Release 6.1

Tellius 6.1 is here! ✨ The latest update cuts the busywork that slows teams down: date hunting, filter guesswork, repeated follow-ups, and manual “prove it” steps. Kaiya is now context-aware, so your answers stay correct when the real world changes.

• Kaiya can now ground internal analytics with external “anchor” details like event dates, policy effective windows, and geographic scope.

• Kaiya keeps your intent and context consistent across an entire thread, and delivers usable outputs sooner so you can validate direction immediately.

• Voice Mode supports natural, multi-turn conversations.

• Improved short-term memory keeps follow-ups coherent across Deep Insights and standard analytics.

• Auto-routing selects the best execution path without manual mode switching.

Finally, results are easier to reuse and share. You can copy or export text, charts, and tables, export full conversations to Word, and export Vizpads to Excel/CSV with filter provenance. The result is fewer re-runs, fewer stakeholder back-and-forths, and a shorter path from “What happened?” to “Here’s what we’re doing next”, with insights that move cleanly into reports, docs, and exec updates.

New features 🚀

Kaiya Web Search: Grounds analytics with real-world context

Introducing web search augmentation in Kaiya to improve answers when a question depends on external, time-sensitive context. This capability is designed to ground and parameterize analytics, not to act as a general-purpose internet research assistant.

• Web search is used to fetch small but important “anchor” details (such as event date ranges, policy effective dates, geography/scope, and other timing windows)

• Kaiya then uses those anchors to run the right structured analysis in SQL/Python, for example by choosing the correct pre/post comparison windows, filtering to the right impacted regions, or aligning results to a policy start date.

• Web search is enabled by default and is triggered automatically only when Kaiya detects that the question needs fresh or external context to answer accurately.

• The anchors Kaiya finds can be kept in short-term working memory with a freshness-based Time-To-Live (high/medium/low), so follow-up questions in the same conversation use the same dates/scope instead of re-guessing them.

You can ask questions the way you naturally would (“after the hurricane,” “after the policy change,” “during the holiday week”), and Kaiya can still apply the correct filters and comparisons without you needing to look up dates and manually apply them. For example, you can ask “Did Thanksgiving impact TRx for Product X?” Kaiya can look up the holiday timing and apply a properly anchored comparison window in your internal TRx analysis.

Talk to Kaiya naturally with Voice Mode

Kaiya 6.1 introduces a voice-first conversational experience that goes beyond basic dictation. You can speak naturally, change direction mid-question, and Kaiya can still follow the intent across turns.

• Supports multi-turn voice conversations and handles natural speech patterns (pauses, corrections, fillers), not just transcription.

• Supports two voice options: GPT Realtime (preferred) and Deepgram (alternate option with multiple voices).

Voice mode is not enabled by default across deployments. Please contact your admin/CSM to enable voice mode in your environment.

Voice makes it easier to explore data when typing is inconvenient and supports faster “brain dump” style analysis. It reduces friction and supports on-the-go workflows, such as sales or field teams needing quick answers before a meeting. For example, a field user asks, “How is TRx trending in the Northeast?” then follows with, “Actually ignore that... Show the top districts this month”, without restarting the conversation.

Better follow-ups with improved short-term memory

6.1 improves how Kaiya keeps and uses context across multiple turns. It maintains better continuity even when switching between different internal capabilities, and it adds core infrastructure that enables richer memory behavior over time.

Short-term memory improvements

Kaiya now carries context more effectively across conversation turns, including across Deep Insights, so follow-ups work more naturally. Context across Deep Insights, Analytics, Metadata, and Summaries is stored together with tighter coupling.

Structured memory scaffolding (foundation)

Kaiya can store and retrieve memory objects using core memory types such as:

• Ontological memory: definitions, relationships, synonyms, formulas related to the data.

• Instructional memory: user or admin instructions/constraints (for example, formatting or business rules).

• Reference memory: references such as results from web searches or external lookups.

Kaiya is being built to pull the “right remembered info” in two ways:

• Always-on rules (deterministic overlays): Some saved context can be automatically included every time Kaiya runs in a specific situation (for example, when using a particular Business View).

• Smart lookup (semantic retrieval): Kaiya can also search its saved memory and bring in only what is relevant to your current question, even if the words are not an exact match.

When Kaiya uses web search, it can temporarily remember important “anchor details” it found (like dates, time windows, and locations) in short-term working memory. How long it keeps that information depends on how “fresh” it needs to be (Time-To-Live based on high/medium/low freshness). This helps Kaiya stay consistent when you ask follow-up questions in the same conversation thread.

Kaiya becomes better at handling follow-ups without repetition, maintaining continuity across different Kaiya capabilities, and delivering more coherent answers across longer multi-turn workflows. For example, after a Deep Insight explanation, you can switch back mid-thread and ask, “Now break that down by district and summarize the key shifts”, and Kaiya can continue using the same context.

No more mode switching: Kaiya auto-routes questions to the right analysis path

We’ve introduced an orchestration layer that automatically routes questions to the right internal capability based on the conversation context.

• Kaiya can route between default mode and Deep Insights mode automatically.

• Users no longer need to manually enable Deep Insights for applicable questions.

• The orchestrator also helps determine whether a question can be handled through SQL-only execution or requires SQL + Python-style processing.

Specifically, it routes the query to:

• SQL-only when the question can be answered by querying tables (filters, joins, aggregates, group-bys, etc.)

• SQL + Python-style processing when the question needs extra computation beyond SQL (e.g., statistical analysis, forecasting, clustering, custom business logic, advanced text handling, more complex transformations).

You can focus on asking questions naturally, while Kaiya handles the execution path selection and reduces mode-management friction.

Copy/Export Kaiya conversation results

Tellius 6.1 makes Kaiya outputs easy to reuse and share by adding copy + export actions across conversational responses, Deep Insights, and agentic workflow results. From the Kaiya chat interface, you can now grab exactly what you need (a single text block, chart, or table) or export the full result in one go, without rework or screenshots.

Copy to clipboard (per element):

• Text → copies as plain text (paste into docs, email, chat)

• Charts → copies as PNG image

• Tables → copies as PNG image

Export (per element):

• Charts → .png or .jpg or .pdf or .csv

• Tables → .csv or Excel

For complete sharing, Kaiya now supports “Export Result as doc” to generate a single editable Word document (.docx) containing the full result in the conversation (text, charts, and tables), with a clean, consistent layout for readability. You can move faster from “answer” to “artifact”: dropping Kaiya insights directly into stakeholder-ready docs, decks, and spreadsheets.

One-click Vizpad export to Excel or CSV ZIP

Tellius 6.1 introduces one-click export for Vizpads, so you can take analysis into Excel or share it externally without downloading charts one by one. You can export all chart data in a Vizpad as either:

• One Excel workbook (.xlsx) with one sheet per chart, or

• One ZIP file with one CSV per chart

Exports preserve the exact state of your analysis at export time, including global, tab-level, selective, and chart-level filters. Each sheet/CSV is named using the chart title, includes column headers. It also comes with a README that records filters applied, charts included, and any truncations or partial failures, making the export easy to audit and share.

To keep exports fast and manageable, admins can configure export guardrails under Settings → Application Settings → Vizpad → Excel Export, including:

• Maximum sheets in a workbook

• Maximum records per sheet

• Maximum columns per sheet

If a limit is reached, the export truncates remaining data and documents it in the README. Exports also respect existing permissions (including row-level security and masked fields).

Point-in-Time analysis: Use the latest value per time bucket

In 6.1, Business Views now include a Point in Time toggle in Advanced Settings. When enabled, time-based analysis in Vizpads treat each time bucket as an end-of-period snapshot: for weekly/monthly/quarterly/yearly groupings, Tellius uses the latest available data point inside that period, instead of treating the period as a full sum/rollup of everything that happened in it.

This applies to time-based analysis such as:

• Year-over-Year (YoY) and other time comparisons

• Top/Bottom and Top/Bottom For Each style analysis

• Supported across daily, weekly, monthly, quarterly, and yearly resolutions

This makes time-based results more predictable and consistent, especially when users switch between time resolutions, and gives admins a simple control to choose whether a Business View behaves like a snapshot over time versus a standard time aggregation.

For example, say, you track Active Customers that update daily. With Point in Time enabled, a monthly view shows the latest Active Customers value in each month (end-of-month), so YoY compares true month-end snapshots instead of mixing partial-month rollups.

Enhancements 📈

Creator-based filtering for superusers across Data assets

In 6.1, superusers can organize and find data assets faster with a new Creator/Author filter. Under the Data module, a Show datasets/BV author(s) dropdown is added for Datasets, Prepare and Business Views tabs that lists all creators and respects existing permissions (you only filter within what you’re allowed to see).

This lets superusers filter large, shared environments by who created an asset, making it much easier to locate the right dataset, dataflow, or Business View. It helps quickly answer “show me what this person built”, reduce time spent hunting for assets, and keep multi-user environments cleaner and easier to manage.

This dropdown will be visible only to superusers. The dropdown also Includes creators who are inactive/deleted (so older assets remain findable).

More resilient SAML SSO across more IdP setups

In 6.1, SAML single sign-on is more resilient across a broader range of real-world identity provider (IdP) configurations, reducing login friction for organizations with different SSO setups. A backward-compatible fallback option is also available via configuration to avoid user impact if an environment needs to temporarily revert while the IdP setup is reviewed.

• Broader IdP compatibility testing: Validated SAML across more security and assertion-delivery patterns used by IdPs, including cases where the SAML Response is signed, the Assertion is signed, both are signed, or neither is signed.

• More flexible validation handling: Improved tolerance for differences in how IdPs send signing and security information, reducing failures caused by strict interpretation of IdP variations.

• IdP-driven authentication remains intact: Tellius does not force a specific authentication method (password, MFA, etc.); your IdP remains the source of truth for how users authenticate.

• Flexible user identifiers: Supports multiple identifier formats (email, username, or custom IDs), instead of requiring one fixed format.

• Safety fallback via configuration: If an environment encounters issues with a specific IdP setup, an admin-configurable option can switch to the prior SAML implementation while troubleshooting is completed.

One-command conversation summaries that capture the full thread

Kaiya now comes with a summary capability designed to consolidate context across the entire conversation.

• Supports prompts like “Summarize the entire conversation.”

• Summaries can incorporate relevant context from Deep Insights, Analytics, earlier summaries, and metadata within the thread.

With this, you can convert long, multi-step exploration into a clean, shareable recap without manually stitching together multiple responses. For example, after a long investigation, you can ask Kaiya, “Summarize what we learned and list the top drivers”, and Kaiya provides a single consolidated narrative.

Automatic titles for charts, tables, and tracked metrics

In 6.1, Vizpad charts and metric tracked in Feed will be automatically named for you based on the columns and configuration you pick. As you build a chart or table, Tellius auto-generates a relevant title so it stays readable and share-ready.

unless you set your own title) it can stay in sync as you change the visual configuration in View/Edit.

If you manually name a chart/table, Tellius won’t overwrite it as you modify its configuration. However, auto-generated titles stay dynamic: it updates as you change the configuration and add/remove filters.

The auto-generated titles consider local filters (filters applied to that specific chart/table), and do not consider global filters.

It also applies to Feeds (Track a metric). Feed/metric names can be auto-generated using the key setup fields (for example: measure, aggregation, dimensions and time span).

Faster analytics with improved time-to-first-result

6.1 includes execution upgrades that reduce latency and make analytics feel more responsive during interactive exploration.

As soon as the first output is ready, Kaiya can render an initial usable result (typically a table) right away, while it continues the remaining work (such as chart generation and reflection/retries) afterward. This improves the “time to first usable answer,” especially for common analytical questions.

This reduces the waiting time before you can start validating the direction of the answer, spotting issues, or deciding your next follow-up question.

Reflection-based auto-correction for complex questions

In 6.1, we’ve strengthened reflection so Kaiya can check whether intermediate outputs match your question’s intent and automatically correct issues when needed.

After Kaiya generates SQL (and, when needed, intermediate tables/charts), a Reflection agent checks whether the results actually match what you asked for. This includes looking for common mismatches, such as:

• Incorrect grain (level of detail)

• Missing constraints or filters

• Partial coverage of the question

If Reflection agent finds a mismatch, Kaiya can adjust the execution plan and re-run the query (or the needed step) so the final output aligns more closely with the question intent.

As a result, you spend less time rephrasing prompts or troubleshooting, even when questions are complex or multi-part.

Advanced Pivot: roll-ups and grand totals at every level

In 6.1, Advanced Pivot tables in Vizpads now support “aggregate at all levels”, so totals and rollups show up not only at the lowest detail level, but also for collapsed or grouped levels in your row/column hierarchy. This includes grand totals (for example, totals by Region and overall totals) so you can read complete summaries without expanding every group.

What you’ll see now

• Rollups for each hierarchy level (e.g., Category → Sub-category → Priority, or Region → Product) even when higher levels are collapsed.

• Row totals, column totals, and grand totals reflect the current pivot structure and filters

• Works for common pivot measures like SUM (e.g., Sales/Units) and AVG (e.g., Avg Discount/Avg Profit), so intermediate totals are visible where users expect them

You get faster, cleaner reporting in pivots—teams can compare totals at the level they care about (Region, Product group, Category, etc.) without drilling into every layer or doing manual rollups outside the Vizpad.

For example, consider a pivot where rows are Furniture → Bookcases → High/Medium/Low, and the value is Average Discount. Now, even if you collapse “Bookcases”, you still see the average discount for Bookcases and for Furniture overall, plus the grand total, without opening every sub-row.

Smarter cascading filters for multi-fact Business Views

In 6.1, Vizpad filters behave more predictably for Business Views that use deduplication and span multiple fact tables.

• Vizpad no longer cascades filter values across unrelated columns when those columns come from different underlying datasets in the Business View.

• Vizpad still cascades filters when the columns are truly related, such as fields coming from the same source dataset (example: filtering Country can still narrow State values).

Your filter dropdowns stay populated with the right values, Vizpads using multiple facts are easier to use, while still keeping the helpful cascading behavior in places where it makes sense. For example, if your Vizpad has two filters: Region (from a geography table) and Brand (from a product table) and pulled data from multiple fact tables. In 6.1, selecting Region = East won’t wrongly shrink the Brand list, but filters like Country → State will still narrow correctly because they come from the same dataset.

Stronger monitoring and diagnostics for multi-step execution

Kaiya 6.1 adds deeper monitoring and diagnostic visibility into multi-step agent execution (including tracing the internal flow). This improves the team’s ability to identify and fix issues quickly, especially in complex multi-agent scenarios.

Faster troubleshooting translates into more stable behavior and quicker fixes for edge cases and complex queries over time.

Stronger security and compliance readiness

Continued strengthening security controls across the platform and remediated items identified through SAST scans. Improved overall security posture and compliance readiness across Tellius.

Kaiya feedback simplified (cleaner feedback loop)

Kaiya feedback is now thumbs-up / thumbs-down + text comments. The prior “edit and rerun” feedback flow is temporarily paused.