A Claude Code plugin that gives your agent predictive ability over your
relational data. Ask a forward-looking question in plain English — "which
customers are about to stop ordering?" — and RelQL connects to your data,
maps the schema, writes a RelQL PREDICT query, and scores
it with the RT-J relational foundation model.
In Claude Code:
/plugin marketplace add RelativeDB/RelQL-cc
/plugin install RelQL@RelQL
Either invoke the command:
/relql which of my customers are about to stop ordering?
…or just ask a predictive question and the relql skill activates on its own. It walks the whole path with you:
- Frame the question as target + population + time window.
- Connect your data — Snowflake, Postgres, BigQuery, MySQL, Databricks, or a CSV / pandas DataFrame.
- Language — Python, Java, or Rust, matched to your project.
- Map your tables to a RelativeDB schema.
- Write the RelQL query.
- Wire retrievers over your storage (temporally correct by construction).
- Score with the native RT-J backend and report ranked results.
Churn / lapse, conversion / activation, spend & demand (regression), usage forecasts, recommendation ranking, and risk / status flags. The shipped RT-J checkpoint executes binary classification, regression, multiclass classification (predicted class + approximate probabilities, via the text head), and ranking (top-k via per-candidate existence scoring). Quantile / interval outputs parse and validate but are not yet executable.
.claude-plugin/
marketplace.json # marketplace "RelQL"
plugin.json # plugin "RelQL"
commands/
relql.md # /relql slash command
skills/relql/
SKILL.md # the orchestrating skill
reference/ # RelQL grammar, query cookbook, per-language APIs,
# connectors, native-backend setup
scripts/
run_prediction.py # runnable Python scorer wrapper
- Docs & language reference: https://relql.com
- Engine & libraries: https://github.com/RelativeDB/RelQL
Apache-2.0. © RelativeDB.