The $3,600 API bill: what self-hosting OpenClaw actually costs

The Mac Mini pitch is hard to argue with: $2/month in electricity, no subscription fees, full control. Here's the part nobody mentions — your API bill.

The token problem

OpenClaw is stateful. Every message includes your entire conversation history in the API request. That context grows.

One power user documented their session context hitting 56-58% of a 400K token window. A simple "what's on my calendar today?" required processing 200,000+ tokens of cached history first.

At Claude's pricing, that's about $1.80 per message. Ten messages a day is $540/month before OpenClaw does any actual work.

Real bills from real users

$3,600 in one month — aggressive automations, session history never pruned

$623 in one month — developer with Telegram, WhatsApp, and scheduled tasks running

$400/month consistently — typical power user, daily use, two chat channels

The pattern is always the same: unbounded session context plus automations that ran longer than expected.

Six things driving your token costs

Session history accumulates indefinitely until it hits the model's context window

Tool outputs (search results, file reads) are stored and resent every turn

Your SOUL.md and config prepend to every API call

Complex tasks require multiple back-and-forth API calls per user message

Using an expensive model for tasks a cheaper one handles fine

Background heartbeat tasks fire API calls when you're not actively using the assistant

The 12-month cost breakdown

ExpenseMac Mini M4VPS ($20/mo)MyOpenClaw Pro Hardware$800 ($22/mo amortized)$0$0 Infrastructure$2/mo$20/moIncluded API costsUnbounded (~$100-400/mo typical)Unbounded$5-30/mo built-in + BYOK option Setup time4-8 hours2-4 hoursTypically ~90 sec with warm pool Maintenance2-4 hrs/month2-4 hrs/monthNone 12-month total~$1,800+ (mid estimate)~$1,680+~$598

Mac Mini beats managed hosting on cost only if your API bill stays under ~$45/month. For most active users, that's not realistic.

What we do differently

We addressed token bloat at the infrastructure level with a 4-layer memory stack:

Sessions persist indefinitely — no daily 4am reset like the default

Old tool results are pruned before they accumulate, not the messages you care about

When the context window fills, old messages are summarized rather than dropped

Before compaction, the agent writes durable facts to daily notes files

Every plan includes built-in AI ($5-30/month depending on tier) — enough for daily use. Add your own key if you want more. The difference: your bill is a number you know in advance.

Start your free trial →