What constitutes a 'tracked call' in Argoos?
One tracked call in Argoos corresponds to one request made to your AI endpoint. This means every time your application calls a model and you subsequently fire the Argoos tracking request, it counts as one tracked call, regardless of the token count.
Which AI providers and models are supported by Argoos?
Argoos is provider-agnostic and supports all major AI providers and models, including OpenAI, Anthropic, Google Gemini, Mistral, Cohere, and any custom models running on your own infrastructure. It works as long as you can send an HTTP request after an AI response.
Does Argoos store my prompts and outputs?
Argoos only stores your prompts and outputs if you choose to send them. Passing `inputText` and `outputText` in the tracking call is optional. If these fields are omitted, Argoos only collects metadata such as model, latency, tokens, and cost, ensuring you retain control over your data.
How does Argoos differ from simply checking if an API returned a 200 status code?
While a 200 status code confirms a call completed, Argoos goes further by assessing if the AI response was actually useful. It validates structural integrity, checks for required fields, ensures confidence scores are within range, and detects if the response deviates from your defined quality standards, catching silent quality failures that status codes miss.
What happens if I exceed my monthly tracked call limit?
Argoos does not cut off your service mid-month if you exceed your limit. You will receive an email notification as you approach your limit, and the team will work with you to facilitate an upgrade, ensuring continuous visibility into your AI product's quality.