Interactive Tour

This tour assumes you have access to a hosted Thyme instance (see Installation) and have committed at least one featureset (see Your First Feature). Open your instance's dashboard URL and follow along.

If you haven't committed anything yet, the Quick Start is the fastest way to get a featureset live before walking the tour.

1. Catalog - what's committed

Open /catalog.

The Catalog page is the system of record for everything you've committed. It has five tabs:

• Featuresets - every @featureset and the features it exposes
• Pipelines - every @pipeline, with input datasets and aggregations
• Datasets - every @dataset, both raw and pipeline-output
• Graph - a dependency graph view of the same objects
• Lineage - the same lineage view that appears on /lineage

Click into any featureset card to see its features, extractors, and Python source. The hover tooltip on the page heading explains what a featureset is - the docs links throughout the UI follow the same pattern.

2. Inspect - query a feature value

Open /inspect.

The Inspect page is a UI wrapper around the GET /features REST endpoint. Enter:

• Entity ID - the key value (e.g. user_42)
• Featureset - pick from the dropdown
• Timestamp (optional) - leave blank for online (latest), or pass an ISO-8601 timestamp for a point-in-time query

Hit Inspect and you'll see the feature values returned by the query server. The Last updated line at the bottom shows how recently the underlying state was advanced.

If a feature returns null or "missing", see Troubleshooting → Features return null / missing.

3. Sources - what's ingesting

Open /sources.

Each card represents one @source decorator. The card shows:

• The dataset the source feeds
• The connector type (Postgres, Kafka, Kinesis, etc.)
• The cursor field and current cursor value (for polling sources)
• The poll interval and max_lateness setting

If the cursor value isn't advancing, walk through Troubleshooting → Source is "stuck".

4. Jobs - what's running

Open /jobs.

Each row is one running job. A job is the runtime materialization of a @pipeline - every pipeline becomes one job at commit time. The columns show throughput, last commit timestamp, and partition count.

Click into a job to see its pipeline operator chain, input/output topics, and (if applicable) the windowed aggregations being computed.

5. Lineage - what depends on what

Open /lineage.

This is the structural graph of your committed objects. Sources flow into datasets, pipelines transform datasets into other datasets, and featuresets read from those datasets. See Lineage for what each node and edge means.

Try this: click any node. The graph dims to gray and re-highlights everything upstream and downstream of the selected node. This is blast-radius mode - the fastest way to answer "what happens if I change this?".

Press Esc or click the background to exit.

6. Query Runs - audit trail

Open /query-runs.

Every CLI or SDK query against the query server creates an audit record here. Each row shows the featureset, entity IDs queried, latency, hit rate (how many requested features had values), the timestamp parameter (if it was a point-in-time query), and the API key fingerprint (in auth mode).

Click into any run to see the full input and output - useful for reproducing a production result locally, or for replaying a failed query against a fix. See Query Runs for the full API.

7. Monitoring - runtime health

Open /monitoring.

The monitoring page embeds the Grafana dashboards described in Operations → Monitoring. You'll see system health, engine performance (events/sec, write latency), query-server performance (latency percentiles, QPS), and definition-service activity.

Use this view to spot stuck pipelines, sudden latency regressions, or commit storms.

Where to go next