A user shares their experience with OpenClaw on a dedicated Mac mini, where switching to subscription billing resulted in credits being used instead of plan limits, prompting a question about the billing mechanism.
Running OpenClaw on a dedicated M4 Pro Mac mini, always-on persistent agent. Claude Max 5x plan. Tonight I switched from API key to subscription via setup-token after seeing a post about it here. Auth worked fine, status confirmed anthropic:manual , and we saw spend on the sub account. not the API account. But here's what I actually observed: The plan usage meters (session, weekly) never moved. Not a single percent. Usage credits started climbing immediately, first exchange in, at roughly $0.40 per back-and-forth. Went back to API key. So my question isn't "does subscription work". I've seen plenty of posts saying it does. My question is: why did my usage go straight to credits without touching the plan limits at all? Is the Agent SDK path just a different billing surface that bypasses the plan quota entirely? Or is there something about my setup specifically? For context: my agent runs a persistent 100k+ token context. With the API key and prompt caching I'm at 99% cache hits and paying for ~600 new tokens per exchange. Without the key, caching doesn't work, so every exchange processes the full context fresh. That alone would make the subscription path expensive even if the plan limits were being used. But the plan limits weren't being used. That's the part I can't explain. Anyone with an always-on setup actually seeing their plan percentages tick up from OpenClaw traffic? Or is the Agent SDK path always credits?
The article analyzes the rapid rise and fall of OpenClaw, questioning whether its popularity was genuine or hyped, and discusses its usage-based pricing model.
Discussion of the current status of using Claude CLI via OpenClaw, referencing Anthropic's policy updates and potential account ban risks, with recent changes to credit buckets for claude -p usage.