DQC

v3.13

Release 3.13 gives the Assistant its own page: one chat that works across your data sources, rulesets and Improvement workflows and makes changes after you approve them. The new Business Partner rule and node check and correct company master data against external sources. Databricks connects with a Microsoft Entra ID service principal.

🚀 New Features

Assistant

  • The new sidebar item "Assistant" opens a chat that is not tied to a single ruleset. It reads your data sources, tables, profiling, rulesets, rules, check results and Improvement workflows with your own permissions.

  • It can create rulesets and rules, change a ruleset's scope, start checks and profiling, manage tags, and build and run Improvement workflows. Every change waits for your approval: "Allow once", "Allow for this session" or "Don't allow". Rules created in a chat stay inactive until you activate them.

  • For a broad request such as "Which rules should this table have?" it proposes a plan first and builds the rules after you confirm.

  • Attach files with the "+" button, by pasting or by drag and drop: images (PNG, JPG, WebP) up to 5 MB and files such as CSV, XLSX, Parquet, JSON, PDF or DOCX up to 20 MB. A chat holds up to 10 files.

  • Chats are personal and keep running when you leave the page. They get a title automatically and can be renamed, deleted and searched.

  • Personal data in table values and in text or table attachments is masked before it reaches the AI model. Images and scanned PDFs are sent as they are.

Business Partner validation and improvement

  • The new rule type "validates business partner data" checks company name, address, VAT ID, website, email and phone against external company and address data. You map the company name, city and country columns and choose the fields to check.

  • A row is validated only when exactly one company is identified reliably. All other rows are listed as "Not validated" with a reason such as "Ambiguous match" or "No reliable match", and can be downloaded.

  • An issue shows the existing value, the found value and the matching details. A rule with rows that were not validated has an unknown quality and is excluded from the ruleset grade.

  • The new Improvement node "Business partner" proposes corrections for the fields you select. Original values are kept: each field gets columns for the proposed value, its confidence and its source.

  • Rule and node consume token credits. The token credit estimate of the Preview node covers the node.

Databricks with Microsoft Entra ID

  • The Databricks connector has a third authentication method, "Microsoft Entra ID", for a service principal with tenant ID, client ID and client secret.

  • "Client ID/Secret" is now called "Databricks OAuth secret". Existing connections stay as they are.

✨ Improvements

Assistant in rulesets and Improvement workflows

  • Both chats use the same engine as the Assistant page: a reply keeps running when you leave the page, and Stop cancels it on the server.

  • After your approval, the ruleset Assistant creates, changes, activates and deletes rules of its ruleset, assigns tags, changes the column scope and starts a check.

  • The Assistant picks rule types from a catalogue of all built-in types and uses Custom SQL or Python only when no built-in type fits or you ask for it. Cross-table, format, freshness and volume rules can now be set up from the chat.

  • Long chats load the latest 40 messages first, with "Load older messages". The message field has voice input, and answers can be copied.

Connector errors that name the fix

  • Errors when saving a data source or loading its tables stay visible in the form and link to the troubleshooting guide of that connector. The "Failed" badge on data sources and tables links there too.

  • Connection errors for PostgreSQL, Redshift, Oracle, ClickHouse, Impala, Snowflake, BigQuery and MotherDuck end with what to check, for example the password, the host and port or the firewall.

  • Azure SQL, Azure Synapse and Microsoft Fabric report login, database, firewall, timeout and TLS problems separately instead of one generic error.

MCP server

  • New tools manage tags.

  • New tools delete rules, rulesets, Improvement workflows, workflow files, tables and uploaded file data sources. The existing protections apply, for example for the last ruleset of a table.

  • Tools that return columns page them to speed up delivery.

🤖 Small improvements and bug fixes

  • A preview or run with an empty result completes instead of failing with "No improvements result was found!".

  • Duplicate survivor nodes on large tables finish.

  • The Preview node keeps the highlight of added columns after reopening.

  • Python rules and nodes can use iter, next, map, filter and further built-ins that failed with "name is not defined".

  • Excel uploads with double quotes in a column header work.

  • The address rule restricts the lookup to the country of a mapped "Country code" column and offers six more address fields, among them "Full address" and "District".

  • If the account settings fail to load, the app shows "We couldn't load your workspace" with "Try again" instead of loading endlessly. Dashboard figures, lists and breadcrumbs update right after a change.

  • The data sources table has a "Last updated" column, and date fields accept typed input.

  • The notification log on the mission detail page is reachable on large screens.

💡 Behaviour changes worth knowing

  • Two overlapping saves of the same rule no longer overwrite each other. The later one is rejected with a request to reload.

⚡ Breaking Changes

  • Custom Python needs a top-level def f(df) or def f(df, data) as its entry point. f = lambda, *args and **kwargs are rejected, also in existing rules and nodes.

  • Tag API: creating, changing and deleting tags requires the "Manage tags" permission, and reading or setting the tags of a rule requires access to its ruleset. Users and API keys without these permissions receive 403.

☁️ Private Cloud

Release candidate image tags for platform version 3.13.0:

  • qui: rc_1c5faadc328571892d4fe93da28915c837f1c727

  • qservices: rc_amd_1c5faadc328571892d4fe93da28915c837f1c727

  • python-executor: rc_amd_1c5faadc328571892d4fe93da28915c837f1c727

  • dqc-mcp (AMD64, required for the Assistant): rc_amd_1c5faadc328571892d4fe93da28915c837f1c727

Platform version: 3.13.0