What AI Agent Platforms Charge For, and When They Stop

Five platforms will sell you an agent that does work on your behalf. Their monthly prices look close enough to line up in a table. Put them side by side and the table tells you almost nothing, because the number on the card is not the whole bill and it is not what decides when the agent stops.

Two things sit underneath it. One is whether the cost of calling a model is inside that price or outside it. The other is what the platform does when you reach the end of whatever it counts. We read what Botpress, MindStudio, Gumloop, Lindy and Relevance AI publish on both, on November 1, 2026, and the five answers are five different shapes. These are not the AI features bolted onto an automation platform, which charge on their own axis and which we compared in how automation tools charge for AI.

Quick Answer

The model bill sits in a different place at every one of the five. MindStudio puts it outside the plan price and says it adds no markup from MindStudio. Relevance AI puts it in a second currency. Botpress folds a fixed allowance into each conversation. Gumloop converts it into credits and charges a fee on top. Lindy spends it from the same credits it spends on everything else. Grouping them that way is our reading; each vendor’s own wording is quoted below.

A free start means something different at every one of the five too. Botpress and MindStudio have tiers that do not expire. Lindy gives you a credit grant that does. Gumloop has no free tier on its pricing page at all, only a trial that wants a card. Relevance AI has retired its free plan outright.

Where the Model Bill Sits

MindStudio keeps it outside and says so in the price itself. Every tier on its pricing page is written as a figure plus the words + usage. The free tier is $0 per month plus usage. Its usage reference says the underlying cost of running models is passed through at cost, with no markup from MindStudio. You are paying the model providers through MindStudio at their price, and paying MindStudio separately for the platform.

Relevance AI keeps it outside too, but gives it its own currency. Platform work is counted in Actions. Model and tool use is drawn from Vendor Credits, which its documentation describes as passed through at wholesale with no markup. Two meters, two balances, two things that can run out independently.

Botpress folds it in at a stated rate. Its pricing page says Every conversation includes $0.10 of AI usage, covered by us. On the paid tiers it prints the result of that rate against the conversation count, so the tier with 250 conversations shows $25 of AI usage included and the tier with 1,500 shows $150. The free tier’s card prints the conversation count and no AI figure at all.

Gumloop converts it and charges for the conversion. Its documentation pegs a credit to half a cent — $1 buys 200 credits — and a single agent run draws on four separate things: the model reading and writing, each connector tool it calls, the time it spends working, and an orchestration fee on top of those three. The fee is eight percent by default on Pro. Enterprise plans can negotiate a different one.

That fee produces the one genuinely counter-intuitive rule in the set. Bring your own API key and the orchestration fee rate doubles to sixteen percent, because the model spend has left the credit balance entirely. Gumloop’s own worked example moves from 459 credits to 93. The rate goes up, the credits go down, and the two numbers point in opposite directions — which means neither one on its own tells you whether bringing a key is cheaper. The provider invoice you now pay directly is not in either figure.

Lindy spends it from the same pot as everything else. Its documentation prices a credit — Credits cost $0.01 per credit on base plans — and says you spend them only while Lindy is working. The model choice is made per task, by Lindy or by you. There is no separate model line.

What a Free Start Means at Each One

Botpress. A tier that does not expire: 100 conversations a month, three agents, three seats. We opened one on November 1, 2026 without entering a card, deployed a bot with a webchat channel, and read the counters. Conversations showed 0 of 100. Storage showed 0 of 1,000 table rows and 0 MB of the 100 MB allowances. A fourth counter showed AI spend at $0.00 of $10.00 — a figure the pricing page’s free card does not carry, although it sits exactly where the ten-cent rate and a hundred conversations would put it. Botpress does not say that is how the free limit is set, so we are reporting the alignment, not a derivation.

MindStudio. Also open-ended: one agent, 1,000 runs a month, access to models through its router without your own keys. The catch is the one already named. The tier is $0 per month + usage, so the platform is free and the model calls are not.

Lindy. A grant, not a tier. The free card reads No credit card required and Free credits last 7 days, against $50 of credits. Seven days is the constraint that matters: the money is generous and the clock is short, so this one rewards knowing what you want to test before you start.

Gumloop. Its public pricing page shows two columns, Pro and Enterprise. There is no free row. What there is instead is a fourteen-day trial of Pro, and its documentation is direct about the terms: the trial requires a card, is a one-time offer per customer, and rolls into a paid Pro subscription when it ends unless you cancel first. Cancel before the end and the card is not charged while you keep Pro for the rest of the window.

Relevance AI. The documentation is unambiguous: The Free plan is retired for new signups and existing accounts. The pricing and documentation pages we read carry no trial either. We have not tested for invited or sales-led routes, so what we can say is that the published ones are gone.

What Counts as One, and Whether You Can Predict It

Every one of the five counts something. What it counts is different enough that the allowances cannot be compared as numbers at all. The builders built into automation platforms have the same problem with their own units, which we took apart in four AI builders, four different gates.

Botpress counts conversations, and defines one carefully: Any exchange with at least two messages from the end user in the billing month counts as a conversation. That is a count of exchanges. It is not a measure of what the model cost to produce them, which is why the AI spend meter exists separately.

MindStudio counts runs, and its documentation settles the manual case in the same sentence: A run is a single execution of an AI Agent. Trigger it by hand or let it fire on its own, and either way it counts as one. The other four describe their units without saying that, but they are counting different things, so the silence is not directly comparable.

Gumloop counts credits, and publishes enough arithmetic that you can work a run out in advance if you know the model cost: divide it by half a cent, add a credit for each successful connector call, add five credits per minute of active processing, then add the fee. It also tells you what is free — Failed calls are not charged. and waiting for your reply costs nothing. Be careful with that one. It is a rule about tool calls, not about runs: if the agent called a model before the tool call failed, that part was still work.

Lindy counts credits too, and this is where the predictability stops. The price per credit is published and the ranges for broad kinds of work are published — a few credits for a lookup, up to a few thousand for something built end to end. What is not published is a way to turn a specific task into a credit figure before you run it. You can budget by category. You cannot quote a job.

Relevance AI counts Actions, and draws the line at the tool: An Action is counted when an Agent runs a Tool — one email or a long workflow, the same one Action either way. Whatever the model did inside it comes off the Vendor Credit balance instead.

What Happens When You Hit the Limit

This is the part a monthly figure cannot show you, and it is the part that decides whether an agent left running overnight can hand you a bill. Three of the five answer it directly. The other two answer part of it.

Botpress runs two rules at once, and they are not the same rule. On the conversation count, paid plans add a hundred-conversation pack automatically at 95 percent of the quota, and the FAQ is clear that this is not something you switch off: You never get cut off. On AI spend, paid plans buy a $10 credit grant at the same 95 percent mark. The free tier gets neither. Credit grant auto-recharge isn’t available on Free/Community plans. and, on storage, Free is hard-capped with no add-ons. Free has no paid top-ups or overages. Paid plans add conversation packs and AI spend grants automatically, under the rules Botpress publishes.

Gumloop stops by default, and its overage has a ceiling of its own. Run out of the monthly credits and the agents stop, because credit overage is something you turn on rather than something that is already on. Turn it on and work continues at half a cent per credit — but Pro overage is always capped., at a default ceiling of a million overage credits per billing period, and Agents stop once the cap is reached. There are two floors here, not one.

Lindy pauses and says what that means. When the pool runs low it pauses the actions that use credits and tells you; when it is empty, the documentation puts it plainly — Nothing breaks, and there’s no surprise bill. Paused actions resume when credits reset at the start of the next cycle.

MindStudio answers for money and not for runs. Set a monthly workspace budget and the rule is explicit: Once the workspace exceeds this limit, subsequent usage will be suspended until the next billing period. Per-user budgets behave the same way. What we could not find anywhere is the other half — what happens when the run count itself is used up and you do not buy more. The pricing FAQ says you will be notified and can upgrade or buy runs. It does not say what happens if you do neither.

Relevance AI has the controls and not the default. Automatic top-ups exist, but they are something you start rather than something that runs: Spend Controls is available to Pro and Team plan users (monthly and annual). Separately, usage limits let you impose your own stop — Usage limits allow you to hard stop usage when you reach a specified limit. Both of those are things you configure. On the pages we read, what happens if you configure neither and the balance empties is not stated.

Who Each One Fits

If the point is to try an agent before paying anything, two of the five are shown to let you. We opened a Botpress free workspace without entering a card, and Lindy’s pricing page says in as many words that none is required. Between those two the difference is the clock: the Botpress allowance does not expire, and the Lindy grant lasts seven days. MindStudio has a free tier too, but it does not say whether a payment method is needed to open one or to run models on it, so we are not counting it here.

If you are putting an agent somewhere it will run unattended, read the stop before the price. A hard cap, an opt-in overage that is itself capped, and a pause are three different answers to the same risk, and only one of the three is a bill. That is the comparison worth making here, and it is not visible on any of the five pricing cards.

If you want to see what an AI step does on a free plan of an automation tool instead, we ran that test in AI on a free plan.

If you need to forecast cost before you commit, the shape of the unit decides how far you get. Gumloop publishes enough arithmetic to estimate a run. Botpress and MindStudio give you a count you can map to your own traffic, with the model cost either inside it or beside it. Relevance AI gives you two balances to track instead of one. Lindy gives you a price per credit and a category, which is a budget rather than a quote.

What We Could Not Confirm

• What MindStudio does when the monthly run count is exhausted and you do not buy more. The budget rule is documented; the run rule is not.
• What Relevance AI does when Actions or Vendor Credits empty and spend controls have been left off.
• How many credits a specific task consumes on Lindy before you run it. The per-credit price and the broad ranges are published; a way to compute one job is not.
• A like-for-like cost for a check that returns no matching data. The five charge for different underlying activities, so an empty result does not necessarily mean no usage. Gumloop comes closest by excluding failed tool calls and idle waiting, but a call that succeeds and returns nothing is not the same thing.

FAQ

Which of these can I actually test for free?

Botpress and Lindy without a card — the first we opened ourselves, the second says so on its pricing page. MindStudio has a free tier but does not state whether a payment method is required. Gumloop’s trial needs a card and converts to a paid Pro subscription if you forget to cancel. Relevance AI has retired its free plan.

Does a free plan mean I pay nothing at all?

Not on MindStudio, where the plan line reads plus usage and the model calls are billed through at cost. Botpress includes an AI usage allowance alongside the conversation quota, and the free workspace we opened showed those as two separate counters. On Lindy the model cost comes out of the credit grant.

Can an unattended agent run up a bill?

On a Botpress paid plan, yes by design: packs and grants are added automatically rather than optionally. On Gumloop, only if you have enabled overage, and even then there is a cap above it. On Lindy, the documentation says there is no surprise bill. For MindStudio and Relevance AI, set a budget or a usage limit yourself rather than relying on a default we could not find.

Why is there no price ranking here?

Because a conversation, a run, a credit and an Action are not the same thing, and we have not run one workload across all five to convert between them. Ranking them on headline price would hide exactly the differences this page is about.

Sources

All pages read on November 1, 2026. Quoted wording is taken from the pages as published on that date.

• Botpress — pricing. Free tier contents, the conversation definition, the per-conversation AI rate, the 95 percent rules and the Free exclusions all come from this page and its FAQ.
• MindStudio — pricing, usage pricing reference and billing settings. The run definition, the pass-through wording and the budget suspension rule.
• Gumloop — pricing, credits and cancelling a plan or trial. Credit arithmetic, the orchestration fee, the overage cap and the trial terms.
• Lindy — pricing and pricing documentation. The free grant, the per-credit price, the work ranges and the pause behaviour.
• Relevance AI — pricing overview, plans and credits, spend controls and usage limits. The retired free plan, the Action definition, the Vendor Credit wording, the top-up prices and the two kinds of control.
• The Botpress counters quoted in this article are from a free workspace we opened on November 1, 2026 and had not yet put any conversations through.