A self-hosted backend platform that combines a Supabase-compatible API surface with an embedded analytical column-store.

Trex preserves the Supabase wire contracts for Auth, REST, and Functions, so existing clients and tooling work unchanged. On top of that it embeds an analytical engine you can point at Parquet on S3, BigQuery, ClickHouse, another Postgres, and other sources — queried in the same SQL as your application data.

The published image is ghcr.io/p-hoffmann/trexsql:latest. The easiest way to run a full stack (Trex + Postgres + PostgREST + Realtime) is via the provided compose file:

git clone https://github.com/OHDSI/trex.git
cd trex
docker compose up -d

Edit the environment: blocks in docker-compose.yml to set JWT secrets, S3 credentials, and any other configuration before starting in a real deployment.

Once up, Trex listens on:

• :8001 / :8000 (TLS): HTTP for the web UI, edge functions, GraphQL, MCP, auth
• :5433: Postgres wire protocol (psql, JDBC, pg_dump)

• Auth: signup, sign-in, magic links, OAuth/OIDC, sessions, API keys.
• REST auto-generated over Postgres at /trex/rest/v1.
• GraphQL auto-generated over Postgres at /trex/graphql, with subscriptions over LISTEN/NOTIFY.
• Edge functions with an encrypted secrets store.
• Analytical engine: federate across Postgres, MySQL, BigQuery, ClickHouse, S3 and others in one SQL statement; read open table formats (Iceberg, Delta, Parquet); full-text, vector, and spatial search.

Trex reuses and extends several open-source projects, including Supabase (Storage, Edge Runtime, CLI), DuckDB, Postgres, PostgREST, PostGraphile, and Apache Arrow / DataFusion. Forks are maintained as submodules and retain their upstream licenses.

Apache-2.0.