Most of you already pay for Claude Code or Codex. Paying us a second time for the same models felt daft, so you can now connect that subscription to Chartnaut and run our agents on it. You decide which models go through your plan and which go through Chartnaut credits, and you can change your mind whenever you want.
You only connect it once
This lives in Chartnaut Desktop, because your subscription signs in on the machine you already use it on. There’s an AI providers page under Settings with a connect button for each one. Sign-in happens in your normal browser, the way it does for everything else, and we don’t keep anything about that account.
You can connect more than one account per provider and pick which is active, which is useful if you’ve got a personal plan and a work one. Subscriptions only work on Desktop, so this is the one feature you won’t find in the browser build.
It matters more here than in a chat app
A chat window that answers questions is cheap to run. Chartnaut’s agents aren’t doing that. They’re writing indicators, definitions, studies and layers, checking each one against the real runtime, dry-running it over history and reading back what it produced before they show you anything. That’s real work and it costs real tokens.
Which means the cost of iterating is the thing that decides how good your research gets. On your own subscription you stop rationing attempts. You ask for the fourth version of a definition because the third marked a few bars you didn’t like, and it costs you nothing extra. That freedom is the actual feature here. The billing is just the mechanism.
What a subscription is worth at API prices
Chartnaut credits are charged at each model’s API rate. A Claude or ChatGPT plan isn’t priced like that. You pay a flat fee every month and the provider eats the difference, and if you’re in an agent most days that difference is big.
Two people have published their own logs. Paweł Huryn ran his Claude Code transcripts through a usage dashboard he built and got $1,588 of usage at API rates in 30 days, over 440 sessions, on the $200 Claude Max plan. That’s about eight times what he paid. His write-up is here.
Kyle Redelinghuys went back further. Eight months of Claude Code, around 10 billion tokens, would have cost him roughly $15,000 at API rates. He paid about $800 for Max over the same stretch, so nearly 19 times. His later months are estimated from session logs rather than counted token by token, and he says as much in his breakdown.
Neither is a lab study, and how far a plan goes depends on how hard you lean on it. A couple of hours a week and it’s closer to a convenience than a saving. We haven’t found anyone who’s published the same kind of log for Codex yet, but ChatGPT Plus and Pro are flat-priced the same way.
So if you’re on Claude Code or Codex already, putting $25 into Chartnaut credits makes no sense. Connect the plan and keep the money. If you don’t have one, a $20 Claude Pro or ChatGPT Plus plan costs less than that top-up. Credits are still there for the browser and phone apps, and for the open-source models.
One more thing
This one costs us money rather than making it, so it’s worth saying why we did it.
We sell a trading workspace. We don’t want to be in the business of reselling tokens at a markup, and anybody paying for Claude Code or Codex and then buying AI credits inside a product is paying twice for the same thing. That was going to irritate people eventually, and I’d rather remove it now than defend it later.
There’s a selfish version of the argument too. When you’re not counting credits you use the agents properly, and people who use them properly are the ones who find out what the platform can really do.
If the connection gives you any trouble, mail me at [email protected] and I’ll help you sort it.
Thanks for reading.
Best,
