Getting started

FeatureMesh has three entry points, depending on how deep you want to go.

Try it in the browser

Every code example on this site is interactive. Click any playground to edit the query, run it, and see results — no installation needed.

Here's a taste: a FeatureQL query that defines a greeting feature and evaluates it for a given name.

WITH
    -- Define features
    YOUR_NAME := INPUT(VARCHAR),
    A_MESSAGE_TO_YOU := 'Hello, ' || YOUR_NAME || '!'
SELECT
    -- Return features
    A_MESSAGE_TO_YOU
FOR
    -- Bind values to features for evaluation
    YOUR_NAME := BIND_VALUE('FeatureMesh')  -- <-- insert your name here

Under the hood, the playground sends your FeatureQL query to a remote FeatureMesh registry, which translates it to SQL. The SQL then runs in a DuckDB WASM module directly in your browser.

This is the fastest way to learn the language. Start with the FeatureQL foundations or jump to FeatureQL for the Impatient if you already know SQL.

Install the Python library

pip install featuremesh

The featuremesh package gives you a Python client that translates FeatureQL to SQL and executes it against your own data. It supports DuckDB, Trino, BigQuery, and DataFusion as backends.

A minimal example with DuckDB:

from featuremesh import OfflineClient, Backend
import duckdb

client = OfflineClient(
    access_token="your_token",  # from console.featuremesh.com
    backend=Backend.DUCKDB,
    sql_executor=lambda sql: duckdb.sql(sql).df()
)

result = client.query("""
    WITH
        ID := INPUT(BIGINT)
    SELECT
        ID,
        DOUBLED := ID * 2
    FOR
        ID := BIND_VALUES(ARRAY[1, 2, 3])
""")

print(result.dataframe)

The library also provides Jupyter magic commands (%%featureql) for notebook workflows. See the full PIP library reference for all backends, configuration options, and magic command details.

Run the demos Docker container

The featuremesh-demos container is a self-contained environment with everything pre-configured: the FeatureMesh registry, analytics proxy, online serving, and sample data.

It includes:

• Jupyter notebooks walking through batch analytics, real-time serving, and business functions
• Pre-configured backends: DuckDB, Redis, PostgreSQL, and an HTTP server
• Sample datasets for an e-commerce scenario

This is the best way to experience the full platform — batch analytics with BigQuery/Trino, real-time serving with Redis and JDBC, and feature persistence — without setting up infrastructure.

Get started: github.com/featuremesh/demos

Which path should I choose?