Changelog

Hex is already the best tool in the world for teams to go from question to answer with data. But sometimes telling your data story is about what happens after you have an answer. How do you put together an engaging story with powerful visuals? How do you combine insights into a narrative that gets people to take action?

We’ve been hard at work on this. Today, we’re announcing improvements to the Hex Agent that allows you to control more of the story: a full prompt → dashboard workflow, better visualizations, and more improvements to our agents.

Check out how we’re expanding the scope of what you can do with AI & data.

🏎️ Prompt → Dashboard​

You now can create a Hex app via our agent with one prompt. Just tell the Hex Agent what you want to build: it can view and create app layouts, create tabs + and edit them.

This takes out the manual work behind building dashboards in Hex.

We’re obsessing over how to enhance the visual stories you can tell in Hex. Stay tuned, we have lots coming soon.

📈 Better visualizations​

The Hex agent can now build more beautiful charts, including dual axis and charts with reference lines. This is on top of our previous work to build a subagent to ensure higher quality charts. Before the subagent, the main Hex agent handled visualization along with everything else and was often eager to move on with the overall analysis, leaving a trail of subpar charts in its wake. Today, the subagent works diligently to refine visuals and present a clean, precise chart.

Want to learn more about how we did it? Read about it in Madeleine Filloux’s blog.

🛠️ Other improvements​

• Add cells to chat with a click - from any cell in a published app you can now click Add to chat and start asking questions in the published app about it! This makes it easy to tell the agent to focus on a specific cell that interests you even if you don’t know the cell’s name to @ mention it.

• Control which data connections are available via MCP - Admins now more flexibility into which data connections the Hex Agent can use with external integrations. This is especially helpful when you have data you don’t want leaving Hex, like PII data, but still want to enable MCP or Slack integration. Each data connection can be toggled on or off for external integrations. When a data connection is excluded from MCP access, the agent will inform users that it can’t fully display the data, and link the user back into Hex to view it.
• New running UI in the notebook - we’ve added a new UI which shows which cells are running. Just hover over the project run button when your notebook is running. You can click a long-running cell to get scrolled to the offending part of your project.
• Kernel Management for Admins - Admins and Editors can now view, and stop, actively running kernels. They can stop individual kernels, or all of them at once - Admins can manage all active kernels in the workspace, and Editors can manage their own kernels. This is helpful when you have critical projects that need to run (including scheduled runs) but are bumping up against kernel limits.

🕵🏼‍♀️ Introducing Context Suggestions​

Every day we see how Hex changes the way people use data – and how good context creates trusted answers. Building your context up can be a lot of work - but it's even harder to know if it's working well and, crucially, how it can be improved.

And we want to make setting up, observing, and improving your context easy.

Today we’re announcing Context Suggestions in Hex. Suggestions automate context curation, turning Hex into a system that gets smarter the more you use it.

🗣️ Working with Context Suggestions​

Behind every thread, Hex’s Review Agent analyzes the interaction and creates “Warnings” where a lack of context is potentially leading to confusing, incorrect, or incomplete answers.

Context Suggestions are then created based on clusters of those warnings, surfaced in a feed that the data team can review and implement.

Here’s an example of what you might see in the suggestions feed. The agent will draft an update to our guides and we can review the changes. Just click “Publish,” or make your own edits.

This opens up a ton of possibilities for data teams to increase their impact and roll out AI to more people without sacrificing trust.

It's all surfaced in the Hex Context Studio - the team can access suggestions, see why the Review Agent is making its suggestion, and take action right then and there.

Check out our help docs for more on suggestions.

📚 Control suggestions programmatically with the CLI​

Suggestions are also surface-able in the Hex CLI so you can incorporate them into your existing workflows for metadata, semantic models, and analytics.

You can list all suggestions and get the details of specific suggestions. Each suggestion also comes with a handy string you can paste into the terminal and easily work it into any scripts you want to run to update guides in external repos or warehouse metadata.

♻️ Roll out AI Analytics to your team and continuously improve it​

Our vision for Suggestions is to turn context into a self-learning system. Context is constantly flowing and evolving - with every new question, it changes a little bit. Your platform should evolve with it.

If you’re getting started with AI Analytics, our advice for you is: just get started! Try out Hex on your data and see where you get great answers and where you need to build a little more. Don’t try and model your entire warehouse before you start asking questions - by the time you’re done, your users will have a completely different set of questions!

🛠️ Other Improvements​

• Run stats improvements - We’ve made some updates to run stats. Run stats now show up in a panel where you can see them side by side with your notebook, which is more useful. You can click on a step in your project run and see all the execution steps and how long it took, which is helpful for identifying performance improvements. Editors can also now see runs for all users, rather than just runs in the current session.

If you’re following AI Analytics you’re probably hearing a lot about context. Context is the rules and guardrails for AI. It’s how we ensure that all the work that AI is doing to answer your data questions is correct, consistent, and relevant to what you need.

But context isn’t a monolith, it’s a heterogenous puzzle. It’s tempting to believe we can build all the definitions we need first and then have a great experience with AI. But every time we interact with data we’re creating more context! It’s a moving target.

Today, we’re announcing a boatload of improvements to help you handle context.

🧠 Automatic User Memory​

User memory is now available in the Hex Agent, which means you should have higher-quality, more stateful responses - that feel personalized to you.

The agent builds up a profile about you to help have more tailored, accurate answers. It also has the ability to go reference recent conversations you've had with the agent. This means you can ask questions like, "What do you think is most useful for me to be paying attention to?" and the agent will be able to suggest interesting insights for you to dig into.

💬 Prompt the Hex Agent from your data apps​

Data apps (née “dashboards”) are where a lot of data insights are shared in reusable, interactive reports. We’ve upgraded the Hex Agent’s capabilities here.

From a data app, you can start a thread with the Hex agent. Before, your answers were strictly limited to what was in the app. Now, if the answer to your question isn’t there, the agent can use all of its context (like queries, charts, and the logical flow the analyst went through) to answer, creating new cells and chart as needed. This way, the app is a valuable set of starting context to help guide your questions, but it doesn’t limit you.

This also provides a great AI entry point for people who are used to getting their insights in dashboards: they can start with a visual report, and use the agent to ask follow-up questions.

Check out our docs for a full overview of what you can do when chatting with your apps.

🧬 Cells from projects-as-context​

Sometimes the best answers are from assets that already exist. Now, when the agent points to a project in Threads, it can point directly to any cell in the project (like a chart or a table) and render it in-line. This makes it faster and easier to look at existing context rather than always creating new content. It’s particularly helpful when there’s a trusted metric you want the agent to reference, and you want to view it in-line.

⚠️ Previews and warnings in GitHub enable programmatic context​

Guides are a great way to give the Hex Agent some rules and regulations for how to analyze your data - from sample queries to important terms. Many data teams are taking on context curation as a core responsibility, and being able to manage that context at scale is crucial.

Now, you can use our GitHub action to automatically sync guides from GitHub to Hex. This allows you to control context for the Hex Agent from outside of Hex.

Now, when you push changes to your guides, the GitHub action will run validation checks and return warnings. A preview link allows you to test guide changes in Hex before you merge them, and you’ll get a GitHub comment summarizing the changes.

Manage the context for your agents programmatically and test context updates before you merge them into production! Managing agents just got more…manageable.

📜 Guides Version History​

A new history sidebar in the Context Workbench shows a timeline of all published versions. Click any version to view its changes and full contents. This works for guides controlled both inside of Hex and externally in a GitHub repo or other source. You can see what the agent was working with at any point in time.

🔍 Semantic Model References​

You can now “View References” on a semantic model to see every project and cell that uses it. This is especially useful when you need to deprecate or remove an old model and want confidence it won't break any projects or dashboards.

🛠️ Other improvements​

• More details in context onboarding - For those of you who are still learning about context: we’ve added some more details to help you get started in Context Studio. This inescapable learning journey will link you deeper into the places in where you can build and improve your agents. You will learn about context and you will like it.
• New API endpoint for programmatically fetching images from an app - Get the rendered PNG image of a chart cell from a completed dashboard run - making it super easy to programmatically pull updated chart images you care about and add them to a doc or slide deck. See the documentation here.
• Include data in your enterprise search with Glean integration - If you’re using Glean to unify your apps into one searchable workspace, you can now add Hex into the mix. Using our MCP connector, Glean can now access Hex Projects and Threads - you can create a new Thread, continue an existing one, or pull in charts, descriptions, summaries, and direct links to Hex content.
• Better Slack + Hex Agent interactions - The Slack + Hex integration now handles more response types in Slack, including plaintext outputs, Python dataframes, and Python charts. Plus we fixed some bugs that were causing messages to be struck through or not correctly responded to. If you’re seeing any issues please let us know!

🕵🏼‍♀️ Introducing the Hex CLI​

Today, we’re announcing the Hex Command Line Interface. Using the CLI, users (and agents like Claude Code) can create projects, write SQL, run analyses, and connect their agents…without leaving the terminal.

👫 Some ways we’ve seen our customers use the CLI​

It’s been exciting to hear the use cases customers are envisioning for the CLI during our beta period:

🤖 Generate locally, ship to Hex - Sometimes, Claude knows just a little bit more about a topic than Hex does, for example, those changes you’re working on in your dbt project right this minute. With the CLI, use any coding tool to write SQL and Python and then push the finished analysis to Hex as a shareable, versioned project.

🤩 Bulk Schema Updates - when a column is renamed, you used to be left with the hefty job of hunting down every affected dashboard and fixing them one by one. Now, list all projects, scan all SQL cells for the old column name, update, and rerun to validate. From a day-long project to a simple script.

🚂 BI Migration 😉 - Do you have a bunch of legacy BI reports that you want to move to an AI native platform but you’re overwhelmed by the project? Create a new Hex project in one command (or create a script to create a whole bunch of Hex projects). Fetch the SQL from your old BI tool, and migrate it to Hex, no clipboard required.

If you’re ready to get started, just paste the following in your terminal (or check out the documentation here)

curl installer: curl -fsSL https://hex.tech/install.sh | bash
homebrew installer: brew install hex-inc/hex-cli/hex

Data is most powerful when the whole org can use it, and when every person who uses it makes it better for the next. Every query, every metric, every insight should feed back into the platform and make the next person's work faster, more accurate, and easier to trust.

We've been calling the stuff that usually slips through the cracks context exhaust: the query you ran to get there, the project someone built six months ago that answered this exact question, the metric definition your team actually agrees on. It just kind of evaporates. Today's release (and basically everything else we’re doing) is about capturing it and putting it to work.

We're also making the agent itself more capable: specialized subagents that handle complexity in parallel, Python in Threads, smarter data discovery. The kind of thing that means your best analyst's instincts aren't locked behind a queue.

The more your team uses Hex, the better it gets. Here's what's new.

📽️ Project context for the Hex Agent​

The agent can now deeply reference other projects. This helps you take advantage of the work other analysts have done, ensuring trusted answers and consistent metric definitions.

• In Threads - The agent can now pull queries, charts, and code from your projects into a thread.
• In a Notebook - Start with an existing project and use it as a jumping off point for a deeper investigation.
• In Workspace context - Projects can now be referenced in your workspace context and guides as examples the agent should review before giving an answer.

Simply @ mention the project you want to refer to (or paste in a link) if you have a specific reference in mind. You can also tell the agent to search for all projects related to [a topic that interests you].

If you’re creating projects in Hex already, you’ve already built a ton of context that can direct our agents. Be sure users have "Can explore" access to important projects in order for them to be pulled in as context in Threads.

For more details on how this works, check out our blog.

👨‍💼 Threads Agent can use Python​

Previously, when you were in Threads, the Hex Agent could only create SQL cells and charts. We’ve given Threads the ability to write Python! This opens up the possibility of all sorts of new conversations in Threads, like asking the agent to create a forecast of a metric that interests you, bringing in sentiment libraries, and letting the data visualization subagent create more customizable charts. Since Threads can now bring in cells from other projects, you can even take advantage of code written by your data science team and run it in your Thread.

🔎 Focused Cell Viewer in Threads​

We’ve built a new cell-viewer in Threads. As Threads generates cells, you can click into them and see the results. There’s a nice, clean UX for flipping through charts and code cells and a nice big viewer for your pretty charts. While in focus view, you can link anyone to a cell directly. When the agent is done, you’re left with a simple, browsable summary of all the steps the agent went through and outputs it made.

Wouldn’t it be nice if you could publish this summary as a shareable app of some kind? 🤔 Stay tuned.

↔ Agent can swap data connections​

Previously in Threads, the user had to select a data connection before sending a message and the agent was restricted to that connection. Now, when you’re asking the Hex agent for answers from your collaboration or prompting tool, it has the freedom to find any data in your workspace.

This is especially useful in Slack and MCP - which were previously restricted to the workspace’s default connection.

We’ve also heard from folks that they are a little worried that this might result in the Slack bot leaking or posting sensitive information into public channels. Fear not - we’ve also added a setting to let admins specify which data connections can be used by the agent in Slack.

🚊 Improved data discovery and visualization with subagents​

A hard data question isn't really one question. The Hex Agent can now break complex requests into focused workstreams using subagents - specialized helpers that each tackle a discrete part of an analysis in parallel that can effectively handle far more without hitting context window limits. The result is faster, more reliable answers without you having to change how you work.

The Data Discovery Subagent finds the right data connections, tables, and schemas before analysis even starts. This is particularly helpful for questions where you may want the agent to search through a complex warehouse setup.

The Data Visualization Subagent handles your charts end-to-end - choosing the right chart type, configuring axes and labels, and applying visual best practices - so you can stay focused on the analysis itself. Here are some nice charts the subagent has made:

🛠️ Other improvements​

• Hex CLI Open Beta - If you’re analyzing data using a coding assistant like Claude Code or Cursor or managing Hex projects in bulk, you can try out our command line tool to incorporate Hex into your code-based workflows. We’re still working on the CLI, but we’re excited to get people using it and to hear your feedback!

  Docs are available here, or just copy paste the shell installer: brew install hex-inc/hex-cli/hex

• Queue Prompts in Threads - Threads now has the same prompt queuing feature we added to the notebook agent a couple weeks ago. Want to correct the agent’s thinking? Or did you just forget something? Queue up your thoughts and have the agent pick them up when it’s finished with its current task.

• Attach charts to notifications - When scheduling a notification now you can attach specific charts to your notification instead of the recipient hunting through a big graphic to find the chart that matters to them. We've been sending pretty graphs in Slack and the team loves it!

• Run status and project metadata in one place — We’ve consolidated run and kernel info into the Run button at the top of Logic view—hover for status, memory, and environment details.