Break down install sources and traffic quality in chat

The agent calls the AppMetrica API with a Yandex OAuth token: first it fetches the list of apps in the account through management/v1/applications, then it pulls Reporting API reports — installs, sessions, active devices, custom events, crashes and revenue — broken down by partner and traffic source, app version, OS, device model and country. On top of that it computes ARPU and retention, and compares channels by paying users rather than by install volume.

AppMetrica hands out raw data differently from reports: the Logs API prepares an export asynchronously — the first request queues the job and the file only arrives once it is ready, so the agent retries until it is. There is no pagination there, so a long period is split into date chunks and the row-level installs, custom events and crashes are stitched together by the agent. That is how a breakdown the dashboard does not offer gets built.

The agent writes only where the access level permits it: it creates push campaign groups, sends campaigns and edits app settings. Before sending it shows which group goes out and with which parameters, and waits for an explicit confirmation. A campaign goes out as a group rather than device by device, and after sending the agent re-reads the campaign status to say whether it landed or fell over.

The connection runs as a sign-in to a Yandex account with a code: the agent obtains and renews the token itself, with no keys to copy by hand. AppMetrica does not finish counting reports instantly — the last hours are still loading, so the agent names the cut-off date next to the figures and never passes an incomplete day off as a total. It holds the request rate itself, and if there is no data for the period it says so instead of showing an empty table.