Your keys, your providers
Set up OpenAI, Claude, Gemini, Grok, Mistral or DeepSeek in AI Integration, test a key in one click, and pick the model. Keys are sealed on the server, never sent back to a browser, and stay out of git and exports.
Operations
Health Check AI Reporting
The numbers of every API checked on your schedule, explained by the AI model you choose, and mailed when you want to be told.
Analytics already counts every call, error and millisecond of your APIs and servers. A report reads them on its schedule, runs checks with your thresholds, and asks OpenAI, Claude, Gemini, Grok, Mistral or DeepSeek, with your own key, to say what needs attention and how to fix it. The mail goes every time, when something is wrong, or only for critical problems, with the tokens and the cost of its run.
See it live
Step through it, slow it down, or open the full canvas.
api-maker/features/ai-reportingLive- Verdict
- —of the checks
- Tokens
- 0read · written
- Cost
- $0of this run
- Code
- 0to write
It reads what Analytics already counts. On its schedule, a report reads the calls, errors, latency, servers and admins of its window: the last hour, day or week.
How it works
It reads what Analytics already counts.
On its schedule, a report reads the calls, errors, latency, servers and admins of its window: the last hour, day or week.
Checks find the problems, with no AI.
Error rates, p95, failing and slow APIs, traffic that fell, servers short of memory. Each check has your threshold.
You choose when to be told.
Every time, when something is wrong, or only for critical problems. Nothing to tell: no mail, no model, no token.
AI says what it means and what to fix.
OpenAI, Claude, Gemini, Grok, Mistral or DeepSeek, with your key. Keep the prompt of API Maker, add your instructions, or write your own.
A clear mail, with its cost.
The numbers beside the window before, the checks, the analysis and the APIs to look at. Every run and every call is kept with its tokens.
What you get
Checks before AI
Error rates, p95, failing and slow APIs, server errors, traffic that fell or jumped, servers short of memory or disk. Each check has your threshold. They decide the verdict and cost no token.
Told when you want
A daily summary, a mail only when something is wrong, or only for critical problems. The same problems are not told twice within the minutes you choose, and one mail can say when they are gone.
The report you need
Choose the window, the parts of the numbers it holds, how many APIs each list has, the tone, the language, and the prompt: the one of API Maker, your instructions after it, or your own with placeholders.
Cost known before it runs
The editor estimates the tokens of a run and its cost at the prices of the model, and the cost of a month of the schedule. Every mail ends with what its run cost.
Every call logged
Every run in the History, every call to a provider in the AI calls, filtered by period, provider, purpose, status and key. Token usage by day, model, purpose and report.
Root user and admins
The root user reports on every admin; each admin on its own account. The root user may share its keys with admins, with a number of tokens a month for each.
An AI Report on Analytics
The Analytics page has an AI Report button: the time range on screen, explained in a window, sent by mail or turned into a scheduled report in one click.
An example
The payment gateway at 3 a.m.
At 03:10 POST /pay starts failing 12 % of its calls. The hourly report sees the error rate pass its threshold, asks the model, and mails the team: payments fail more than usual, here are the calls, check the key of the gateway. Nobody had to watch a dashboard, and the run cost a fraction of a cent.
Mention the payment APIs first.Compare with the targets of our SLA: 99.9 % success, p95 under 800 ms.Say which team owns each API: /orders is the shop team, /pay the payments team.Added after the prompt of API Maker; the numbers of the report always come after them.
≈ 3.4k tokens in · up to 8k out · $0.0013 to $0.02 a run · 720 runs a month at most · $0.94 to $12.19The month is the most it could cost: a report which mails on problems asks the model only when a check finds one.
Watch it
On the real admin panel, with real data.
Good to know
- The analysis is as good as the model: a larger model writes a deeper one and costs more tokens.
- Mails need the mail server of the root user (SMTP). Until it is set, reports still run: read them in the History.
- The numbers come from Analytics: what it does not collect, a report can not see.
Questions
Which AI providers work?
OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Mistral AI and DeepSeek, with your own keys. A gateway or a proxy of your company which speaks the same API works too, and any model id your key reaches.
Does a report send our data to the AI provider?
Only the numbers of Analytics a report holds: calls, errors, latencies, paths of APIs, servers. You choose its parts. Rows of your databases are never sent, and a report with AI turned off sends nothing to any provider.
How much does it cost?
A report reads a few thousand tokens and writes a few hundred words. The editor shows the estimate at the prices of the model before you save, every mail says what its run cost, and a run with nothing to send asks no model.
Can admins use the keys of the root user?
When the root user shares them: with every admin or the ones it chooses, the providers it chooses, and a number of tokens a month for each. An admin's own key always comes first.
Can I write my own prompt?
Yes. Keep the prompt of API Maker, add your instructions after it, or write your own with placeholders like {data}, {checks} and {verdict} in double braces. A prompt without the data gets the data after it.