Redis Feature Form quickstart

Register providers, define a feature, materialize it to Redis, and serve it.

Redis Feature Form is a feature platform for defining, deploying, and serving machine learning features, with Redis as the low-latency online store at inference time. This quickstart covers one workflow end to end — register providers, apply a definitions file, and read back the result — to confirm Feature Form is set up correctly in your environment.

Before you begin

Before you begin, you'll need:

• A running Feature Form deployment with durable state
• An existing workspace
• A working auth path for ff
• Reachable Postgres and Redis endpoints for your offline-store and online-store providers
• Sample data loaded that matches the quickstart definitions file

Install the ff CLI

The Feature Form CLI ships as the redis-featureform package on PyPI. Do not run pip install featureform — that's an unrelated upstream project. Install into a virtual environment:

python3 -m venv .venv && source .venv/bin/activate
pip install redis-featureform
ff --help

1. Confirm identity and workspace access

If you haven't yet, log in with ff auth login. Then:

ff rbac whoami
ff workspace get --name <workspace-name>

Use the returned workspace ID in later commands.

2. Confirm the built-in env secret provider

ff secret-provider list --workspace <workspace-id>

If a provider uses a secret reference like env:VAR_NAME, make sure VAR_NAME exists in the runtime environment that resolves secrets.

3. Register your providers

See Register providers for steps to register the offline and online providers before applying resources.

4. Review your definitions file

Open the Python definitions file you plan to apply and confirm it declares the resources you want in the workspace. See Typical file structure for the canonical shape.

5. Apply the file

ff apply \
  --workspace <workspace-id> \
  --file path/to/your/definitions.py \
  --wait \
  --wait-for finished

For a dry run first:

ff apply \
  --workspace <workspace-id> \
  --file path/to/your/definitions.py \
  --plan

6. Inspect the results

ff graph workspace stats --workspace <workspace-id>
ff graph feature list --workspace <workspace-id>
ff catalog list --workspace <workspace-id>

You should see the expected graph entries plus the catalog locations created by the applied resources.

7. Query the materialized data

Read sampled rows from a dataset, training set, or feature view in your applied graph. Use a resource name from the ff graph feature list or ff catalog list output in step 6:

ff dataframe query <resource-name> \
  --workspace <workspace-id> \
  --kind dataset \
  --limit 5 \
  --insecure

See Query data for the full set of flags and supported kinds.

Next steps

• Serve features at inference time from a feature view in your graph.
• Update features to iterate on a definitions file you've already applied.