How Pyxis works
What is collected, what is rejected, and where the privacy boundary sits.
Pyxis is built around one rule: store aggregates, never individuals. Everything that enters is checked against an allowlist first.
Your API (OpenTelemetry) --> /v1/metrics --+
Your dApp (wallets-kit) --> /v1/wallet-events --+--> PostgreSQL (row-level security) --> Pyxis app
Your website (tracker) --> Umami ------------------------------------------------------> Pyxis appWhat is stored
| Stored | Not stored |
|---|---|
| Request counters and latency histograms per route, per minute | Traces, logs, request or response bodies |
| Wallet event counts per minute (event, provider category, network) | Wallet addresses, signatures, transaction XDR, free text |
| SHA-256 of key secrets; project membership by Stellar public key | IP addresses, query strings, the key secrets themselves |
Web analytics counts come from Umami and follow Umami's own definitions and data handling.
Tenant isolation
Every project's rows carry a project id, and PostgreSQL row-level security enforces that a request can only see its own project. The application connects as a non-owner database role (pyxis_app), so the policies cannot be bypassed by the application. Each request sets its tenant for the duration of one transaction.
Reported, not verified
Wallet events are engagement signals reported by a browser. They say what the frontend claims happened, not what the ledger confirms.