BYOK and the economics of personal AI

We get this question once a week: "Why isn't Opus included in the regular plan?"

Short answer: because including it would make the plan twice as expensive for the 95% of users who don't need it.

The cost shape

A typical message in personal-AI usage is a few thousand tokens of context, a few hundred output tokens. On Sonnet (Claude 4.6), that's about one credit — a millicent or two of vendor cost. On Opus (Claude 4 / Fable 5), the same message costs 5-10x more.

A "credit" in our pricing is a normalized unit pegged to roughly $0.001 of vendor cost. The Pro plan gives 20,000 credits per month for $39. Most heavy users land between 10,000-15,000 credits a month — about 80% of the cap, with comfortable headroom.

If we bundled Opus into the same flat plan, the math snaps. To cover the case where 10% of usage might shift to Opus, we'd have to underwrite 10× the per-message cost on those calls. The honest pricing for an Opus-inclusive plan is roughly $99-129/month — and the 90% of users who use Sonnet for everything would be paying for headroom they never touch.

What BYOK actually is

"Bring Your Own Key" sounds like a workaround. It's actually the most-aligned billing model we could find.

You generate an API key on Anthropic's console. You paste it into Niyra. We encrypt it (AES-256-GCM) at rest. When you use Opus-mode features (Apex), the LLM call goes through your Anthropic account on your billing relationship. The cost shows up on your Anthropic invoice, transparent and per-call.

We don't take a cut. We don't mark up the inference. You pay vendor pricing direct, and our $39/month covers everything else — the runtime, the integrations, the memory, the channels, the voice infrastructure, the support.

When BYOK beats credits

Below 10,000 credits per month: stay on subscription credits. We're cheaper than vendor pricing for casual use because we're amortizing infra across many users.

Above 10,000 credits per month of LLM-heavy work: BYOK starts winning. You're paying vendor pricing direct without our small margin, and you're not pre-buying credit headroom you don't need.

Specifically heavy use: research, deep analysis, contract review, long-document generation. These benefit from Opus quality and burn credits fast. BYOK lets you pay the actual vendor cost of premium reasoning instead of buying it through a subscription markup.

What Apex is

Apex is the Opus-only mode in Niyra. You toggle it per conversation. The model becomes Opus 4 / Fable 5 instead of Sonnet. Reasoning quality goes up; latency goes up; cost goes up roughly 5-10x.

For most tasks — daily email triage, scheduling, summaries, reminders — Sonnet handles cleanly at fast latency and a fraction of the cost. Apex is for the conversations where the quality difference matters: research synthesizing many sources, contracts with subtle clauses, the investor letter you only get one shot at.

Because Apex pricing would dominate any flat plan, it's BYOK-only and alpha-gated today. The alpha gate is a learning thing — we want to understand the workflows people actually run on Opus before opening it to everyone.

What we don't believe in

Hidden markups. You see exact credit cost on every reply. Live, before you send the next message.

Auto-downgrade. If you run out of credits or vendor budget mid-conversation, we pause and tell you. We don't silently swap from Opus to Sonnet — that would change the quality of the work in flight without your consent.

Lock-in pricing. Cancel anytime; unused credits refundable within 14 days; account export is plain CSV/JSON of your data, no encryption-keyed prison.

The principle behind it

The vendor cost is the vendor cost. Trying to abstract it away into a flat plan is how billing-model arbitrage starts — and personal AI is a product where the cost dynamics matter too much to bury. BYOK keeps the economics transparent, and a small per-credit margin on the casual path keeps the lights on.

If you're running Niyra a couple of hours a day, the subscription plan is the right call. If you're running her as a chief of staff for a substantial fraction of your week, plug in your BYOK key and pay vendor direct. Both paths are first-class.