A platform you build on, or a thing you install

Botpress describes itself as the enterprise grade AI agent platform for customer support, with no technical ceiling and no per seat cost. Its own feature list runs to an AI native helpdesk, helpdesk integrations connecting to Zendesk, Intercom or Freshdesk over OAuth, an AI Agent Studio combining no code and pro code tools, an Autonomous Engine described as model powered reasoning for complex multi step support tasks, knowledge bases, human handoff with full conversation context, AI Triage that classifies and routes tickets, Tables for storing customer and conversation data, deployment to WhatsApp, Instagram, Messenger, Slack and email, integrations including HubSpot, Notion, Jira, Salesforce and Calendly, several named model providers to choose between, and an Agent Developer Kit for custom integrations and logic.

The bill is built on conversations. Free is $0 with 100 conversations, no top ups and no overages, 3 seats and 3 AI agents. Plus is $150 a month billed annually or $189 billed monthly, with 250 conversations a month included, additional packs of 100 at $65, 3 seats, unlimited AI agents, the WhatsApp channel, whitelabel webchat that removes Botpress branding from the widget, a Help Center on a custom domain and technical support over live chat. Team is $750 a month billed annually or $939 monthly, with 1,500 conversations included, packs of 100 at $50, unlimited seats, teams, routing and role based access control, and team analytics. Enterprise is custom.

Their definition of a conversation is the most precise on any of these pages, and it is unusually unfavourable to them, which is a point in their favour. Any exchange with at least two messages from the end user in the billing month counts as one. A conversation spanning two months counts in both. Multiple anonymous sessions each count separately. Three minutes of a voice call counts as one. Model handled, human assisted and emulator conversations are all counted the same. They also publish why they bill this way rather than by resolution: conversations are predictable, so you know your monthly cost before the month starts, whereas resolution pricing creates disputes about what resolved means and raises your bill as the bot gets better.

Model usage is a second layer, sold at cost. Plus includes $25 of AI usage and Team includes $150, with the page stating that AI usage is the cost of calling a language model and that every conversation includes ten cents of it, covered by them. At 95 percent of your AI spend limit they automatically buy a $10 credit grant on your behalf at cost, and unlike conversation packs those grants do not expire at month end, carrying over for up to twelve months. Storage is tiered too: table rows at 1,000 on Free, 100,000 on Plus and 500,000 on Team, with vector and file storage scaling alongside and an expansion add on at $40 a month. All of that is what Botpress published on the checked date. It moves, so read their pricing page and ours before treating any of it as current.

Written to be used rather than to be waved at. If any of these describe you, this is the wrong product and the rest of the page will not change that.

• You have something to build. The Studio, the Autonomous Engine and the Agent Developer Kit are a platform with, in their own phrase, no technical ceiling, and an agent that reasons through a multi step task is a different animal from one that retrieves a passage. There is no logic layer here of any kind, and nothing to extend if you outgrow what we do.
• The agent has to act on something. Tables store customer and conversation data, AI Triage classifies and routes, and the helpdesk integrations connect to Zendesk, Intercom or Freshdesk over OAuth. Askably cannot call an API, read a record or write anywhere, and no setting changes that.
• You want the model to be your decision. Several providers are named on their own page and model usage is billed at cost with the arithmetic printed. If someone at your end has an opinion about which model touches customer data, or wants to see the raw usage cost, they can have both.
• Their pricing argument is genuinely good and you should read it even if you buy elsewhere. Billing by conversation rather than by resolution means you know the monthly cost before the month starts, and it avoids a meter that goes up as the assistant gets better at its job. That is a real critique of half this market.
• Seats are not a cost. Team is unlimited seats with teams, routing, role based access control and team analytics, and even Plus includes whitelabel webchat and a Help Center on your own domain. For an actual support operation with people in it, that shape is coherent in a way a flat widget price is not.

• The smallest paid plan is a tenth of the price. Plus is $150 a month billed annually, or $189 month to month, for 250 conversations. Starter here is $29 a month for 1,000 replies. A reply and a conversation are different units and you should compare their definition against ours, but a factor of five on the entry fee is not explained by the unit.
• Nothing has to be built before anything works. Index your files or crawl your own site, set a threshold, paste one script tag. There is no studio to learn, no agent to design, no table to model and no developer kit to read.
• Answers arrive with numbered references to the passages behind them, so a wrong answer is traceable to the document that caused it rather than debugged through a reasoning trace.
• The refusal is a threshold you set rather than behaviour you engineer. Under your match score, the question stops before a model is called and the visitor receives your fallback sentence, and retrieved material is handled as data rather than as instructions to follow.
• There is no second meter and no auto recharge. Botpress runs a conversation allowance and an AI spend allowance side by side, with a $10 credit grant bought automatically at 95 percent. Here the reply allowance is the whole story: 1,000 at $29, 5,000 at $99, unmetered on Agency.

Read off their own site on that date, quoted narrowly. Prices and packaging in this category change often, so treat this as a snapshot and check both before you decide.

Listing this is more useful than filling it in. Every line here is something that was not clearly published, or was behind a sign-up.

• Whether the Free plan's 100 conversations are a monthly allowance or a one off allotment was not clear. The comparison table printed 100 against Free where the paid tiers carried a monthly marker, and no wording resolved it.
• Whether the $25 and $150 of included AI usage reset each month, and what happens to any unspent part of them at month end, was not stated alongside the figures.
• The published settings note that the $10 recharge amount and the 95 percent trigger are not configurable yet and that a monthly spend cap is coming, so there is currently no published way to cap total model spend.

Should I just use Botpress?

If you are building an agent that has to reason through steps, call systems and act, yes, and it would be a waste of your afternoon to argue otherwise. If you want a widget that answers questions from your own pages by Friday, the platform is overhead you will pay for twice, once in money and once in time.

How does a reply compare with a conversation?

Theirs is coarser. A conversation is any exchange with at least two end user messages in the billing month, so several answers sit inside one. A reply here is a single answer returned to a visitor. Count your own traffic both ways before comparing $150 with $29.

Can I choose the model?

Not here. Botpress names several providers on its own page and bills the usage at cost. What you set here instead is the match threshold, the caution level, the reply length and how many passages are retrieved.

Is there anything to deploy or maintain?

No. One script tag, and either uploaded files or a crawl of your own site that obeys robots rules and follows your sitemap. There is no studio, no agent definition, no table schema and no developer kit.