Data & AI workbench
Data intelligence
you can verify.
In your cloud, with your models, under your rules — and every answer shows its work: the SQL it ran, the tables it read, the passages it quoted.
Which countries brought in the most revenue? — Japan and the US bring in 63% of revenue — far ahead of the rest.
01
Verifiable
Every answer carries its SQL, and a figure the data does not back is flagged. Migrations are proven by running both sides; rules warn you when they go stale.
02
Sovereign
Your installation and your database. Claude, GPT or Gemini under your own contracts — or a model you host. Interface in English and Spanish.
03
Governed
Qry acts as the person who asks: their grants and their row filters, on every query — down to the row.
Measured, not promised
Same questions, same data — side by side
Qry against the AI assistant built into a leading data platform, on the same warehouse and the same questions.
- Everyday questions answered right
- Qry: 31 / 31Built-in assistant: 29 / 31
- Hard questions answered right
- Qry: 5 / 5Built-in assistant: 1 / 5
- Median time to answer (shorter is better)
- Qry: 8 sBuilt-in assistant: 14 s
Public sample datasets (TPC-H and a retail sample), the same warehouse, no instructions or skills in either. October 2026.
The difference is method. The other assistant guessed values while writing SQL — it searched for “approved” where the data says “Y”. Qry looks at the data first, and moves to Python when a question needs statistics.
With another model family — GPT through Azure AI Foundry, in the EU — Qry also answered 31 of 31 and 5 of 5. Accuracy does not hang on one vendor.
Every change of model, prompt or skill is measured against a bank of questions before it ships.
How we measureAlso measured
37.5 M
rows in 25 s
Bronze to gold with Lakeflow SQL tasks inside a Databricks 2X-Small warehouse — lineage included.
47 / 47
validations passed
Forge moving an Oracle test estate to AlloyDB: tables, procedures, views, a trigger and scheduler jobs.
One workbench
From a question to a pipeline, in one place
The same permissions, the same history and the same models across everything your team does with data.
Conversations that show their working
Ask in your own words. Qry finds the tables, writes the SQL for the engine they live in, runs it, charts the result and tells you what it means. When SQL is not enough, it writes and runs Python.
- The query, the tables and every step are one click away
- Rewind to any point of a conversation and take another path
- Share an answer with a link, up to the point you shared
Dashboards you can ask about
Ask for a dashboard and refine it. Drill down, compare periods, forecast a series, flag anomalies — and ask a question about what is on the screen.
- “Explain this” on any tile, Q&A over the whole page
- Export to CSV or Excel, share a public link
- Checkpoints, undo and import between installations
Notebooks for SQL, Python and Spark
Cells in SQL, Python and PySpark — the Spark cells on your own cluster — with secrets kept out of the code. Schedule a notebook and every run gets its own clean runtime.
- AI help that writes cells and fixes the ones that fail
- Versioned in Git, merged cell by cell
- Scheduled runs with variables like {{yesterday}}
Workspaces that know your business
Put your documents in a workspace and ask about them. Search finds both meaning and the exact term you typed, and answers cite the passage they used. Skills teach Qry your business rules — and tell you when one stops matching the data.
- Citations that open the document at the quoted passage
- Shared conversations, files and memory per team
- Rules by scope: company, datasource, team or person
Pipelines that run in your engine
Lakeflow builds bronze, silver and gold layers with SQL that runs inside your warehouse — nothing moves through Qry — with quality gates that stop the chain and lineage your catalog can read.
- Jobs as a graph or as YAML, kept exactly as you wrote them
- Run on demand or on a schedule; parameters per run
- Stuck runs closed and stale tables flagged automatically
Governed by design
Qry acts as the person who asks
The model never chooses whose permissions it uses. Every query a conversation runs goes through the same check as the catalog — role, attributes, row filters — and anything that cannot be checked is not shown.
Grants from datasource to table
Users and groups get access to a datasource, a catalog, a schema or a single table. SAP HANA can be granted by module.
Row filters by attribute
Tag tables, write a policy once — “only my region”, “only my company” — and Qry applies it to every query that touches them.
Single sign-on
Microsoft Entra ID or Google, with directory groups mapped to Qry groups.
Fails closed
Search, the catalog and metadata hide what they cannot verify, instead of showing it.
Qry checks every query against your grants; the database account it connects with still sets the ceiling. We recommend read-only accounts.
How Qry handles securitySovereign
Your cloud, your models, your rules
Your installation
Run by IXEN or inside your own Google Cloud, Microsoft Azure or Kubernetes. Either way, each organisation has its own installation and its own database.
Your model agreements
Claude, GPT and Gemini through your Anthropic, Azure AI Foundry, Google Vertex AI or OpenAI account — prompts stay under contracts you already have. Any OpenAI-compatible endpoint for models you host.
Your people
Interface in English and Spanish, built to WCAG 2.2 AA. Budgets and quotas per user and group, so cost has an owner.
IXEN, which builds and runs Qry, is certified at the HIGH category of Spain's National Security Framework (ENS).
What runs where, and what leavesConnectors
Connect what you already run
Native connectors, browsed with the permissions of each user. Files and MCP servers on top.
- BigQueryWarehouses & lakehouses
- DatabricksWarehouses & lakehouses
- SnowflakeWarehouses & lakehouses
- Amazon RedshiftWarehouses & lakehouses
- Arrow Flight SQLWarehouses & lakehouses
- OracleDatabases
- SAP HANADatabases
- SQL ServerDatabases
- PostgreSQLDatabases
- TeradataDatabases
- Starburst / TrinoEngines, apps & streams
- Cloudera ImpalaEngines, apps & streams
- SalesforceEngines, apps & streams
- Apache KafkaEngines, apps & streams
- Files & MCPOn top: CSV, Excel, PDF and MCP servers
Oracle back to 9i through your own Spark cluster. Through Trino or Starburst, whatever your cluster reaches.
Who it is for
One workbench, two kinds of team
Migration
Leaving Oracle, Teradata or SQL Server?
Forge migrates the code, not just the tables — procedures, views, functions, triggers and scheduler jobs — and proves the result by running both sides and comparing them.
- objects migrated
- 34 / 34
- validations passed
- 47 / 47
Our Oracle test estate, migrated with auto-waves.
Migration with ForgeAn Oracle procedure translated to BigQuery: the first deploy is refused because BigQuery has no SELECT … INTO, Forge repairs it from the error and deploys it in 1.1 seconds.
See it on your own data
We'll set up Qry against a dataset you choose and walk you through it — one hour, no slides.