results. Botpress documents usage and event analytics plus an Analytics Agent. If your project needs to justify itself to a finance director, both can now produce something presentable; test the specific report you need during the trial.

Testing. Voiceflow ships a proper test suite: scripted test conversations, AI-simulated users acting out a scenario, exact-response and LLM-as-judge checks, routing and tool-call checks, batch execution and saved past runs. The Studio Tests tab is manually triggered, but its separate test platform adds scheduled runs that execute without anyone pressing a button, on one-off or recurring cron-style schedules. Botpress is no longer manual-only either: its developer toolkit ships automated evals that simulate conversations and assert on responses, tool calls, state and workflow progression, including LLM-as-judge scoring and a dedicated regression type, though this is a code-and-CLI workflow rather than a no-code screen. The old claim that neither platform has a regression suite is now wrong in both directions. Voiceflow’s testing is the more accessible for non-developers; Botpress’s is the more automatable for engineering teams.

Data protection and hosting: a UK procurement checklist

Both are North American companies processing conversation data that will frequently contain personal information. Both publish Data Processing Agreements with UK-relevant transfer terms. Certification is evidence of security practice, not proof of UK GDPR compliance; you still have to do the work below. Before you deploy either with UK personal data:

1. Read the DPA and check how it takes effect. Voiceflow’s is incorporated into its standard terms rather than signed separately; Botpress publishes its own. Larger deployments should still get contractual confirmation in writing, and an agency operating the agent for a client should map the full controller, processor and subprocessor chain first.
2. Confirm roles. In the normal case you are controller and the vendor is processor for conversation data; Botpress’s DPA also makes it a controller of usage data. Record this in your Article 30 register.
3. Check transfer safeguards. Both DPAs incorporate Standard Contractual Clauses with UK-specific provisions. Confirm the version you sign includes them.
4. Get the subprocessor list. Voiceflow’s sits in its DPA annexes; Botpress’s in its DPA schedules. Diarise the change-notification window so objections are possible.
5. Ask about hosting and residency. Voiceflow offers private or custom cloud hosting at Enterprise only. Ask both vendors, in writing, where your conversation data rests, and put the answer in the contract.
6. Set retention deliberately. Botpress documents 30 days for logs, 90 days for conversations, messages and events, and indefinite retention for files unless an expiry is set, with earlier deletion via its APIs. Voiceflow commits to deletion on request; confirm retention specifics for your plan in writing.
7. Test deletion. Run a subject-access and erasure drill before go-live, covering conversations, files, recordings and knowledge sources.
8. Decide what the bot may hear. Health data is special-category data under UK GDPR; children’s data attracts extra protection, and financial data is sensitive in the everyday sense even though it falls outside Article 9. Treat all three with care, and design the agent to deflect them rather than collect them.
9. Map the third parties. Whichever LLM, speech and telephony providers you actually configure will receive conversation data. Name them in your privacy notice.
10. Consider a DPIA. Automated conversations with the public at scale, voice recording, or anything touching vulnerable users will often justify one. It is cheaper than the alternative.

Alternatives worth a look

These two are not the only options, and for some briefs both are the wrong shape. If all you need is a website agent answering questions from existing content, a simpler knowledge-base tool such as Chatbase gets there in an afternoon. If the real job is triaging support tickets, the automation built into your existing helpdesk may beat bolting a platform onto the front of it, and Botpress Desk is now competing for exactly that ground. If the agent is voice-first at serious volume, compare the specialist voice platforms before settling. If you need full control of hosting and data, open-source frameworks remain the route, at the cost of maintaining them. And if the work is internal rather than customer-facing, the agent tooling built into products like ChatGPT Work and Claude Cowork may cover it without a bot-building platform at all.

How to run a trial that tells you something

A fortnight of structured testing will answer the question better than any comparison article, including this one. Pick your twenty most common real enquiries and use those, not invented examples. Build the same agent on both platforms; it is a day’s work and settles arguments no spec sheet can. Send a small slice of real traffic, then read the usage figures, because estimating costs before you have data is guesswork. Test the handoff to a human properly, including out of hours. Read fifty transcripts start to finish. Try deliberately awkward questions, including ones you would rather the bot refused to answer. And set your Voiceflow tests to run on a schedule, so the suite keeps checking the agent after you stop thinking about it.

Budget for a small paid month on both. Voiceflow’s free plan is a one-time credit grant, and Botpress’s free tier is capped at 100 conversations with no top-ups, so neither will carry a fortnight of real traffic. One oddity to watch: Botpress’s own sources disagree on whether testing in the Studio emulator consumes your allowance. Its emulator documentation says those conversations are not saved and do not count against quota, though AI operations do, while its pricing material has suggested otherwise. Watch your workspace meter during the trial rather than assuming either way.

Getting it built properly

The platform choice is the easy part, and it matters less than people expect. What separates an agent that deflects support tickets from one that irritates customers is the quality of the content behind it, a clean escalation to a person, and somebody reading the transcripts every week and fixing what they find. That last part is the one nobody budgets for, and it decides whether the project is still running in a year. It is also where our AI and automation work tends to earn its keep.

Not sure which way your numbers point? Send us your expected channels, monthly traffic and team size, and we will estimate the likely Botpress and Voiceflow costs before you commit to a build. No charge, no obligation: tell us about your project.