Limited Availability
Product

Build workflows on Apache Iceberg.

Query open tables, build visual data flows, train models, and publish results with the permissions your team already uses.
Analytics

Work directly with Apache Iceberg tables.

Load files, join tables, and use snapshots to reproduce a result after the source data changes. Gnok streams query results as Apache Arrow.

A selected query over sample customer reviews appears above three rows of suggested replies.
Inspect a query and its results together in the worksheet.
Manage your tables
Load files in the workspace or with COPY INTO, evolve table schemas, and query an earlier snapshot by ID or timestamp.
Reuse business logic
Group dependent writes in a transaction, run them in a stored procedure, or register a reusable WebAssembly function.
Connect your tools
Use the workspace editor or connect clients through Arrow Flight SQL, PostgreSQL wire, JDBC, or the Python SDK. Contact us to enable client access.
Explore analytics examples
Data flows

Turn raw tables into reusable datasets.

When Data Flows is enabled in your workspace, build transformations on a visual canvas. For a regional report, filter orders into two yearly branches, total each by region, and join the results.

The two fiscal-year totals feed into Match fiscal years by region, followed by the regional close output.
The fiscal-close flow groups each year by region before matching the two periods.
Build and inspect each step
Filter, reshape, aggregate, and join inputs. Preview rows and inspect unmatched keys before relying on the output.
Validate before publishing
Check required values, unique keys, or reconciled totals. Publish a version when the output meets those checks.
Put the result to work
Attach a published version to a dashboard as a live dataset, or write it to a destination table when enabled.
See how a data flow is built
Machine learning

Train models and compare predictions.

Train on the tables you already use. Score new rows and record the model version alongside each prediction.

For identical inputs, the initial ONNX model scores 0.694873 and the activated candidate scores 0.789680.
Two versions of a sample fraud model score the same inputs.
Train on your data
Fit models for tasks such as churn classification, demand forecasting, and customer segmentation. Tune and evaluate supported models.
Bring an existing model
Import an ONNX model or register a model endpoint you already host.
Check a candidate version
Score the same inputs with the active and candidate versions before switching. Compare predictions with outcomes as they arrive.
Keep models current
Backfill feature groups and update supported models as new observations arrive with online learning.
Explore training and model versions
AI and search

Find answers in your own data.

Build a support search that finds the relevant refund policy and drafts an answer an agent can check against the source.

The worksheet retrieves a sample refund policy and returns an answer that tells the reader to contact support within 30 days of purchase.
The drafted answer uses the sample policy’s 30-day window, original payment method, and processing time.
Combine keyword and vector search
Filter documents by category or date, then rank matches by words, meaning, or both.
Work with text at scale
Classify feedback, extract issues from reviews, summarize results, or draft a reply for each row.
Ask questions in plain language
Ask about a table, inspect the generated query, and ask a follow-up using the same session context.
Set rules for model access
Workspace administrators approve the provider, model, operations, and monthly token budget before language-model calls run.
Explore the policy search example
Streaming

Score new transactions as they arrive.

Keep new events in the same Iceberg tables you query. Use a stream to capture new rows for scoring or downstream processing.

The Gnok worksheet shows a transaction stream query and its results across 32 sample transactions.
Compare stored predictions with the source transaction rows by transaction ID.
Ingest new events
Write incoming records to Iceberg and query them alongside historical data.
Compare different signals
Apply a trained fraud model and flag amounts that depart from a recent moving average.
Refresh downstream results
Use scheduled tasks or dynamic tables to keep derived results current where enabled.
Follow a transaction through the stream
Notebooks and dashboards

Explore results and share them with your team.

Follow a question from a worksheet or notebook to a published dashboard. Give readers filters they can use to answer follow-up questions.

Sample revenue increased in all three regions. East rose from $2,200 to $3,000, Central from $1,700 to $2,300, and West from $1,350 to $1,700.
The notebook example compares January and February revenue by region across two years.
Explore in a worksheet or notebook
Inspect query results, save an example for your team, or pass a result into Python to calculate growth.
Publish a dashboard
Use queries or a published flow as the source for charts and tables. Add a region selector so a reader can explore their results.
Start from a working example
Open shared tutorials in your workspace and follow the steps with sample data.
Follow the notebook-to-dashboard workflow
Security and governance

Publish one dashboard with the right view for each role.

Apply workspace policies when someone opens shared results. On the invoice dashboard, an analyst can work across regions while the East manager sees only East invoices.

The East manager sees only invoice E-102, with the customer email masked.
After the manager refreshes, the disputed invoice disappears and the remaining email stays masked.
Control which data each role can see
Give roles access to tables, filter rows by policy, and mask sensitive columns such as customer emails.
Manage workspace identities
Use passkeys and MFA, connect an identity provider with OIDC or SAML, and provision members with SCIM.
Review access changes
Inspect grants and audit events when you need to trace who changed a policy or access rule.
Compare the analyst and manager views

Try a workflow in your own workspace.