Skip to main content
Observe is where you watch real traffic flow through Maitai. Once you’re sending requests (see Quickstart), Maitai organizes them into a hierarchy you can browse , Application > Intent: and lets you drill all the way down to one request to see exactly what happened. Come here when you want to answer: “What is my app doing in production, and why did this specific response come out the way it did?”

​
Start with the structure

Maitai groups traffic so you can navigate by product area rather than scrolling a raw log.

Applications

The top-level container for a product or service, its traffic, intents, and config.

Intents

Action types within an Application (also called action_type).

Intent Groups

A cross-application layer for shared monitoring, testing, and datasets.

​
Then drill into the traffic

Sessions

Debug multi-turn behavior (including agent runs) via session timelines and task details.

Request Overview

Inspect a single request, create reruns, and compare prompt/model changes.

​
Watch quality on live traffic

Two tools continuously judge real production traffic as it flows through Maitai:

Sentinels

Intent-group-level watchers that evaluate each request and flag faults.

Monitors

Sample live traffic, judge it against configurable rules, and store verdicts over time.
New to the Maitai object model? Core Concepts explains Company > Application > Intent, plus Sessions and Intent Groups, in one page.

​
Where to go next

Once you can see a failure, the Test pillar is how you catch it before shipping, curate a test set and prove your fix. For a guided end-to-end path, see the Quality Loop recipe.