Introducing Threads

TL;DR: Threads brings the power of agents to your organization, letting everyone ask questions and get back trusted answers. It taps into governed data like semantic models and gets better the more you use Hex.

Join me for a live virtual demo of Threads on October 7th at 11 AM PT / 2 PM ET!

These days you can throw a rock and hit a “chat your data” product — in fact, I tried to build one more than two years ago. Back at Hashboard, we had played with ChatGPT and thought we could wire up an LLM to our semantic models and — boom — business users would be answering their own questions.

But it turns out this is really hard, and it didn’t really work. And we weren’t alone: what everyone is finding is that standalone AI-data-chatbots struggle to get to the level of quality users need, don’t have the right context, and aren’t integrated into existing analytic workflows.

Now at Hex, we’re taking a new approach to this: Threads.

Threads

"I want to chat with the data!" If you’re on a data team, you probably hear this type of demand on a weekly basis. Executives, product managers, customer success reps, account executives — everyone expects to be able to ask a question and get an answer, just like they do in ChatGPT or Claude.

But you also want the answers to be trusted — and integrated with the rest of the data workflows on your team. That’s what Threads brings.

Our customers tell us that Threads is the first AI agent for non-technical users that really pushes over the chasm, and clears the quality bar needed to make the tech actually useful for folks who don’t code.

The big, not-so-secret reason it's so good is because it's built on the same framework and foundation as our Notebook Agent, which is already changing the way data teams do their work. Now, Threads is bringing that same magic to everyone else who wants to work with data.

But don’t take my word for it — try it yourself in Hex! This isn’t a preview release; it’s a public beta you can sign up for and start using today.

A partner for reasoning about data

When you ask a question in Threads, it doesn’t just spit out a chart and hope for the best. Instead, it plans, uses tools, and iterates through results — drawing on semantic metadata and context to get you the best possible answer.

Threads will feel like the thinking mode of other AI products you’ve interacted with. This is a big departure from most text-to-answer analytics solutions that try to one-shot their way to a simple chart and summary.

In Threads, you can follow along with the agent’s step-by-step reasoning and see how it’s working through your question. The agent has access to powerful built-in tools to search across different semantic models and database tables, write SQL, spin up viz cells, profile data, and even look up Hex documentation.

Not only does the visibility into the agent’s reasoning make it easier to audit, you’ll also notice far more useful outputs as a result of this deep thinking process. If you ask an open-ended question, Threads will show you different possible approaches, and might even ask you for your feedback on which direction take.

Data teams in the driver’s seat

We believe the data team will be central to maintaining and governing the quality of AI responses, similar to how they control and manage a company’s most important dashboards. This is a huge opportunity for them to go broader and be more impactful – but their ability to govern AI context will be critical.

So, we designed Threads to give data teams control. It can be enabled for “Explorer” seats in Hex, and data teams decide what data they can access — whether that’s just semantic models, or a broader set of endorsed warehouse tables.

While we have ideas and guidance on setting up agents for success, you ultimately know your organization best. Hex gives you the tools to manage what data is trustworthy and how Threads should use it.

Gets better as you use it

Hex gives you a lot of ways to provide context to agents for better answers over time. As people use Hex, your answers are always improving. Check out the Threads quickstart guide, but here are the sources it draws on:

• Projects and published apps: Threads can look at work that already exists in Hex to see if a question has been answered before. Endorsed projects are prioritized and treated as authoritative, so your knowledge compounds as you build.

• Semantic models: Sync your own from dbt, Cube, or Snowflake, or build them directly in Hex.

• Workspace rules: Add unstructured context like business rules or terminology, so Threads speaks your organization’s language.

• Endorsed tables: Marking tables as endorsed helps Threads use them as trusted context.

• Schema documentation: Any documentation pushed into your data warehouse is indexed and searchable.

What’s surprised me the most, as a former head of data, is how comfortable I feel advising orgs to configure Threads to use both semantic data and warehouse tables directly. The agent heavily prioritizes governed data, only falling back to raw tables when it doesn’t have solid modeled data to go on — so you get breadth without losing trust.

Built for collaboration

Threads is powered by Hex notebooks under the hood, so it’s easy for a business user to kick off a Thread and tag in their data team if they hit a roadblock. Inside a notebook, you can inspect, troubleshoot, or dig deeper into any charts or queries — and use the Notebook Agent to speed things up!

If other folks would benefit from the output, you can easily turn it into a dashboard and publish it. Now with the answer as a published app, Threads can draw on that context in future responses — fueling a virtuous cycle. The more assets you build in Hex, the better Threads gets.

Batteries included

Before Threads, the most compelling solutions I saw from our customers were home-grown solutions they glued together themselves. Infrastructure and data providers are offering language models and text-to-SQL tools, but they aren’t plugged into where your data team is working. This requires setting up a whole new stack — maybe a custom frontend, maybe something wired into Claude, maybe a standalone chat app duct-taped together.

Threads takes all of that off your plate. It’s fully integrated into Hex, removing the overhead and complexities of managing an in-house solution.

Agents that start with the data team

At Hex, we believe the value of AI in analytics starts — and succeeds — with the data team. That’s why our first two agent releases last month focused on the data team’s core workflows:

• Notebook Agent: the best way to do deep data analysis with AI right inside Hex’s notebook.

• Semantic Model Agent: maintaining semantic models manually takes a ton of work. Our Workbench Agent makes it faster than ever to model data. Unlike some tools, semantic models aren’t required to use Hex, but they make the experience even better when you do.

As part of our Fall 2025 Launch, we’ve also dramatically improved these two features. The Notebook Agent can now access your entire semantic catalog and query semantic models with new tools. The Workbench Agent also gained a number of big improvements — from the ability to tag projects to better development workflows for accepting and rejecting the agent’s changes.

More to come

We’re focused on accelerating everything the data team does with AI. Threads is in public beta today, and we’re already building what’s next: deeper integrations, richer context, and higher-quality answers.

We see a future where data teams expand to own this new type of experience, and we have a huge amount of work planned to help you manage, monitor, improve, and scale insights over time. If you give Threads a try, let us know what you think!