BYOK explained: using AI with your own API key
BYOK gets talked about as though it is obviously better. It is better for some people and worse for others, and which group you are in depends on facts about your usage that no marketing page can know. Here is enough detail to work it out.
What BYOK means
Bring your own key means an application uses AI credentials that you supply, for a provider you chose, instead of bundling AI usage into its own licence or subscription.
In SideNote Pro's case that is literal. The application ships with no AI credentials, no bundled credits and no endpoint of its own. You configure a provider profile with a base URL, a model and your key, and requests go from your machine to that endpoint directly.
| Bundled AI subscription | BYOK | |
|---|---|---|
| Who holds the API key | The vendor | You |
| Where requests go | Through the vendor's server, then to a model | From your machine to your provider |
| Who chooses the model | The vendor, usually from a short list | You, from anything your provider offers |
| How you pay | Flat recurring fee | Metered, by the provider, for what you use |
| Cost when idle | The same as when busy | Nothing |
| Cost when heavy | The same as when idle | Scales with usage |
| Setup effort | Sign in and go | Create an account and a key first |
Why people choose BYOK
- Model choice. You are not restricted to whichever models a vendor has a commercial arrangement with, and you can use a new model the day your provider ships it.
- The right model per task. A fast cheap model for rewriting a sentence, a heavier one for a hard problem, chosen per request rather than per subscription.
- Independent visibility. Your provider's dashboard shows your usage directly, not filtered through an intermediary's summary.
- One less party. Requests do not pass through a vendor-operated relay. That is one fewer organisation holding your prompts.
- Local models. BYOK is what makes pointing the same application at
localhostpossible at all. - No lock-in. Changing provider is editing a base URL, not migrating your working life.
- Nothing wasted on idle months. If you use an assistant heavily some weeks and not at all others, metered billing matches that shape.
Where BYOK is the worse choice
We are not going to tell you BYOK is always cheaper, because that is not true and you would find out.
- Heavy daily use of an expensive model can cost more than a flat subscription. Metered billing is only cheaper below the crossover point, and long conversations with large context reach that point faster than people expect.
- It is not automatically more private. You removed the vendor from the path. Your provider is still there, receiving everything, under its own retention and training policies. If you have not read them, you have not gained privacy, you have changed who holds the data.
- Setup is a real barrier. Creating an account, adding a payment method and generating a key is a genuine ten minutes before the first useful answer.
- You own the operational problems. Rate limits, quota exhaustion, a deprecated model, an expired key: nobody else is going to notice these for you.
- Compatibility is not uniform. An OpenAI-compatible endpoint may not support image input, may ignore
reasoning_effort, and may not return usage metadata.
A reasonable way to decide: if you would be annoyed at paying a subscription during a quiet month, BYOK suits you. If you would be annoyed at watching a meter during a busy one, it does not.
Keeping metered cost predictable
The discomfort of metered billing is mostly about visibility, and it is addressable.
- Set a hard spending cap in your provider's dashboard. This is the real limit, and it is the first thing to do after creating a key.
- Set the optional daily estimate limit in SideNote Pro as a local seatbelt. It works from local estimates and cannot see other applications using the same key, so it complements the provider cap rather than replacing it.
- Use local usage tracking to see which provider and model combinations dominate your spending.
- Read reducing your AI API costs once. Most of the savings come from two or three habits.
What the SideNote Pro licence covers
This is the point most worth being unambiguous about. The one-time SideNote Pro licence covers the Windows application. It does not include any AI usage, and it never will while the product is BYOK.
You pay your AI provider separately, according to that provider's own pricing and terms, for what you actually use. If you use a local model on your own hardware, you pay nothing beyond the electricity and the machine you already own. See pricing on the home page.