Explanation
Architecture & the pipeline
Elevarq is not a single binary — it is a pipeline of focused components, each with one job and a clear boundary. Understanding how they fit explains why setup has the shape it does and where your data does and does not go.
The pipeline
Three components sit between your databases and your team's workflow, and data moves one way along the chain:
PostgreSQL
databases you operate
Elevarq Signals
read-only collector
Elevarq Analyzer
turns snapshots into findings
Elevarq Workbench
triage findings and push tickets
- Signals connects to each database with a read-only, least-privilege role and gathers the statistics PostgreSQL already exposes, packaging them as portable snapshots. It writes nothing back.
- Analyzer consumes a snapshot and produces findings — prioritised, explained recommendations. It works from the snapshot, not a live connection to your database.
- Workbench is the surface you install and operate. It ingests findings, presents them for triage, and pushes them outward to your ticket system. It holds its own state in an embedded database.
The split is deliberate: the component that touches your data (Signals) is small, read-only, and runs close to the database; the component you operate day-to-day (Workbench) never needs a path to your data at all.
Deployment model
Workbench is single-tenant, single-host, customer-controlled. It installs inside your own infrastructure and is operated by you — there is no Elevarq-hosted SaaS and no multi-tenant path. Each install:
- holds one organization's data;
- activates one license at a time;
- runs as exactly one process — one container, one pod, or one VM (see License tiers & entitlements).
To serve multiple end customers, you run one Workbench install per customer, each with its own license.
Which way data flows
Everything points one way and away from your data. Signals reads statistics; the Analyzer turns a snapshot into findings; Workbench pushes findings outward into your tracker. By default Workbench makes no outbound network calls — license activation and refresh are local artefact operations, and analyzer imports arrive inbound. Nothing writes back to your databases, and nothing phones home.
The boundary that makes this safe to run is covered in the data-boundary & safety model.