Getting Started
Set up Levered end-to-end -- connect your warehouse, define metrics, create an optimization, and integrate the SDK.
This guide walks you through setting up Levered from scratch. By the end you will have a live optimization that serves personalized variants to your users and learns from real conversion data in your warehouse.
Prerequisites
- A data warehouse -- BigQuery or Snowflake with exposure and reward event data (or tables you plan to populate).
- A Levered account -- Sign up at app.levered.dev if you haven't already.
Choose your setup method
Using Claude Code? Install the Levered plugin and Claude will handle the entire setup automatically — including CLI installation:
/plugin marketplace add levered-hq/claude-plugin
/plugin install levered@leveredThen tell Claude what you want to optimize.
Using another coding agent? Install the Levered CLI. All commands support --json for structured, machine-readable responses. See the CLI reference for the full command list.
curl -fsSL https://cli.levered.dev/install.sh | bash
levered loginSet everything up yourself through the Levered dashboard UI. Each step in the guide below includes dashboard instructions.
Setup flow
Levered does not store your event data. Instead, it connects to your warehouse, reads exposure and reward events via SQL, and trains bandit models on that data. The setup has four steps:
- Connect your warehouse -- Give Levered read access to your BigQuery or Snowflake instance.
- Set up your data pipeline -- Create the exposure and reward tables Levered reads during training.
- Define metrics -- Write SQL queries that tell Levered what counts as a successful outcome.
- Create an optimization -- Define the design factors (headline, CTA, layout) and link them to a metric.
- Integrate the SDK -- Serve variants to users with the JavaScript SDK or REST API.
Once live, Levered trains on your warehouse data on a recurring schedule and continuously shifts traffic toward better-performing variants.
Next step
Start by connecting your data warehouse.