AI Health Reports for Your APIs by Email: OpenAI, Claude, Gemini

Intro

0:00 Health Check AI Reporting. API Maker checks the numbers of every API, on your schedule. The AI model you choose explains them. And you get a mail when you want to be told.

Numbers nobody reads

0:13 Every API platform collects numbers: calls, errors, speed, and servers. But people read a dashboard only when they have time, never at three in the morning. An alert fires on one number, with no context. A script that asks an AI model has its key pasted into it, and nobody knows what it costs. With API Maker, checks look at the numbers of every API, on your schedule. The AI model you choose says what they mean, and what to fix.

0:44 Its keys are sealed on the server, and never go into git. And every run shows its tokens and its cost, before and after. In this video: AI Integration, with the keys of six providers, a report, from what it reads to who it tells, a preview, a run, and its mail, the tokens, and every call, logged, and the AI Report button of Analytics.

AI Integration

1:10 A report needs a model, and a model needs a key. So we start where the keys live. On the dashboard, under Operations, open AI Integration. There are six providers: OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Mistral AI, and DeepSeek. Each one shows its logo, its models, and their prices. And here is what happens to a key. It is sealed on the server with AES-256-GCM. It is never sent back to a browser, and never written into git.

1:44 Calm Home uses OpenAI. Click Set up. Paste the API key. After this, the panel never shows it again. To change it, you type a new one. Then pick the model. Each one has a short note. Under the list, you see its price for a million tokens, read and written, and how much it can read at once. GPT-5 mini is fast and cheap. It is good for reports.

2:15 Many companies send their AI calls through their own gateway, to watch them and to limit them. Advanced takes its address. Here, it is the AI gateway of Calm Home. If you leave it empty, API Maker calls the provider itself. Advanced also takes your own prices, when your contract is not the list price. Before we save, we want to know that the key works. Test asks the model one short question, with these settings.

2:44 The answer comes back with the time it took, and the tokens it used. Click Save. OpenAI is ready. It is the first provider, so it is the default one. A feature that names no provider uses it. One more thing about keys. The root user can set up keys too, and share them with every admin, or with the admins it picks. Each admin gets a number of tokens each month.

3:11 A call that would go over it is refused before it is sent. An admin's own key always comes first. And a call made with a shared key is also logged for the root user. The root user pays for it, so the root user sees it.

A report

3:27 Now, the reports. Go back to the dashboard, and open Health Check AI Reporting. It has four tabs: the reports, their history, the token usage, and every AI call. There is no report yet, so the page offers three ideas: a daily summary, a mail when something is wrong, and a wake-up call for critical problems. Calm Home wants to hear about problems, not about every quiet hour. So we start from Tell me when something is wrong.

3:59 The editor has five parts. First, what the report reads: the last hour of Analytics, and which parts of it. That is the health, the kinds of API, the busiest, slowest, and failing APIs, the servers, and the time before, to compare. Every API adds a line for each API, and the estimate at the bottom grows. Fewer parts mean fewer tokens. Next, when it runs. It can run every few hours, every day, on some days of the week, once a month, or with a cron expression.

4:32 This idea runs every hour, at five past, in the time zone of Calm Home. The editor says the schedule in words, and how many runs the next thirty days have. Now the most important part. Before any model is asked, checks look at the numbers. An error rate over two percent is a warning. From ten percent, it is critical. Then the p95 of all calls, and of each API, the server errors, and traffic that fell or jumped.

5:03 And servers that are short of memory or disk. Each check has its own limit. The checks cost nothing, and they decide the result: healthy, needs a look, or critical. Then the mail. Type the address of the team. Then choose when to send: every time, when something is wrong, or only for critical problems. A problem told a moment ago is not told again for six hours. But a new problem is told at once.

5:32 And one more mail can say when the problems are gone. Last, the model and its prompt. Pick the provider and the model. Here, the default one. Pick a tone: short, detailed, or for managers. And pick the language it writes in. The prompt of API Maker already asks for a summary, what needs attention, what goes well, and what to do first. Add my instructions puts the words of the team after it.

6:00 Here: mention the checkout first, and say which team owns each API. My own prompt replaces it, with placeholders for the data and the checks. At the bottom, you see the cost of one run: the tokens the model reads, the most it can write, and the price, at the list price of the model. Then the cost of a month of this schedule, at most. A report that sends only on problems asks the model only when a check finds one.

6:30 Click Save. The report waits for its next hour.

A run and its mail

6:34 So, what happens at five past the hour? Every server looks at the reports every thirty seconds, and the first one to take the moment runs it. It reads the numbers of Analytics, runs the checks, and decides. If there is nothing to tell, the run is noted, and no model is asked. If there is something to tell, the model gets the numbers as short tables, with the problems the checks found.

6:58 Its answer goes into the mail, and the run goes into the history. We do not have to wait for the schedule to see a report. The shop had a busy few minutes, with some errors. We open the report, and click Preview with AI. The result: needs a look. Here are the checks that found something, each with its numbers. Here are the model, the tokens it read and wrote, and what they cost.

7:24 Then the mail, as it would be sent, with what the AI says: a summary, each problem with its fix, and what to do first. A preview sends nothing. Run now runs the report at once, and sends its mail, whatever its checks say. The History keeps every run: when it ran, its result, if the mail went, the model, its tokens and its cost, and what started it. A run opens on its numbers and its checks.

7:55 See the mail shows it exactly as it was sent. First, four numbers next to the time before, then the checks. Then what the model says. Then the APIs with errors, the slowest ones, and the servers. And at the end, what this run cost. Send again sends it to anyone you type. Which server sends these mails? Outgoing Email, on the dashboard. Here, Calm Home sends through the mail server of the root user.

8:23 There is nothing to set up. The root user decides which admins may use its server. An admin can also use a mail server of its own. A test mail checks it in a moment. Sign-in codes always go through the server of the root user.

Tokens and calls

8:40 Models charge by the token, so API Maker counts them. Token usage shows the calls, the tokens read and written, and the cost of a time period. You can see it by day, by purpose, like reports, previews, tests, and Analytics, by model, and by report. AI calls lists every call to a provider. You can filter it by time, provider, purpose, status, and key, or search it. Here, we show the previews only.

9:08 A call opens on its tokens, the estimate made before it, the prices, the model that answered, and its tries. A busy provider is tried once more. The clock sets how long the calls and the runs are kept.

The AI Report of Analytics

9:22 Reports run on a schedule. But sometimes you are already looking at Analytics, and you want an opinion now. Analytics opens on its last hour. We pause it, and click AI Report. We pick the model, the depth, quick or every API, and the language. Click Make the report. It explains the time range on the screen, in the same words as a mail. You can send it from here, or schedule it.

9:48 The editor of a report opens, with this time range already filled in.

Good habits

9:53 A few good habits. Make a key for API Maker alone, with a spending limit at the provider. Start with a report on problems, and a fast model. Quiet hours then cost nothing. And keep the prompt of API Maker. Add your instructions after it, instead of replacing it.

Recap

10:13 To sum up. The keys of six providers, sealed and tested. Reports that read what Analytics collects, with checks you set. A mail every time, on problems, or only when it is critical. Every run and every call, with its tokens and its cost. And an AI Report in Analytics, one click away. Next: sandboxes and runtimes. API Maker. Build faster, spend smarter, deliver better.