If you want to use your own AI key and control spend, begin with a job you can describe and a budget you can review. The useful questions are practical: what should the AI help with, who will check its work, and who will watch the usage?
Consider Maya, who runs a homeware shop in Lisbon. She wants help drafting captions for a collection and answering familiar product questions. Before connecting a key, she can list those jobs, prepare examples and decide what a useful result looks like. That gives her a way to judge the trial beyond whether a reply sounds polished.
What it means to use your own AI key and control spend
Bring your own key means creating an API key in your business's account with an AI provider and adding it to the service where you want to use AI. The key lets that service make requests through that provider account. Keep a record of which account owns it and who is responsible for reviewing its use.
The provider account, the model and the task are separate decisions. A key identifies the connection; a model determines which AI handles a request; your instructions describe the work. Write down the intended job before choosing any of them, so your trial has a clear purpose.
Create a provider key and add it in AI Studio
In Kwamle Media, add your business's provider key in AI Studio and choose the model. You can connect OpenAI or Google Gemini. Anthropic Claude is also supported. The studio includes spending controls, and the dashboard records the provider and model used for each AI request. The platform's AI can draft replies and captions and create images for posts.
- Sign in to the provider's own console with the account your business controls. Confirm who can manage it and review usage.
- Create a new API key using the provider's current instructions. If it offers permission settings, select only the access your intended work needs.
- In AI Studio, open Providers and add the key for the matching provider. Keep it out of message fields, captions and instructions.
- After setting the provider budget described below, choose an available model and try a short caption or a sample product question. Read the result before widening the trial.
Keep your first instructions specific: the product facts, the customer's language and the tone you want. A useful test should show whether the answer respects those facts and asks for missing details instead of inventing them.
Choose a model for each kind of work
Match the model to the job. For Maya, an answer about caring for a mug and a draft introducing a collection need different checks. Use these broad categories to organise your tests:
- Fast models: try these for short, familiar questions with clear answers. Check that they follow your instructions and ask for missing details.
- Large models: test these on wording that needs more care, such as explaining differences between products or drafting a collection story.
- Models with long context: consider these when answering depends on a longer conversation or substantial reference material. Include only information relevant to the task.
Treat these as starting points for testing. Run the same sample questions through your candidates, note whether the answer is usable and record how much editing it needs. Include the languages your customers use. A model that produces an attractive draft still needs checking for facts.
Set a provider budget and review usage
Decide what you can spend over a review period, then set a spending limit using the controls available in the provider's console. Read the current explanation of that setting: does it stop requests or only send an alert? Confirm which account or project it covers and when the period resets.
If a control only warns you, agree who will act on that warning and pause discretionary work. Check the current limits in the provider's official documentation. A budget needs someone to review it, especially when you add a new task or change models.
- Review the provider's usage view after the first tests and on a regular schedule that suits your business.
- Note which jobs you ran and which models you chose, so changes in activity have an explanation.
- Look for repeated drafts, longer instructions or extra image requests when spending rises.
- Reduce work you do not need and test one change at a time before increasing the budget.
Read usage alongside the results your team actually kept. If captions repeatedly need rewriting, examine the instructions and source material before requesting more versions. For a busy period, decide in advance which tasks matter most and who can approve extra work. Keep that decision in your team's working notes.
Keep the key private and replace it safely
Treat the key as a private credential. Keep it out of public posts, shared documents and screenshots, and give access only to people who need to manage the connection. Keep a note of its purpose and owner without copying the secret into that note.
Kwamle Media encrypts your data in transit using TLS and stores it on encrypted infrastructure. Your team still needs to handle any copy of the key privately.
Rotating a key means replacing it and retiring the old one. For a planned replacement:
- Create a replacement key in the same provider account using its current instructions.
- Update the provider key in AI Studio and run a small test with the intended model.
- After the replacement works, revoke the old key in the provider's console.
- Review usage for unexpected activity and update your private record of who manages the connection.
If you believe the old key has been exposed, revoke it promptly. Reconnect with a replacement and test before using AI again.
Start with one job you can check
Start with a familiar job, a model you can evaluate and a budget someone owns. Keep the first trial small enough to review every result. Expand only when you can explain the usage and the work is useful to your team.
The bring your own AI key guide shows where those choices fit in Kwamle Media. Use the guide to choosing an AI model to plan the questions and content you will test.
