Skip to main content

Releases

View 2024 releases
Latest release:

🌐 Introducing: The Knowledge Engine​

Every time an analytics agent is asked a question, it makes dozens of small decisions about which context and data to prioritize and how to interpret a user’s intent. But every new thread re-searches and re-reasons from scratch, spending time and credits, opening the door for inconsistencies and making judgment calls that are hard to observe at scale.

What if the knowledge agents discover along the way wasn’t thrown away, but retained, curated, and compounded? Now, Hex does exactly that, with our graph-based Knowledge Engine powering Hex 4.0. Read more in our launch blog.

How it works​

Each night, a background curator agent reviews the day’s work in your workspace and identifies any new learnings from defined context, or judgment calls that agents made. It then reconciles them by merging duplicates, reinforcing existing knowledge, and flagging conflicting guidance.

Knowledge list in Context Studio showing each piece of knowledge with its status, purpose, last updated date, and retrieval popularity

Once reconciled, each piece of knowledge is ranked by what it means and how the agent should use it.

  • Verified knowledge is backed by defined context: a guide, a semantic model or an endorsed project
  • Supported knowledge is a well-supported decision corroborated by available context
  • Learning knowledge is a decision the agent had to make that is neither supported nor denied by available context
  • Contested knowledge is where the agent has to make a decision where sources didn’t match up
  • Debunked knowledge is shown to be incorrect, but used as a cautionary graph record

When someone asks a question in Hex 4.0, the agent will search the Knowledge Graph first for relevant knowledge instead of rediscovering context from scratch. This means not only fewer tool calls, but also fewer situations where agents divert from your trusted logic.

And the new choices or discoveries they do make? Those get reviewed in the background and reconciled with your graph.

🔭 Observing the facts agents know​

The graph functions as an observability surface into how agents are leveraging your existing context, what they’re learning from it, and what they’re having to come up with in the absence of clear guidance. So you can feel more informed about what context is working, and what you need to build or adapt.

Retrieval counts by knowledge purpose, with trend lines for resource selection, data interpretation, and definition

👩‍💻 Getting started​

If you’re using Hex 4.0, the Knowledge Engine will run every night and it will start powering your threads automatically. You’ll see knowledge powering agent thinking, and your threads will be faster.

In the initial beta, the Knowledge Engine only builds from sources that are accessible to everyone in your workspace. As we expand the beta, we'll evolve our authentication controls so as to safely expand its learning opportunities.

If you want to start exploring your knowledge, Admins and managers can head to Context Studio → Knowledge to review contested knowledge, which might indicate inconsistencies in your context layer or places where agents aren’t agreeing.

As graph usage builds up, use the newfound visibility to guide your decisions on where to invest in context management.

Manage learning and retrieval in Settings → AI & agents → Knowledge, and check out the Hex Learn docs for details.

The Knowledge Engine is rolling out in public beta alongside Hex 4.0. If you want to learn more about how we’ve rebuilt our entire platform to make analytical work faster, cheaper, and more accurate, join us at the Hex 4.0 Live Tour.

AI has made it easy for everyone to work with data, but it’s causing a lot of chaos.

People are downloading spreadsheets, screenshotting chat threads, building random HTML dashboards, and sending them around with no governance, permissions, or observability.

That leaves data teams with a tough call: let everyone use AI with their data (and sacrifice trust), or trust the answers (and people will probably work around you and use AI, anyway).

Hex 4.0 breaks that tradeoff. We’re introducing an all-new version of Hex, rebuilt from the ground up around code. With 4.0:

  • It’s easier than ever to ask questions and build things with data, from Hex or external agents
  • It’s finally possible to review and trust agent-generated analytics code
  • Apps run way faster, thanks to an all-new execution engine

Read more about the vision behind 4.0, or join us for the Hex 4.0 live tour to see it in action.

🧱 Hex projects are fully code-defined​

Every Hex 4.0 project is built on a code framework designed for data work. That gives the Hex Agent more flexibility to build what you need, whether that’s a chart, a dashboard, or a more custom interactive app.

Being code-defined also lets us simplify the agent’s harness and tooling, so it can build apps more efficiently.

That’s what makes everything else in 4.0 possible. Agents are great at writing and understanding code, and projects built from it can be diffed, versioned, and reviewed by default.

🔌 Build from anywhere​

Because projects are entirely made of code, any agent can see and edit their full contents. Ask Claude, Codex, or Cursor to build a data app and push it into Hex. It becomes a governed, shareable project with live data connections, real permissions, and usage logs.

You can also pull an existing Hex app into your agent of choice and keep working on it there. There’s nothing proprietary to lock you in, it’s all just code.

🔎 Observability across your agents​

Whether your team is asking data questions from ChatGPT, Slack, Claude, or Hex itself, you can now see every interaction in one place.

Context Studio includes external agent interactions alongside activity in Hex, giving you a full view of where and how your team is working with data.

🕵️ A new UI, built for reviewing and understanding​

As agents build more with data, you need an easy way to check their work. The new project UI assumes agents write most of the code, so it’s built for reviewing and understanding what they built.

Click any number or chart in an app and choose Inspect to see the logic behind it, which resources it used (like a semantic model), and how it connects to the rest of the analysis. From there, open the code to see the full DAG in a logical layout, or make a quick cell edit without leaving the inspect view.

This comes in especially handy when someone sends you a vibe-coded app to review. If weekly active users are higher than in your latest deck, inspect the metric to see if it counts logins while the deck counts completed actions.

⚡ Ridiculously fast execution engine​

Whether you’re building an app, checking its logic, or exploring the results, the experience should feel fast and delightful. We replaced the old Python kernels with an all-new execution engine and compute architecture. Apps load and respond dramatically faster.

Here’s how load times changed for a few of our internal dashboards, comparing the same apps in Hex 3.0 and Hex 4.0.

Dashboard load times in Hex 3.0 vs 4.0

You’ll especially feel the difference in big, interactive apps with lots of filters and drill-downs. When you make a change, Hex reruns only the cells that depend on it and reuses cached results for everything else.

🚀 What to expect in the public beta​

The public beta is available to all customers on a Hex contract. On a self-serve plan? Email beta@hex.tech to turn it on.

Creating a new 4.0 project from the + menu
  • Editors, Managers and Admins can create a 4.0 project using the + button on the home page.
  • The current version of Hex remains the default and isn’t going away. Your existing projects will keep working, and nothing will be migrated automatically.
  • Want to upgrade an existing project? Create a new 4.0 project, then give the agent a link to your existing project and ask it to convert it.
  • Some features you may be familiar with aren’t available in 4.0 yet. During the beta, please test it, run evals, and share your feedback!

We can’t wait to see what you build!

🔒 Governed mode​

Previously, workspace admins could restrict the data available to the Hex agent for Explorers by enabling Endorsed mode, which limited access to tables, models, and projects marked as “Endorsed.”

Governed mode takes things one step further and lets admins define their own rules for which data sources the Hex agent can access for Explorers, rather than relying solely on the “Endorsed” classification.

Some ways to set it up:

  • Semantic models only: Have the agent answer using only your defined metrics, without querying warehouse tables directly or exploring projects.
  • Projects only: Have the agent use dashboards and analyses your team has already built. It can run projects and change inputs, but won’t write new warehouse queries.
  • A custom mix of sources: For example, you could use all semantic models, only warehouse tables marked Gold, and only projects marked Certified or Trusted.

Already using Endorsed mode? Nothing changes. Your settings carry over as a Governed mode configuration that allows only endorsed data.

Governed mode controls what the agent uses, not what people can access. Anyone with access to a data connection can still query it directly. To restrict data access, set permissions in your warehouse.

Admins can set up Governed mode in Settings > Governed mode.

📊 Access usage and credit grants data​

Admins need a clearer picture of who’s using credits, and on what.

Agent Usage and Credit Spend API endpoints let Admins pull historical agent activity and credit purchase data from their Hex workspace.

Use the data to:

  • Create daily usage reports to track credit consumption trends
  • Attribute costs by user or group for chargebacks
  • Monitor budgets by setting up alerts for spending thresholds
  • Generate finance reports with detailed use and purchase data

Access this data via API, or pull it directly into your agent or terminal through the Hex MCP server or CLI.

Available to workspace Admins on Enterprise plans.

⬆️ Import Apache Ossie models​

If you’ve already defined your business logic in Apache Ossie, bringing it into Hex shouldn’t mean rebuilding the model by hand. Now you can convert an Ossie model into Hex semantic-project YAML with the Hex CLI:

hex context semantic-project convert \
--from=ossie \
--to=hex \
--input your-model.ossie.yaml \
--output ./hex-model/

This one-way conversion saves files locally for review.

To bring them into your workspace, configure sync for a semantic project in Settings → Data sources → Semantic projects. Test the changes on a dev branch, then merge to main to import them into Hex.

🗂️ Agent activity classifications​

Where someone opens the agent doesn’t tell you much about what they’re working on. People often explore data in a notebook, and a thread that starts as a quick question can end with a finished app.

Context Studio now categorizes agent activity by what they’re for, not where they started. Instead of separate "Threads" and "Notebook agent" categories, filters and charts now show three types:

  • Analysis: asking questions and exploring data
  • App building: building or improving an app, dashboard, or notebook
  • Modeling: building and editing semantic models

Managers and Admins can now see how their workspace uses the agent: how much is analysis and how much is building for others. Filter by type in Context Studio or with hex thread list --classification.

⚡ Fast chart edits in Generative Apps​

A small chart edit shouldn’t require the full Hex Agent. Hex can now route targeted changes (like moving a legend, adjusting formatting, or adding a metric to the tooltip) to a smaller, faster model to reduce both wait time and token usage.

If a request goes beyond a simple property change, it goes to the full Hex Agent as usual, so you don’t have to decide which to use.

💬 Chat with Generative Apps​

Data apps are built to answer questions, but the best insights often lead to new ones. Explorers can now chat with published Generative Apps to ask follow-up questions, investigate metrics, and keep exploring in an agent Thread without leaving the app.

Tip: @-mention a specific cell or name the tab or section you’re viewing, along with any relevant filters or inputs.

Learn more about Chat with App.

📎 Schedule CSV exports for Generative Apps​

You can already set scheduled notifications for your Generative Apps, giving your stakeholders updated insights when they need it. Now you can include a CSV export.

Scheduled exports use the project’s underlying cell outputs — not transformations made in the app interface. Before scheduling an export, confirm that its cell outputs contain only data appropriate for the notification’s recipients.

🤖 New models to pick from​

Different tasks call for different models. Six new options are now available in Hex:

  • Opus 5.5 for hard, open-ended investigations
  • Fable 5.1 for complex, nuanced work
  • GPT 6 Sol for maximum reasoning depth on tough problems
  • GPT 6 Astra for involved analytical work
  • GPT 6 Luna for quick, lightweight tasks
  • Grok 4.6 for another model to try across everyday workflows

GPT 6 Sol, Astra, and Luna also support an optional Fast mode for quicker responses, at a higher credit cost.

Availability: Fable 5.1 is available on multi-tenant and non-EU, non-HIPAA single-tenant stacks.

To enable Fable 5.1, go to Settings → AI & Agents and turn on Allow data retention for specific models. No action is needed for the other models, they're already available in the model picker in the prompt bar.