Release date: June 2026

Datafor 9.00 is a major release. It rebuilds the AI Assistant on a new agent runtime, redesigns the report editor and the management console, introduces AI user licensing, adds the KPI Trend Card, and delivers a large set of query-engine performance and stability improvements.

Highlights

1. AI Assistant rebuilt on a new agent runtime

The conversational analysis preview introduced in 8.00 has been replaced by a new AI Assistant. It opens from the AI Agent card on Home and from the report editor toolbar.

What changes for users:

• Streaming answers with visible progress. Each stage of the analysis (understanding the question, matching it to the model, querying, concluding) is shown while it runs.
• Multi-turn conversations. Follow-up questions build on the previous result: "only East China", "break it down by month", "last year", "top 10", or "switch the metric". Suggested follow-up questions appear under each answer.
• Clarification instead of guessing. When a metric, filter value, or time field is ambiguous, the Assistant asks and offers candidates; the analysis resumes from your answer.
• Complex questions. A question that needs several queries is planned as a small set of steps. The result of every step is shown, and the conclusion is built only from those results.
• Model welcome card. After selecting an analysis model you see a model brief, the number of metrics and dimensions, and suggested questions. Quick Actions start guided analyses (metric check, trend, breakdown, comparison) without typing a question.
• Better results. Automatic chart recommendation, KPI cards for single-value results, stacked, funnel and line charts, percentage formatting, relative-time handling, and calculated-measure aware questions with shortcut calculation templates (growth, share, ratio).
• History. Conversations can be reopened and replayed.
• Language. Questions are understood in the user's language and answered in the same language.

For administrators, the new Common Questions page maintains the suggested questions of each model, and a Semantic maintenance workflow maintains business-friendly names, synonyms and descriptions of the dimension fields and business metrics the Assistant relies on.

AI Assistant · Common Questions · Business Semantics for AI

2. LLM management and metadata retrieval

• Redesigned LLM configuration. Multiple model profiles on OpenAI-compatible endpoints (including GPT-5.5 class models), a connection test, and default assignment of models to the agent's roles. Development and production use the same configuration model.
• Hybrid metadata retrieval. Model metadata is retrieved with vector search combined with business-term grounding, which reduces mismatches between the words in a question and the fields in the model.
• Vector index management. Indexes can be deleted and rebuilt from the AI Agent settings; embedding authentication failures are reported instead of failing silently.

LLM Configuration · Preparing Data for AI

3. AI user licensing

Licenses now carry an AI users capacity in addition to Creator and Reader seats; Settings → System → License shows the three capacities side by side. Administrators enable the AI Assistant per user with the Enable AI Agent switch in Users; the number of AI-enabled users is checked against the license, and AI features (assistant, insights) are only shown to enabled users. Multi-tenant deployments refresh the capacity per tenant.

Upgrade note: after upgrading, existing users are not AI-enabled. Enable the users who should have access to the AI Assistant.

4. Redesigned report editor

• New Style, Data and Components panels with consistent typography, faster rendering and single-column chart lists.
• Redesigned toolbar and unified dialogs (Save, Save As, confirmations).
• Snap to grid and alignment guides while dragging; cleaner selection outlines.
• Fit to page and automatic canvas fit to the window.
• Redesigned Actions → Interactions panel with a Linked components picker.
• The filter area can be resized by dragging.
• Components with auto-refresh show a reminder while the report is being edited.
• Interface translations updated for all supported languages.

5. Mobile layout designer

The mobile layout designer was rebuilt: a component list that shows where each desktop component lives, free drag and resize on a device-width canvas, collapsible panels, per-component style settings that apply only to the mobile layout, and improved root-canvas zoom and drag.

Mobile Layout View

6. New component: KPI Trend Card

A card that shows the current value of a KPI, compares it with a target (a target measure or a static target), colors the status by direction (higher or lower is better, evaluated against the target or the previous value), and draws a trend area over a time or category dimension. Marker value, unit, status icon and fonts are configurable.

7. New management console

The web console was rebuilt with a new Home, login page and navigation.

• Home offers quick actions (Create Report, Connect Data, Create Model, AI Agent) and Recent work, which now includes analysis models. Models can be marked as favorites.
• Settings uses secondary navigation; the JDBC drivers, OAuth2, CORS and white-label pages were redesigned.
• Users: new create/edit dialog, user import with attachment validation, export of selected users.
• Models, Datasource and other resource lists use inline actions, fixed action columns, consistent loading states and dialogs.
• Access control layout was reworked; delete and cache-clearing actions ask for confirmation.

8. SDK and embedding

• initialFilterValues sets the default values of filter components when a report is embedded (read-only/preview mode).
• URL default_ parameters can be injected into date pickers, date ranges and flat selectors; several SDK crashes were fixed.

SDK Embedding · Pass Parameters Through URL

Model designer

• Refreshed layout and attribute panel; a minimap for large canvases.
• The model tree scrolls to the field selected on the canvas.
• Time levels of attributes are recognized and used as defaults for the default time field of measures.
• Annotation names were normalized (Datafor prefixes removed, values trimmed), producing cleaner metadata for AI.
• Excel import: date formats are recognized correctly, and a type change that fails is reported as a warning instead of blocking the import.

Performance and stability

Query engine (Mondrian):

• Batched cell requests are deduplicated; pure-dimension table queries on multi-fact models are significantly faster.
• Relationship semantics (1:1, 1:n, n:1, m:n) are used when generating SQL. One-to-many and many-to-one joins no longer trigger whole-table GROUP BY subqueries; de-duplication is only applied to real many-to-many links.
• Fact Count is resolved from the measure group whose fact table is directly related to the dimensions on the axes.
• Memory: tuple cache retention is capped, native tuple materialization releases its references, and cell request keys are retained for a shorter time.
• Native aggregate and slicer constraints are preserved correctly (fixes wrong totals when aggregate requests are combined with filters); constant crossjoins get a fact anchor; calculated-member filter expansion is bounded.
• Time levels stored as epoch milliseconds are converted correctly; numeric native filters return a clear error message; schema relation loading is deterministic.
• A repeatable MDX regression suite guards these paths.

Platform:

• Metadata lists are cached per tenant user; license checks in fetchUser are faster; BouncyCastle is no longer required at runtime for license validation.
• Result cache: datafor.query.cache.result.enable=true no longer raises errors, and null values are no longer cached in dynamic queries.
• The simplified query API used by the AI Agent returns detailed error information and enforces row limits; annotations and measure expressions are exposed in metadata.

Important bug fixes

• Excel/CSV export of custom components no longer returns "no data"; time-axis selections are restored when a report is reopened.
• Custom blocks: spacing, position offset, missing selection state and abrupt loading fixed.
• Clustered column data labels, color mapping and overflow scrolling in panels fixed.
• Character-set handling for JavaScript resources fixed (server configuration).
• First-screen "error code" flash, wrong "fit to screen" target, toolbar overlap and slider thumb alignment fixed.
• Report title shown white-on-white in the top bar, stale white-label logo cache, and user export fixed.
• File dataset Excel import: date recognition and type handling fixed.
• Version check failures no longer flood the log.