AI agent platforms charge in one of six ways: per task, per credit, per workflow run, per seat, unlimited within fair-use windows, or nothing at all if you self-host. The headline price matters less than the billing unit, because the billing unit decides what happens to your bill when an agent takes more steps than you expected. This guide explains each model with real list prices taken from each vendor’s own pricing page on 12 October 2026, then shows what those prices add up to for a solo user, a 5-person team and a 20-person team. We build Shogo, which uses the unlimited-within-windows model, so we’ve also said plainly where Shogo is not the cheapest option.
Why the billing unit matters more for agents than for automations
A classic automation does the same steps every time. If a workflow has four steps and runs 1,000 times, you can work out the cost on a napkin.
An AI agent doesn’t work that way. It decides how many steps to take. It might look something up once or five times, retry a failed tool call, or think for longer on a hard request. That means two runs of the same agent can cost very different amounts on a platform that charges for each step or each token. The more an agent does on its own, the less predictable a metered bill becomes.
So the useful question when comparing AI agent platform pricing isn’t “what does the cheapest plan cost?” It’s “what am I charged for, and who carries the risk when the agent does more work than planned?”
The six AI agent pricing models
1. Per task or per step
You pay for each action the platform completes.
- Zapier is the clearest example. Its free plan includes 100 tasks a month, the Professional plan starts at $19.99 a month and the Team plan at $69 a month for up to 25 users, with the price rising as you pick a higher task tier. Zapier says on its pricing page that AI steps, code and its SDK now all draw from the same shared task pool, and that “AI model tier, code runtime, and connector type each affect how many tasks an action uses.” When you hit your limit, Zapier switches you to pay-per-task billing at a higher per-task rate, unless you turn that off, in which case your workflows pause.
Who it suits: predictable, rule-based workflows where you know how many steps each run takes.
Where it hurts: agents that loop, retry or call several tools per request, because every one of those actions is a task.
2. Credits
Credits are a platform’s own currency. Different actions cost different numbers of credits, often depending on which AI model you use and how long the agent runs.
-
Make has a free plan with 1,000 credits a month. Its paid plans start at $9 a month for Core (10,000 credits), and Pro starts at $16 a month. Make’s help center explains that most modules use 1 credit per operation, but some AI modules use a “dynamic” number of credits based on tokens consumed. If you run out, Make automatically buys extra credits at a 25% higher cost than the credits in your plan.
-
Lindy bills every active user, starting at $29.99 per user per month for 3,000 credits, with larger credit tiers above that, all pooled across the team. Lindy’s own guide puts everyday requests at 2 to 250 credits, “deep work” at 250 to 1,000 and “big builds” at 1,000 to 2,500. Extra credits cost $10 per 1,000, and unused monthly credits don’t roll over.
-
Gumloop charges $37 a month for its Pro plan with 20,000 credits. Its documentation lists four separate charges on every agent run: the model’s cost at $0.005 per credit, 1 credit per connector tool call plus that tool’s own charge, 5 credits per minute the agent is actively working, and an 8% orchestration fee on top of the other three.
Who it suits: teams that want access to many models and are willing to monitor usage.
Where it hurts: forecasting. A credit price tells you very little until you know how many credits your actual agents burn, and that number moves with model choice, run length and tool calls.
3. Per workflow run (execution)
- n8n charges per execution, which it defines as “a single run of your entire workflow. It doesn’t matter how many steps are in the workflow or how much data it processes.” Its Cloud Starter plan was listed at €24 a month for 2,500 executions when we checked (n8n shows prices in euros), and its free cloud plan includes 50 executions.
Who it suits: long, multi-step workflows. Ten steps cost the same as one.
Where it hurts: high-frequency triggers. n8n’s own estimate is that a workflow running every 5 minutes uses about 8,600 to 8,900 executions a month, more than three times the Starter allowance. And if your workflow calls an AI model with your own API key, you also pay the model provider separately.
4. Per seat
You pay a fixed amount for each person, usually with some usage included per seat.
- Dust lists its Pro plan at €24 per seat per month with 8,000 credits per seat, and its Max plan at €120 per seat per month with 40,000 credits per seat. Lindy also bills per active user, so in practice it is a per-seat plan with a credit pool attached.
Who it suits: teams that want a bill that only changes when headcount changes.
Where it hurts: two places. Costs grow with every person you add, even occasional users. And many “per-seat” plans still meter usage underneath, so a seat price alone doesn’t tell you whether heavy users will run out.
5. Unlimited within fair-use windows
Instead of a monthly pool that can run dry, usage is capped by time windows that reset on their own. You don’t get charged more for using the platform heavily. The limit is how much you can do inside each window.
- Shogo uses this model. Its paid plans are unlimited usage within two rolling windows, a 5-hour burst window and a 7-day weekly window, with higher windows on higher tiers and no monthly credit pool. Basic is $8 a month for a single user on a fast model. Pro is $20 per seat per month with all models. Business is $40 per seat per month with the highest windows, SSO, audit logs and per-member spending limits. Enterprise is uncapped. If you want to keep working past a window, you can turn on optional overage, billed at the model provider’s cost plus 20%, up to a hard cap you set. We explained why we moved to this model in Shogo plans are now unlimited.
Who it suits: people who use agents heavily and want a fixed bill, especially for agent work where the number of steps varies run to run.
Where it hurts: “unlimited” is not “infinite.” If you run very heavy work in a short burst, you can hit the 5-hour window and have to wait for it to reset, or pay for overage. Anyone comparing platforms should ask what the windows are on the tier they’re buying, not just read the word “unlimited.”
6. Self-hosted, open source
You run the software on your own server. There’s no subscription to the vendor, but there are still costs: the server, the AI model usage (you pay providers like OpenAI or Anthropic directly), and the time to maintain it.
-
n8n offers a self-hosted Community Edition on GitHub, and its pricing page notes “unlimited executions if you self-host.”
-
Dify offers a free Community Edition you deploy with Docker. Dify describes it as “for open-source enthusiasts, individual developers, and non-commercial projects” under the Dify Open Source License, so read that license before using it commercially. Its hosted Professional plan is $590 per workspace per year with 5,000 message credits a month and 3 team members.
-
Who it suits: teams with an engineer who can run and update a server, and anyone with strict data-location rules.
Where it hurts: “free” ignores your own time and your model bill, which is often the largest line item for agents anyway.
A middle option: hosted by default, deployable in your own environment. Some platforms run on the vendor’s cloud by default but can also be deployed where your data lives. Shogo works this way: it usually runs on Shogo’s cloud, and it can also be deployed in your company’s own cloud account, inside a VPC, or on-premises. That suits teams with security or data-location requirements who don’t want to maintain open-source software on their own.
One more: custom pricing only
Some platforms don’t publish prices at all. When we checked, Relevance AI’s pricing page showed only an Enterprise plan with custom actions and custom vendor credits, and a “Talk to sales” button. That isn’t a billing model you can compare, so if price predictability matters to you, factor in the sales process too.
What these list prices add up to at three team sizes
These are entry list prices from each vendor’s pricing page, before any overage, top-ups or model provider bills. They show how each billing model scales with headcount, not what your exact bill will be.
| Platform | Billing model | Solo user | 5-person team | 20-person team |
|---|---|---|---|---|
| Zapier | Per task | from $19.99/mo (Professional) | from $69/mo (Team, up to 25 users) | from $69/mo (Team, up to 25 users) |
| Make | Credits | from $9/mo (Core, 10,000 credits) or $16/mo (Pro) | Depends on credit tier | Depends on credit tier |
| n8n Cloud | Per execution | €24/mo for 2,500 executions | Depends on plan and executions | Depends on plan and executions |
| Lindy | Per user + credits | from $29.99/mo | from $149.95/mo | from $599.80/mo |
| Dust (Pro) | Per seat + credits | €24/mo | €120/mo | €480/mo |
| Dify Cloud | Per workspace + credits | $590/yr (Professional, 3 members) | $1,590/yr (Team, up to 50 members) | $1,590/yr (Team, up to 50 members) |
| Shogo | Unlimited within windows | $8/mo (Basic) or $20/mo (Pro) | $100/mo (Pro) | $400/mo (Pro) or $800/mo (Business) |
Three things stand out:
-
Per-seat plans get more expensive in a straight line. Lindy, Dust and Shogo all scale with headcount. At 20 people, a per-seat plan is a real budget line.
-
Per-workspace and per-task plans look cheaper for big teams, until you look at usage. Zapier’s Team plan and Dify’s Team plan don’t grow with headcount, but the task tier or credit allowance you need does grow as more people run more agents.
-
The price you can forecast and the price on the page are different things. A credit or task plan’s real cost depends on how much your agents actually do. A seat or windows plan’s real cost is mostly the number on the page.
If you’re worried about AI costs getting out of hand from a different angle, our piece on tokenmaxxing looks at why token spend is a poor measure of whether AI is actually helping.
How to choose an AI agent pricing model
-
Your workflows are fixed and simple: per task (Zapier) or per execution (n8n) is easy to forecast. If you’re already on Zapier and the task bill is growing, see our Zapier alternatives comparison.
-
You want many models and will monitor usage weekly: credits (Make, Gumloop, Lindy) give flexibility, as long as someone watches the meter. Our Make alternatives guide covers the options if credits stop working for you.
-
You want a bill that only changes when headcount changes: per seat, and check whether the seat includes a usage pool that can run out.
-
Your agents take an unpredictable number of steps and you use them daily: unlimited within fair-use windows avoids a pool running dry mid-month. Ask what the windows are.
-
You have strict data rules: self-host if you have an engineer to run it, and budget for your model provider bill. If you’d rather not maintain it yourself, look for a platform that can be deployed in your own cloud, VPC or on-premises, as Shogo can. Our Zapier alternatives comparison notes which tools can be self-hosted.
For a broader comparison beyond price, see our guide to the best AI agent platforms.
Where Shogo fits, and where it doesn’t
Shogo is an open-source AI agent that builds the agents and apps you describe. It’s priced so heavy use doesn’t change your bill: paid plans are unlimited within 5-hour and weekly fair-use windows, with no credit pool and no per-task charges. It works across every major model provider and there’s a free plan with daily free usage. Shogo usually runs on Shogo’s cloud, but it can also be deployed in your own cloud account, a VPC or on-premises.
Where Shogo is a good fit on price:
-
Solo users and freelancers who want unlimited usage for $8 a month on Basic, or all models for $20 on Pro.
-
Small teams who use agents every day and don’t want to forecast credits or tasks.
-
Teams who want a hard cap on any extra spend. Overage is off unless you turn it on, and you set the cap.
-
Teams with security or data-location requirements, since Shogo can be deployed in your own cloud, a VPC or on-premises instead of Shogo’s cloud.
Where Shogo is not the cheapest or best fit:
-
Light, occasional use. If you run a handful of simple automations a month, Zapier’s or Make’s free plans may cost you nothing.
-
Large teams where few people use agents. At 20 seats, Shogo Pro is $400 a month. A per-workspace plan like Dify’s Team tier, or Zapier’s Team plan, has a lower base price if only a few people do the heavy work.
-
Very heavy burst workloads. A single user doing very heavy work in a short burst can hit the 5-hour window. Business has higher windows, and Enterprise is uncapped, but that’s a step up in price.
-
Basic is single-model. The $8 Basic plan is one user on a fast model. You need Pro for advanced models.
Next: how to reduce AI agent costs by matching the model to the task.
FAQ
Which AI agent platforms don’t charge per task?
n8n charges per workflow run rather than per step, so a 10-step run counts once. Per-seat platforms like Dust charge by person, though Dust also includes a credit allowance per seat. Shogo’s paid plans are unlimited within 5-hour and weekly fair-use windows, with no per-task charges and no credit pool. Self-hosted open-source tools like n8n Community Edition and Dify Community Edition have no vendor charge at all, though you still pay for servers and model usage.
What is the cheapest AI agent platform for a small team?
It depends on how heavily the team uses agents. For light use, free plans from Zapier (100 tasks a month) or Make (1,000 credits a month) may be enough. For a 5-person team using agents daily, a per-seat plan with no usage pool, such as Shogo Pro at $20 per seat ($100 a month for five), keeps the bill fixed. Credit-based plans can look cheaper at the start but get harder to forecast as usage grows.
Is there an AI agent platform with a free tier?
Yes. Zapier, Make, n8n Cloud, Dify Cloud and Shogo all have free plans, and Lindy offers $50 in free credits that last 7 days. Free tiers are usually capped by tasks, credits, executions or daily usage, so they’re for trying a real workflow, not running a team on.
What does “unlimited usage” mean on an AI agent platform?
It usually means there’s no monthly pool you can use up, but there are still limits on how fast you can use the platform. On Shogo, paid plans are unlimited within a rolling 5-hour window and a 7-day window, and higher tiers have higher windows. Always check the windows on the specific tier, because “unlimited” never means infinite throughput.
Are self-hosted AI agent platforms really free?
The software can be free, but running it isn’t. You pay for the server, for AI model usage directly with providers like OpenAI or Anthropic, and for someone’s time to keep it updated. Some licenses also restrict commercial use. Dify, for example, describes its free Community Edition as being for individual developers and non-commercial projects.
Can Shogo run in my own cloud or on-premises?
Yes. Shogo usually runs on Shogo’s own cloud, but it can also be deployed in your company’s cloud account, inside a VPC, or on-premises. That’s useful for teams with security, compliance or data-location requirements that rule out a shared cloud.
How does Shogo’s pricing work?
Shogo has a free plan with daily free usage, Basic at $8 a month for one user on a fast model, Pro at $20 per seat per month with all models, Business at $40 per seat per month with higher limits, SSO and audit logs, and custom Enterprise plans that are fully uncapped. Paid plans are unlimited within 5-hour and weekly fair-use windows. Optional overage is billed at the provider’s cost plus 20%, up to a cap you set. Full details are on the Shogo pricing page.
Why do credit-based AI agent platforms make costs hard to predict?
Because an agent decides how much work to do. Gumloop’s documentation, for example, charges for model usage, each tool call, each minute the agent is working, and an 8% orchestration fee on top. Two runs of the same agent can use very different numbers of credits depending on how many steps it takes.
Prices checked against each vendor’s own pricing page on 12 October 2026. Vendors change pricing often, so confirm on their pages before you buy.
Last reviewed and updated: October 2026.
