AI Integration¶
AI Integration is where you give API Maker the keys of the AI providers you already use. Features of API Maker then use them : Health Check AI Reporting first, and the AI Report button of Analytics.
In short
- Set up one provider or more : OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Mistral AI, DeepSeek.
- The server seals each key. It never goes back to a browser, and never into git.
- Test asks one short question and shows the answer, the time and the tokens.
- The root user may share its keys with admins, with a number of tokens a month for each.
- Every call is logged with its tokens and its cost.
| Where | The root user : Settings → AI Integration. An admin : Dashboard → Operations → AI Integration, or the user menu. |
| Who | The root user and admins. A developer uses the AI of its admin through the features. |
| Providers | OpenAI, Anthropic, Google Gemini, xAI, Mistral AI, DeepSeek. |
| Keys | Sealed on the server with AES-256-GCM (passDBEncryptDecrypt). Never shown again, never in git, never exported. |
| Costs | Estimated at the list price of each model, or at the prices you write. |
The providers¶
| Provider | Models offered first | Get a key |
|---|---|---|
| OpenAI | GPT-5.1, GPT-5, GPT-5 mini, GPT-5 nano, GPT-4.1, GPT-4.1 mini, GPT-4o mini | platform.openai.com/api-keys |
| Anthropic Claude | Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1, and the ones before | platform.claude.com |
| Google Gemini | Gemini 3 Pro (preview), 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite | aistudio.google.com |
| xAI Grok | Grok 4, Grok 4 Fast, Grok 3 mini | console.x.ai |
| Mistral AI | Mistral Large, Medium, Small | console.mistral.ai |
| DeepSeek | DeepSeek Chat, DeepSeek Reasoner | platform.deepseek.com |
A model which is not in the list works too. Press List models in the dialog of a provider : API Maker asks the provider which models your key reaches. You can also write any model id by hand.
Set up a provider¶
- Press Set up on the tile of the provider.
- Paste its API key.
- Choose the model. The dialog shows its prices and how much it reads.
- Press Test. API Maker asks one short question with these settings, before anything is saved. You see the answer, the time and the tokens.
- Press Save.
| Setting | What it does |
|---|---|
| API key | Sealed on the server as soon as it is saved. To change it, write another one. Empty keeps the one you have. |
| Model | The model the features use. The default one is good for reports : fast and not expensive. |
| Thinking effort | For the thinking models of OpenAI and Anthropic : Low, Medium or High. More effort gives a deeper analysis and costs more tokens. |
| Longest answer | The most tokens an answer may have, thinking included. A report needs about 1,500. Keep room for the thinking of reasoning models. |
| Wait at most | How long a call may take before it is given up. |
| On | Off, nothing uses the provider. Its key stays. |
| The default provider | Used when a feature or a report names no provider. |
| Address (Advanced) | Another address which speaks the same API : a gateway or a proxy of your company. Empty for the API of the provider. |
| Your prices (Advanced) | US dollars for a million tokens, when your contract is not the list price. They make the estimates, never a bill. |
Which model for reports ?
A health report reads a few thousand tokens and writes a few hundred words. A fast model is enough : GPT-5 mini, Claude Sonnet 5.5 or Haiku 5.5, Gemini 2.5 Flash, Mistral Medium, DeepSeek Chat. Take a larger model when you want a deeper analysis.
Share the keys of the root user¶
The root user may share its keys with admins. An admin then uses them when it has no key of its own.
- Share with : nobody, every admin, or the admins you choose.
- Providers : the providers you share. Empty shares every provider you set up.
- Tokens a month : the most tokens each admin may use of your keys in a month (UTC). 0 means no limit. A call which would go over the limit is refused before it is sent, and the admin is told why.
The page shows how many tokens each admin used of your keys this month. The log of the calls of the root user lists these calls too : you pay for them.
An admin sees on its own page whether the root user shares its keys, which providers, and how much of the month is used. Use them when I have no key of my own turns them off for its account. Its own key always comes first.
What is logged¶
Every call to a provider is a row of the log : the test of a key, a report, a preview, the AI Report of Analytics. Read it in Health Check AI Reporting → AI calls.
| Field | What it says |
|---|---|
| Tokens | Read, written, the thinking among them, the ones read from the cache of the provider. |
| Estimate | The tokens read, as API Maker estimated them before the call. |
| Cost | At the prices of the model : the list price, or yours. |
| Status | Answered, cut short (the longest answer was reached), declined by the model, failed, over budget. |
| Time | How long it took, every try included. A busy provider is tried once more. |
| Key | Your own key, or a key the root user shares. |
Where it is kept¶
| What | Where |
|---|---|
| The settings of each person | am_ai_integrations of the main database. The whole document is sealed, only its owner and its date are in the clear. |
| The log of the calls | am_ai_calls of the database of the logs (or the main database), kept 90 days by default. |
None of them is part of a commit or a pull of git, nor of an export of the settings. The keys stay on the servers of the installation.
Keys are secrets
Make a key for API Maker alone, with a spending limit at the provider. Remove it at the provider when you stop using it.
Related¶
- Health Check AI Reporting : reports by mail, written by AI.
- Logs and dashboards : the numbers Analytics collects, and its AI Report button.