Quickstart

Install Feldera, create an incremental SQL pipeline, then query features from your serving layer. Estimated time: 10 minutes.

Step 1: Install

Pull the Feldera Docker image and start the runtime:

docker pull feldera/feldera:latest
docker run -d \
  --name feldera \
  -p 8080:8080 \
  feldera/feldera:latest

Verify runtime is up:

curl http://localhost:8080/health
# {"status":"ok","version":"0.9.2"}

Step 2: Set a pipeline

Create a pipeline configuration file listing input connectors, SQL transforms, and the feature outputs to serve.

name: txn-features
version: 1

connectors:
  - name: transactions
    type: kafka
    config:
      bootstrap_servers: "localhost:9092"
      topic: payments.transactions
      format: json

outputs:
  - name: card_velocity
    type: http
    key: card_id

Step 3: Write incremental SQL

Put standard SQL in a .sql file. Feldera compiles it into an operator graph that updates results incrementally when new events arrive.

-- features.sql
CREATE VIEW card_velocity AS
SELECT
  card_id,
  COUNT(*) AS txn_count_1h,
  SUM(amount) AS spend_1h,
  COUNT(DISTINCT merchant_id) AS unique_merchants_1h,
  MAX(event_time) AS last_seen
FROM transactions
WHERE event_time > NOW() - INTERVAL '1 hour'
GROUP BY card_id;

Step 4: Run the pipeline

Upload config and SQL, then start via the API:

# Create the pipeline
curl -X POST http://localhost:8080/v1/pipelines \
  -H "Content-Type: application/json" \
  -d @pipeline.json

# Start it
curl -X POST http://localhost:8080/v1/pipelines/txn-features/start

Use the CLI:

feldera pipeline create txn-features --config pipeline.yaml --sql features.sql
feldera pipeline start txn-features

Step 5: Query features

After startup, fetch feature values by key. Results include all events processed so far and update incrementally:

curl http://localhost:8080/v1/features/txn-features/card_velocity/card_9182

# Response (sub-100ms)
{
  "card_id": "card_9182",
  "txn_count_1h": 7,
  "spend_1h": 412.50,
  "unique_merchants_1h": 4,
  "last_seen": "2026-07-10T14:22:38Z"
}

Connectors

Feldera supports these input connectors by default:

Core concepts

Incremental updates

Feldera compiles SQL into an operator graph, with each node storing the state needed for correct results as data arrives. A new event updates only affected nodes. No full-table scan or batch window is required.

Delta state

Every operator node stores changes as a delta, not a full snapshot. SUM and COUNT update by processing the new increment instead of the entire input relation.

Feature serving model

Running pipelines expose outputs as key-value pairs. Call GET /v1/features/{pipeline}/{view}/{key} for the current value for a key. For batch retrieval, use the scan endpoint to export every key at once.