Why MoClaw Supports BYOK (Bring Your Own Key)

U.S. consumers waste an estimated $32.84 billion per year on subscriptions they don't use, and AI is making the problem worse. MoClaw is also a $20/month subscription, but only for the infrastructure that runs your agents. The AI calls themselves run on a bring your own key (BYOK) model, billed directly by your provider and never marked up by us.

We're going to defend that split, show you when it wins and when it loses, and admit the things BYOK on a managed platform genuinely can't do.

Key Takeaways:

$20/month covers a dedicated cloud machine with persistent memory, scheduling, multi-channel messaging, sandboxed execution, and 1,000 platform credits.
Bring your own API key (BYOK) and your AI calls are billed directly by OpenAI, Anthropic, or Google. We never mark them up.
BYOK doesn't make MoClaw blind to your prompts. Your data still passes through our machine. It does move your AI calls under stricter API-tier privacy terms.
Heavy users on long-running Opus jobs can pay more on BYOK than on ChatGPT Pro's $200/month flat cap. We'd rather say that than hide it.
Everything we store for you is one-click exportable. Switching providers is a settings change.

Infrastructure as Subscription, Compute as Usage

Most AI products bundle two very different things into one fixed monthly fee. The first is the infrastructure that runs the product: servers, scheduling, sandboxes, memory. It costs the vendor money 24/7 whether you use it or not. The second is the compute that powers the AI itself, which is metered down to the token. Charging a flat fee for both feels simple, but it tilts the economics. Light users subsidize heavy users, and forgotten subscriptions silently fund the company.

According to a C+R Research survey, 42% of consumers have forgotten about a subscription they're still paying for. AI products are not a special category. ChatGPT bills $20/month, Claude Pro $25/month, and the new wave of agent platforms (such as Manus, Genspark, and Lindy.ai) keep the same shape. Manus runs credit-based pricing starting at $20/month, but a single complex task can burn 500 to 900 credits, which makes a "monthly plan" feel a lot like a prepaid card with bad fees.

MoClaw is also a $20 monthly subscription. We are not claiming to be subscription-free. We're claiming the line is drawn somewhere defensible. The cloud machine is fixed-cost capacity, so a flat fee is honest. The AI calls are not, so they're metered by your provider. You see every token in their dashboard, not ours.

That's the deal. The rest of this article is about whether it's the right deal for you.

What You Actually Get for $20

Sarah is a freelance market researcher who runs one agent every night: scrape three industry newsletters, summarize, drop the digest in her inbox by 7 AM. The agent uses about 60,000 tokens of Claude Sonnet per night on her own Anthropic key, which costs her around $0.30 a day. Her MoClaw bill is $20 a month. Her Anthropic bill is roughly $9. Total: $29 a month for an agent that would otherwise cost her four hours a week to run by hand.

Her $20 covers a dedicated cloud machine that:

Runs agents on a persistent file system, so memory and skills survive between runs
Schedules tasks (cron-style) and exposes webhooks and chat channels (Telegram, Slack, email)
Sandboxes browser automation, file IO, and tool use
Includes 1,000 credits for built-in features like image generation that we host directly

For comparison, an equivalent self-hosted OpenClaw on a small cloud VM runs $30 to $80 a month for the server alone, before the API costs you'd be paying anyway.

When BYOK Wins, and When It Doesn't

Most pricing posts cheat on the TCO question by ignoring the cases where their model loses. Here's the version with the loss case included.

Usage profile	MoClaw + BYOK monthly bill	Closest single-vendor option	Who wins
Light: a few agent runs a week, mostly Haiku/Mini	~$22 to $30	ChatGPT Plus $20 (no agents)	MoClaw, easily
Moderate: daily agent, Sonnet-class models	~$35 to $80	Manus Plus $40 to $200	MoClaw, on cost and control
Heavy: long-running multi-agent flows on Sonnet	~$120 to $220	ChatGPT Pro $200 (no agents, OpenAI only)	Toss-up
Very heavy: dozens of Opus tasks per day	$300+	ChatGPT Pro $200 (no agents, OpenAI only)	ChatGPT Pro on cost; MoClaw on capability

If you'd hit ChatGPT Pro's $200 cap every month and you're happy living inside OpenAI's ecosystem, ChatGPT Pro can be cheaper than BYOK. We'd rather say that out loud than dress up an open-ended bill as automatically lower.

Most readers still come out ahead on BYOK for two reasons:

You only pay for what you use. Cancel a project, cut a workflow, take a vacation, and your AI bill drops the next day. A flat plan doesn't.
You aren't locked into one provider. Route coding to Claude, image work to Gemini, fast classification to Haiku. Switch overnight when prices drop. The model market moves fast: GPT-4o is already roughly 10x cheaper than GPT-4, and you should be able to capture that without renegotiating a plan.

What BYOK Actually Does for Your Data

Most BYOK marketing implies "your data never touches our servers." That isn't true on any managed agent platform, including ours. If we run your agent, we process your prompts and outputs in memory to do that. The honest question is what changes when the API call itself goes out under your key instead of ours.

What BYOK genuinely changes:

The AI call is governed by API-tier terms, not consumer terms. OpenAI's API privacy policy explicitly says they do not train on API data. The consumer ChatGPT product runs under different terms. Anthropic, Google, and others draw the same line. Moving from a consumer subscription to BYOK quietly upgrades the privacy posture of the LLM call itself.
You see the bill, line by line. Every token shows up in your provider's dashboard. No black-box credit conversion, no "this task cost 700 credits, why?".
Provider switching is a settings change. Swap one key for another and your agents pick up the new model. Conversation history, skills, and memory live in your MoClaw workspace, not inside any single provider's silo.

What BYOK does not change:

Your prompts pass through MoClaw's machine. We orchestrate the agent loop, run tools, and forward the model call. We do not log, store, or train on those prompts, and they never leave your dedicated workspace except as the outbound call to your provider. But we can see them at execution time. That's true of any managed agent host, and we'd rather you hear it from us.
It is not a substitute for self-hosting if your threat model is "no third party can ever process this data." If you genuinely cannot let MoClaw process the prompt, the answer is OpenClaw on your own server. We'd rather lose the customer than mis-sell.

What we do commit to: everything we store on your behalf, including agent configs, memory files, and conversation history, is exportable in one click. Our privacy practices are public and updated as we change them.

How BYOK Works in MoClaw

Setup is a settings change, not an integration:

Get an API key from OpenAI, Anthropic, or Google AI Studio.
Paste it into Settings → API Keys.
Run an agent.

You can hold keys from multiple providers simultaneously and route different tasks to different models. Keys are encrypted at rest and only decrypted in memory at the moment of an outbound call. We never log, display, or transmit the raw key elsewhere. Rotation is one click on your side, and immediate on ours.

You don't have to use BYOK at all. The 1,000 included credits cover what most users need for built-in features like image generation, web browsing, and hosted skills, and you can buy additional credit packs if you'd rather see one number on your bill. BYOK is the path for users who want metered AI compute and provider freedom. Credit packs are the path for users who want one number to put on an expense report.

Who Shouldn't Use MoClaw + BYOK

A short, direct list:

You only need a chatbot. A ChatGPT subscription is genuinely simpler. You're not running agents, scheduling tasks, or wiring in messaging channels.
You'd hit the $200 ChatGPT Pro cap every month and you only need OpenAI models. The flat cap may save you money. You give up agent capabilities, model choice, and exportability, but if those aren't valuable to you, the cap wins.
Your accounting team requires one fixed bill. BYOK is variable by design. Use credit packs or pick a flat-priced agent platform.
Your threat model excludes any third-party processor. Self-host OpenClaw on infrastructure you own.

For everyone else, users who want managed infrastructure, transparent metering, model freedom, and the ability to leave with their data, MoClaw + BYOK is built for you.

FAQ

What is BYOK (Bring Your Own Key)? BYOK lets you connect your own API key from OpenAI, Anthropic, or Google to MoClaw. Your AI calls are billed directly by the provider at their published rates instead of through us, and you keep full visibility into per-token usage in their dashboard.

Do I need BYOK to use MoClaw? No. Every plan includes 1,000 credits, and you can buy more in credit packs. BYOK is for users who want their AI compute metered through their own provider account.

Is my API key safe? Yes. Keys are encrypted at rest and decrypted in memory only at the moment of an outbound call. We never log or display the raw key. You can rotate or revoke from your provider's dashboard at any time.

Does BYOK mean MoClaw can't see my data? No, and we don't claim it does. Your prompts pass through your MoClaw machine so we can orchestrate the agent. We don't log, store, or train on them, but a managed platform that "couldn't see your data" couldn't run agents for you. If your threat model excludes third-party processing entirely, self-hosted OpenClaw is the right answer.

How does BYOK compare to ChatGPT Pro at $200 a month? ChatGPT Pro is a flat $200 a month for unlimited use of OpenAI's chat experience, with no agents. MoClaw + BYOK is $20 a month plus metered API calls, with full agent capabilities and any provider you want. If you'd reliably exceed $180 a month in API calls and only need OpenAI models, ChatGPT Pro can be cheaper. If you want agents, model freedom, or transparent metering, MoClaw wins.

What happens if I switch providers? Add a second key, point your agents at the new provider, and you're done. Conversation history, memory, and skills stay in your workspace. There is no migration step.

The Bottom Line

	Subscription AI (ChatGPT/Manus)	Self-hosted OpenClaw	MoClaw + BYOK
Monthly fixed cost	$20 to $200	$30 to $80 server	$20
AI compute cost	Bundled	Pay your provider	Pay your provider
Agent capabilities	Limited or none	Full (you build/maintain)	Full (managed)
Data path	Their servers	Your server	Your machine on our infra → your provider
Provider lock-in	Yes	No	No
Setup time	Minutes	Hours	Seconds
Maintenance	None	Ongoing	None
Data export	Limited	Trivial	One click

Infrastructure as subscription, compute as usage. We don't want a forgotten-subscription dollar from you, and we don't want to mark up your tokens. We want to charge you for the cloud machine that's genuinely running, and let your provider charge you for the AI you genuinely use.

Try MoClaw with your own API key →