BYOK economics — why we pay only for security

May 1, 2026 · IsoKron team · 3 min read

Bring your own keys, pay your upstream provider directly, and let us pay for the operator-side security critic on every compile. Here's the economic argument for why this is the right shape.

pricing
byok
economics

The middleware tax

Pick any current AI development platform and ask one question: who pays for the tokens? In most cases, the answer is "the platform pays its frontier provider, and you pay the platform with a markup."

The markup is the middleware tax. It's typically 20–100%. Sometimes it's described as "billing convenience." Sometimes it's described as "value-added pricing." Functionally, you're paying twice for the same model invocation: once for the model itself, and once for the right to use the model through this particular intermediary.

The tax adds up. A team running serious volume on a frontier model can easily pay an extra $1,000–$10,000/month for nothing the underlying provider doesn't already offer.

What BYOK actually means

Bring Your Own Keys is the alternative. You supply API keys for Anthropic, OpenAI, Google AI, or xAI. You pay your provider directly. IsoKron is not in the loop on the bill.

When IsoKron compiles your project, it makes calls against your keys. Your dashboard at Anthropic (or OpenAI, or wherever) shows exactly what was spent. There's no IsoKron-side line item. There's no markup to argue about because there's no transaction between us and a vendor on your behalf.

The savings can be substantial. A team paying $4,000/month in middleware-marked Claude invocations would pay roughly $2,000–$2,500 if the same calls went directly to Anthropic. That's the cost of having someone else hold the keys.

Why we eat the Critic cost

Every compilation in IsoKron passes through a Layer 4 Critic — a security review pass that catches prompt-injection attempts and parasitic-chaining patterns in the compiled output before it commits to your knowledge base. The Critic uses Anthropic Sonnet 4.6 as the primary reviewer and may invoke Haiku 4.5 as a secondary Tier 2 check on suspicious outputs.

We pay for these Critic invocations, not you. That's the only platform-side token spend in the v1 economic model.

Why pay for security ourselves? Three reasons:

The Critic is operator infrastructure, not customer infrastructure. It defends the platform (against malicious inputs that could poison the shared system over time). Customers who paid for their own Critic would have a financial incentive to skip it. We don't want that incentive to exist.

The Critic's prompt template is part of the platform's defense posture. If customers paid for it via their own keys, the prompt would be visible in their provider's logs. We'd rather keep the template confidential.

The cost is small. Critic spend runs around $0.03–$0.05 per compilation. At any plausible customer volume, that's a fraction of a percent of revenue. The economic argument for shifting it to customers isn't compelling.

Budget caps anyway

Even though customers don't pay for Critic invocations, we cap our own spend per workspace to protect the platform from pathological customers (or compromised ones running compile loops). Limits:

24-hour soft cap: $5 (email warning)
24-hour hard cap: $20 (workspace pauses new compilations)
30-day soft cap: $50 (email warning)
30-day hard cap: $150 (workspace pauses)

The hard caps auto-pause new compilations until the window rolls over. Customers operating well below these never see them.

What this means for you

If you're running anything serious on AI codegen today, BYOK is probably already where you should land. The platform that holds your keys is incentivized to keep you locked in. The platform that asks you to bring your own is incentivized to be worth using on its own merits.

We don't claim BYOK is the only honest pricing shape. There are real use cases for "platform manages keys" (sales-led enterprise where the buyer doesn't want to provision Anthropic). But for indie operators, small teams, and any technical buyer comparing platforms on cost transparency: bring your own.

Bottom line

IsoKron is free in v1 at the platform layer. You pay your upstream provider directly for the model calls IsoKron makes on your behalf. We pay for the security Critic that reviews every compile. The math works out to "you save the middleware markup and we eat the security cost." That's the deal.