Honestly nowadays I'm paying for most of the tokens just because it's so cheap. It makes zero difference to me and it just minimizes friction. If you want to be complete, you'd also get them to sign up for an OpenAI account. Keep in mind there are nuances like rate limits; some clients don't have high rate limits but want a high‑rate‑limit application built quickly. I usually use my own OpenAI API key.
I don’t sell AI agents, but running one is cheap: unless you’re using the most expensive model (GPT‑4.5) and paying for everything yourself, the cost is marginal. With cheaper models like GPT‑4.1 mini, you won’t notice the expense unless you’re being spammed constantly.
They’re essentially free to run—about $20 a month plus any token usage. If you want to replicate the setup, just follow one of my courses from start to finish. The agent is just the harness with connectors and context for your environment. Platforms like Grockbot abstract away credential management, making it easy to connect everything and unlock more value.
Um, so here's the really interesting thing about what's happened over the last year or so when I started my agency: OpenAI and AI credits in general were super expensive—I was spending thousands of dollars every month on OpenAI credits for my writing business and several hundred bucks per month for things like enrichment and personalization of email first lines. Since then, costs have dropped precipitously. Where they were before versus where they are now, you can realistically run AI 24/7 on every corner of your business unless you're using the absolute best model, which is only a few percentage points better than the previous generation. You'd be spending about three bucks a month, so you can cover most of the OpenAI credits yourself. It's not like it used to be; now it's very easy and simple. If you do this, you typically avoid the rate‑limit issues I see a lot of people run into on lower‑tier accounts—OpenAI tier five is pretty trash, you might be limited to a few thousand tokens per minute, which causes problems. So OpenAI credits are a nuanced topic, but I would not charge new clients for OpenAI credits; I’d bundle all the AI fees into my pricing because I have partnerships with various providers. In your case, maybe you can’t say that yet, but if you get a partnership you could. Regarding custom tools—if it’s something like Make.com, I’d have the client pay for it, but the good news is that Make.com‑type tools usually cost under $50 or maybe under $100 a month, so it’s low enough that it doesn’t matter much. There are a couple ways to approach offers: you could swallow everything up and charge the client a slightly larger fee (the simplest, lowest‑friction approach), but that adds liability on your end and complicates handoffs—if you and the client split, how does the client get access to what you built? Do they get it or do you take it away? I prefer to have the client own most of their infrastructure, aside from the OpenAI API key, which is simple enough to walk them through in ten seconds if we ever do a handoff. That way I have zero liability, they have the accounts, and I just use their password to access them. It’s simple and easy for everybody.
Most of the time the client pays for services like Apollo, Appify, OpenAI, and NAN. You can use your affiliate links so you get a cut, but the client is the one covering the subscription cost. They enter their payment details directly into the platform, so you never handle their credit card info. The risk is low, and as a beginner you generally don’t charge the client for those recurring costs directly.