Skip to main content
Latest release:

Data teams are playing a crucial role in governing the quality of AI responses, just like they did with trusted dashboards in the pre-AI era.

This week’s updates include tools for understanding agent behavior, a cleaner entry point into self-serve analysis on top of semantic models, and more improvements to Threads!

👀 Agent observability in the new Context Studio

We’re bringing observability and context management together in the new Context Studio — giving you the tools to manage what data is trustworthy and how agents should use it.

On Team and Enterprise plans, Admins and Managers can open up the dashboard in the Context Studio to understand how agents are being used, what users are asking, and where performance excels or needs improvement.

With this view, you can:

  • Track conversation volume
  • Identify your most active users
  • Spot patterns in user feedback
  • Filter by agent type, workspace role, time range, and more

Admins can also dive into individual conversations to:

  • Review the agent's reasoning
  • See what context it referenced
  • Diagnose where improvements are needed

Managers get access to aggregated metrics but not individual conversations.

When you're ready to make adjustments to the agent’s behavior, you can hop over to the Context sources tab, where we’ve unified settings for workspace rules, data endorsements, and semantic projects in one place.

There is so, so, so much more coming soon here — so share your feedback and stay tuned.

🪟 Views on semantic models

We’ve introduced views as a new layer on top of semantic models, designed for clarity and readability.

As your data models grow, you can create views in the Modeling Workbench to rename and reorganize dimensions and measures — giving business users a cleaner, more intuitive starting point for self-serve analysis.

Views can help slim down wide tables, simplify complex join trees, or even flatten relations. Now you can scale the complexity of your underlying models, then surface only curated views that are easy for end users to explore and understand.

📁 CSV uploads to Threads

You can now upload CSVs to Threads, and the Hex Agent can craft queries against the data in the file provided. The agent is also capable of joining the CSV data with existing warehouse data, enabling cross-source insights from a simple file upload.

This is available to users with the Explorer role or higher who have permission to access non-semantic data in Threads. Admins can configure agent data access and data connection defaults in the Threads settings page.

🧵 More Threads improvements

We’ve made a bunch of additional improvements to Threads!

You can now copy and download charts, pivots, and tables in Threads. Want to add a chart from Threads to your presentation slides? Or import the data in a table into a spreadsheet? You can copy results to your clipboard, or download them as images or CSV files.

When a chart or pivot is built directly on a semantic model, we’ll tag the name of the model directly in the footer of the visualization. This provides a standardized way to check the data source at a glance.

Chart on a semantic model

We’ve also made Threads faster at answering simple quick questions. When you ask a question, the agent will gauge the complexity; if it’s deemed an “easy” question, the agent will spend less time thinking and return a quicker response.

Give these improvements a spin by trying our new Threads browser shortcuts! hex.new has always been the shortcut for creating a new project in Hex, and now you can enter threads.new in your browser’s address bar to create a new thread. For an even quicker shortcut, use threads.new?prompt= followed by your question.

📐 User-level agent instructions

If you want to steer the agent to behave in certain ways — whether it’s the SQL style or just the agent’s tone — you can now add user-level rules files to customize the styling and preferences of the agent’s responses.

From the custom instructions modal, you can also see the agent workspace rules that Admins have added for your organization.

Other improvements

  • Better bulk column editing experience: The Columns menu in the lower right corner of SQL and table cells got a UX refresh and new capabilities! You can now reorder, hide, and pin columns from one tab, and bulk format them in a separate tab.
  • Chart upgrades by the Notebook Agent: If you use the Notebook Agent to modify a legacy chart cell, the agent will automatically upgrade the chart cell on your behalf in order to make the changes you ask for.
  • Start Threads from workspace search: You can use the keyboard shortcut Cmd+K to open workspace search, type in your question, and start a thread from there.

We’re always working to make Hex better — and that includes investing tender loving care into core workflows like building viz, caching, and publishing apps. This week’s updates are focused on quality-of-life improvements that make it easier to customize charts, get fresh results, and understand performance.

⏱️ New app run and caching UX

Our new and improved app run settings are now part of the publishing workflow and make it easy to keep apps up to date.

Most of the time, you want apps to reflect the latest data — and now they will. For new projects, the default is to auto-refresh published results: when someone clicks the refresh button, it updates the results for everyone, not just the person who ran it.

Additionally, if results are stale when you open a published app, Hex will kick off a refresh in the background. This means less time waiting for old results to re-run!

For existing apps, Editors can turn on auto-refresh or update results for everyone right from the published app. If you prefer a fixed snapshot instead, you can always restore a previous run from the run log and turn off auto-refresh behavior.

Other improvements include:

  • A simpler, clearer app run settings menu
  • More consistent display of data age on published apps
  • Better handling for errored background runs
  • New estimated run time information

Want to dig in? Read the full rundown in our docs.

📊 Chart cell upgrade

You may have noticed that chart cells recently got an upgrade! All newly-created chart cells come with an updated UI and new capabilities.

Behind the scenes, we've converged our chart cells with Explore, so the experiences are unified across both interfaces. You can now access powerful features in chart cells that were previously exclusive to Explore, including:

  • Semantic models support: Chart cells now accept semantic models as a source, allowing you to build visualizations using governed dimensions, measures, and joins — right in your notebook.
  • Spreadsheet-style calculations: Define dimensions and aggregate measures directly in chart cells. Aggregative measures (e.g., weighted average, ratio of sums, or any custom aggregation) are computed dynamically based on how your chart data is grouped.
  • Top N: It’s now possible to specify the top (or bottom!) n values for a base-axis, color, or facet directly in a chart cell, cutting down the number of steps required to produce these types of visualizations.
  • Custom start-of-week settings: Choose Sunday or Monday as the start of the week for charts and pivots. No upstream data wrangling required!
  • New and improved config panel: We cleaned up the UI and added clickable pills to change aggregate types and date truncations.
  • Drag and drop fields: You can now drag and drop fields into the config panel, or directly into the visualization empty state.
  • Responsive layout: The config and data sidebars now auto-collapse based on screen width, so you can keep the Notebook Agent open without crowding the chart editing experience.
  • And more here on drilling, viewing underlying data, and filtering on aggregates and measures!

Chart cells created before this update will continue to run, but won’t include the new features listed above. To upgrade, just hover over the “Legacy” tag in the top right corner of an existing chart cell and click “Upgrade.”

🏃 Improved run stats

We’ve made it easier to understand and troubleshoot performance with better run stats. You can now access reliable information about completed runs in your current session (from the last 48 hours) from the Help menu in the notebook view, or from the Debug View of an app run.

Improved run stats

For cells that send queries to the warehouse (e.g., SQL cells and chart or pivot cells using query mode objects), we now show additional details including:

  • Whether the query hit the cache
  • Time spent querying vs. streaming data
  • How much data was streamed

Other improvements

  • Hex Agent in Slack for multi-tenant EU stacks: All multi-tenant customers (including those in the EU) can now install the Hex app in Slack, mention @Hex in any channel, and get answers directly in-thread.
  • Stop scheduled runs: It’s now possible to kill actively running scheduled runs from a project’s run log!
  • Smarter cell awareness for Notebook Agent: The agent now has context on which cells are published in your app, so you can say things like “review all cells in published app.” The agent also knows which cells you’re focused on or have selected. So when you ask, "Can you move the chart you just created under this cell?" it knows exactly what you mean — no @ mentions needed.
  • Notebook Agent in Add Cell Bar & Shortcuts: It’s easier than ever to access the agent when you need it, right from the Add Cell Bar. You can also use keyboard shortcuts: Cmd+G for accessing the agent from the cell bar and Cmd+I for opening the agent sidebar.
  • Customize Top N rankings: Instead of being limited to the existing aggregate or measure plotted in your visualization, you can now rank top N by any aggregate or measure.
  • New GPU compute profiles: For compute-intensive workflows, we now offer A10G and L4 GPUs — bringing newer architectures, more GPU memory, and better cost-effectiveness than V100s.
  • Delete Threads: Clean up your workspace with the option to trash threads from the homepage (via the three-dot menu) or in bulk from the Threads tab. They’ll be permanently deleted after 30 days.
  • Clear all chart configs: Sometimes, starting over is the move. Look for the circular arrow icon in the chart config panel to clear all fields and formatting in one go.
  • Click to fix “too many data points”: The “too many data points” warning now offers actionable solutions: click buttons to change the time grain, apply an aggregation, or use top N to reduce the number of data points plotted.
  • Bug fix for chart loading state: Charts used to show an error icon next to columns while in a loading state. This has been fixed.
  • Better query mode performance: SQL cells that return in query mode now run in parallel, leading to faster overall run time.
  • @ mention dataframes improvement: You can now @ mention dataframes in the Notebook Agent even if they’re not in kernel memory. This is handy when you revisit a project and haven’t rerun all the cells!

🤖 Hex Agent in Slack & MCP

Introducing two new ways to tap into Hex's advanced agentic capabilities: a native Slack integration and Model Context Protocol (MCP) support.

We’re bringing trusted context and answers from the Hex Agent directly to where you're already working.

Ask a question by mentioning @Hex in Slack, and the agent gets to work:

  • 🔍 Searching your warehouse and semantic models
  • 🧠 Writing SQL and analyzing results
  • 📊 Responding with any relevant charts and links

Want to get started? Check out our Slack integration docs and MCP server docs.

Read our full announcement to learn more.

Note: The Hex Agent in Slack is available to Explorer+ seats on Team and Enterprise plans for US multi-tenant stacks. Admins can install the Slack integration for their workspace. The MCP server is available to Explorer+ seats on Team and Enterprise plans. Extended AI usage limits are included for free during the early access period.

We’ve shipped a lot of big features lately: the Notebook Agent, Threads, the Modeling Workbench, Explore capabilities, and more. This week’s updates are all about enhancing these agentic workflows in Hex, and doubling down on the semantic context and curation tools needed to make AI accurate and useful in your analysis.

🧰 Fix with agent

Debugging code and SQL is tedious, especially when you’re working with a new, unfamiliar dataset. The Notebook Agent can now help!

From an errored Python or SQL cell, you can now Fix with agent. Instead of being limited to looking at one cell and only editing the contents in that cell, the Notebook Agent has full project context and can view/edit upstream cells and compiled SQL.

“Quick fix” is still great for simple syntax errors that are constrained to a single cell, but “Fix with agent” can handle more complex errors that may require changes across multiple cells. You might even find that its explanations help level up your SQL and Python skills!

🤖 Notebook Agent improvements

We’ve been very busy making our Notebook Agent smarter, faster, and more efficient in Hex. Here’s everything we’ve shipped lately:

  • Powered by Claude Sonnet 4.5: The Notebook Agent and Threads are now running on Anthropic's latest model, with significant improvements in speed, capability, and complex task handling for analytics tasks.
  • Using Hex docs: The Notebook Agent and Modeling Agent can now leverage all the technical documentation on our Learn site in its answers, and point you to relevant tutorials and templates. The next time you have a question about Hex’s features, just ask the agent!
  • Implemented Notebook DAG context awareness, so the agent can make more intelligent, dependency-aware decisions about editing
  • Better at handling long-running cells that do heavy queries and analysis
  • More careful about editing vs. creating new cells for a tidier notebook
  • Better at telling you if it doesn’t have a tool to fulfill a request — like input cells, moving cells or deleting cells. (But good news! These are all tools that are coming 🔜)
  • Radically improved ability to write and edit jinja
  • Agent can now run for twice as long without checking in, and is also much more reliable about stopping early and not wasting time
  • Better at recognizing when a user has a quick question vs. a vague problem that needs alignment, in which case it’s better at checking in to coordinate on a plan rather than just diving headlong into doing things
  • Prefers chained SQL to write clean, easier to read SQL
  • Improved usage of Hex charts for nicer interactive visualizations
  • Generally improved response tone to be more concise, higher signal and less fluffy
  • Reduced creation of markdown or Python summary cells, unless explicitly requested
  • Better understanding of components and sections, and correctly placing cells in the notebook

The agent should feel much more intelligent with all these improvements! We’d love to continue to hear your feedback on all this — just drop us a note.

🧵 Threads improvements

Earlier this month, we launched Threads — a conversational interface for everyone to answer questions with data — and the feedback has been awesome to see!

User feedback on Threads

Since then, we’ve made some improvements to make the Threads experience even more delightful:

  • Readiness indicator in browser tabs: Let Threads work in the background while you grab a coffee or respond to an email. When the agent is finished thinking, you’ll see an indicator in the title of your browser tab, so you can jump back in when results are ready.
  • Duplicating threads: Pick up where your coworker’s conversation left off by making a copy of a thread that has been shared with you. The prompt history and previous results will remain upstream for you to reference.
  • @-mention data sources: When you want the Threads agent to use a particular semantic model or warehouse table to answer your question, you can specify a data source with an @ in your prompt.

Please keep the feedback coming, and let us know what you’d like to see next!

🧱 Databricks Metric Views integration

Last year, we introduced Semantic Sync in Hex, enabling teams to turn on trusted self-serve via pre-existing semantic models.

We’ve now added support for Databricks Unity Catalog Metric Views — in addition to syncing with dbt MetricFlow, Cube, and Snowflake Semantic Views. You can now:

  • Browse metric views directly from your Databricks connection in Hex.
  • Query with SQL in the notebook, without redefining metrics.
  • Build data apps that are backed by governed metrics, ensuring consistency from exploring to publishing.

Whether you choose to sync external models or build them natively in Hex’s modeling workbench, these trusted measures, dimensions, and business logic allow the Notebook Agent and Threads to generate more consistent, accurate results.

Read more in our announcement of Hex’s Databricks Metric Views integration.

🔝 Top N and more Explore improvements

Explore continues to grow even more powerful and feature-rich as we add more capabilities for digging into data and building visualizations. The latest addition to Explore’s toolkit is Top N, which helps you answer questions like “Who are our top 10 revenue-generating customers?” without requiring a line of code.

Top N allows you to specify the top (or bottom!) n values for a dimension in a chart or pivot, and choose whether the “other” values are bucketed in the viz.

This is only one of many improvements to Explore over the last six months, so we compiled a round-up of what’s new in Explore (e.g., totals in pivots, aggregate calcs, drilling, and more)!

🧭 New navigation sidebar

With how much the Hex platform has changed and grown over the last year, it was time to refresh our sidebar for consistent, intuitive navigation!

Lean on the nav sidebar as your go-to hub for creating and engaging with assets:

  • Start a new project or thread
  • Jump back into recent work, and find favorites
  • Search your workspace
  • Respond to notifications (all in one place!)
  • Access Settings or Support

It's available from any page in Hex, so folks who are new to your workspace will have an easier time finding their way around. (And our new homepage is a great jumping-off point with a big ol’ prompt bar for asking the agent a question!)

Other improvements

  • Subtotals and sort by values in pivots: In addition to totals, it’s now possible to add subtotals to pivots that have multiple rows or columns. You can now sort by aggregated values in pivot tables.
  • SSH tunneling for Databricks: Our Databricks data connector now supports SSH tunneling for secure connections to databases behind firewalls or in private networks.
  • Custom data retention periods: Admins can now tailor data retention timelines to meet stricter compliance requirements and internal policies. This custom data retention is available for cell outputs and cached warehouse data of 7, 14, 30, 60, or 90 days.
  • Version history in Modeling Workbench: Now you can see a history of changes made to a semantic model and who made those changes. This includes checkpoints created before the Modeling Agent makes a change.
  • Project filters can reference semantic model columns: Project filters are now hooked up to semantic models, so you can build a fully no-code, interactive app based on modeled data.
  • Support for semi-additive measures: Hex's semantic modeling now supports measures on snapshot tables. These semi-additive measures allow you to control which dimensions measures can be aggregated over. They automatically subselect the correct rows, such as only using the latest month of data.
  • Manager role can edit semantic projects: After an Admin creates a semantic project, Managers can now edit them — broadening the pool of folks who can work on semantic models without giving out full Admin access.

🧵 Fall Launch: Threads & more

Introducing Threads — a conversational interface for analytics, made for everyone who wants to answer questions with data. It’s integrated with your team’s existing data workflows, and has access to powerful tools anchored in trusted context.

Today is our biggest release ever, and by far the coolest stuff we’ve built — including new semantic modeling features, Notebook Agent improvements, better no-code viz in Explore, and more!

The highlight? Threads helps you unlock self-serve that actually works:

  • ❓ Ask questions in natural language
  • ⚓ Ground answers in context curated by the data team
  • 💡 Get deeper insights from Hex’s advanced, built-in tools
  • 🛡️ Keep the data team in control of auditing and curation

Threads is available to Explorers+ in public beta on Team and Enterprise plans. Extended AI usage limits are included for free during the early access period.

Ready to dive in? Check out our Threads quickstart guide.

Read our full announcement to learn more.

🤖 The Notebook Agent

Introducing the Notebook Agent — bringing advanced agentic capabilities directly to where data teams are already working. Ask a question, get to a first draft in minutes, and iterate with precision.

The agent can help you do things like:

  • 🗺️ Plan and refine where you want to take your analysis
  • 🔍 Search your data to find the right table
  • ⛓️ Create and chain cells using SQL and Python
  • 📊 Visualize your data with native pivot and chart cells
  • 💡 Summarize and synthesize your findings in Markdown cells

The Notebook Agent is available to Editors+ in public beta on all paid plans. Extended usage limits are included for free during the early access period.

Ready to dive in? Check out our Notebook Agent docs and best practices guide!

Read our full announcement to learn more.