New MCP connectors let Zoë read the context your stack already holds, and act in the systems your team already uses.
​

If you're the one who owns your data stack, you know the fear about adopting a new agent: "Great, now I have to maintain yet another context layer."

It's a fair concern. Your business logic isn't in one place (impossible in any large company), it's scattered across platforms, each holding definitions someone fought hard to get right.

Drop in an agent that starts from zero and its definitions will drift away from your other tools. You're stuck with either wrong answers or a massive amount of manual synchronization, and both of those land on your desk. What if the agent played nicely with your existing data context?

Today, we're launching MCP connectors for Zoë.

Zoë connects directly to your existing tools as an MCP client, reads the context that's already there, and sets herself up. No rebuild. The years of logic locked in your other tools becomes the starting point.

Then the loop closes:

• Context flows in over MCP, on a schedule.
• Zoë gets smarter as people ask real questions, and standardizes her own context.
• Updates flow back to your source of record, written directly or proposed as a PR on a schedule.

Zoë uses the same definitions across analytics and operations. And every question makes the next one smarter.
​

The loop doesn't stop at analysis

Data → Analysis → Insight → Action → Behavior ↺ back to Data
​

Use cases: data to action

Each one starts with analysis Zoë already does well, and ends with an action she writes back through MCP. The final row is the general pattern: if you've automated it, Zoë can trigger it from an insight.

You stay in control

As an admin, you control all of it: which connections exist, which tools Zoë can call on each one, and when credentials get rotated. Set it up in Workspace Settings → Extensions → MCP.
​

Read the docs →

In short: Zoë moves from a place you go to read numbers, to a teammate who reads the numbers, acts on them across your stack, and closes the loop.

A few updates shipping this week and next.
​

Live now!

Artifact Run History
Track every scheduled Artifact delivery from the Run History tab in the Schedule Artifact Delivery modal. Confirm sends, troubleshoot misses, and jump to the Zoë chat behind any run.
​Docs →

Coming next week!

Chat side by side with the Context Manager
The Context Manager and chat are going to be built directly next to each other for easier development instead of the current pop-up modal. Plus a full-size view option for the Context Manager.

​MCP: Token-based authentication
Connect to Zenlytic via MCP using token-based auth. Docs →

Working with agents to create rich outputs like interactive dashboards is a blast, but every time you want to make one small edit, it takes forever to iterate and rebuild the whole asset just to change the small thing you asked for.

“It takes so long to make small changes” was one of the top pieces of feedback on artifacts right after we launched them.

Today, we're launching a visual editing experience for artifacts, which is WAY faster than just asking the agent to change something.

It lets you be more specific about what you want changed in an easier-to-use form factor and applies your changes in seconds instead of minutes for large artifacts.

There's still lots of work left for us to do to make the artifact experience feel snappy with such rich outputs, but this is a great step forward in making editing fast and easy for artifacts.

Here's how it works →

Managing context for a data agent is hard.

There are so many special cases, so many scenarios where a simple question actually kicks off a whole day’s worth of nuanced work.

You want it to know how your team defines a qualified lead, the weird returns logic from your old ERP, which orders count as fulfilled. But dump all of that into every question, and quality falls off a cliff.

Agents get worse, not better, when you drown them in context they don't need.

Zoë has had Skills to solve this problem for a while, but now she can create and manage those skills for you. She can craft her own reusable bits of know-how, which she pulls in only when a question actually needs them.

This makes it dramatically easier to create skills, which in turn reduces context bloat and helps Zoë perform better. Catch her getting something wrong, and she updates it so it doesn't happen again.

The logic compounds without the context bloating.

Check out how it works →

Every analytics tool I've used asks you to define your KPIs up front. Revenue, active users, churn, retention, etc. You pick your definitions, write the SQL, and maintain it forever.

The problem is nobody actually knows all their KPIs on day one. Humans are really bad databases.

Stop for a second and list all the cheeses you know. It’s hard to even get to 10 cheeses in 60 seconds!

In data, new questions surface new metrics. Definitions drift. The list is never easy to recall or even completely done.

So when you're setting up a new tool, you're stuck. Either you spend weeks (sometimes months) defining metrics before you get any value, or you skip it and the numbers come out wrong.

Now (with self-learning) Zoë can build them out herself. Docs > Here

Ask her a question that needs a metric she doesn't have. She'll work out the definition from your data, propose it back to you in plain English, and save it for you once you confirm. Net revenue retention, weekly active accounts, qualified pipeline, whatever you need. She writes the SQL, names it, and adds it to your model.

The metrics library grows as you actually use the tool, not as a prerequisite to using it.

Here’s how it works.

PS: Lmk in the comments if you actually got above 10 cheeses

Every analytics agent lives or dies on one question: "Do your numbers match the report I already trust?"

When they don't, no one cares why. The tool just doesn't get used.

Reconciling is brutal. You're chasing filters, joins, time grains, definitions, weird edge cases someone added two years ago and never wrote down. It takes hours (I’ve spent way too many weekends on this…).

Now, as part of Zoë’s self-learning ability, she can do it herself!

Drop a screenshot of any report into Zoë: a Tableau dashboard, a Looker look, an old spreadsheet, a slide from last quarter's board deck. Tell her to match the numbers.

She reads it, takes a swing, compares her answer to yours, figures out where the gap is, updates her own context, and tries again. Until everything ties.

Then she saves what she learned, so the next related question already has the right context.

Seeing this in beta was one of the biggest “I can’t believe AI can do this so well” moments for me.

Check out how it works →

It's way too difficult to see the context your agents are working with when you have to go to a different screen to see it.

I constantly find myself going back and forth between the agent's context and its answers to make sure that I'm giving it the right information so that it can answer my questions well.

Browser tabs are flying everywhere, and it ends up being a complete mess.

And now with Zoë’s ability to edit her own data model, the use case for seeing that context changes. It’s more important than ever to see the context from the agent's perspective, so you know exactly what is happening.

As part of Zoë's new self learning capability, we’re launching our upgraded context manager, which makes viewing the context for Zoë (our AI data analyst), easy and instant.

This one is going to see some great improvements over the next few weeks and months as well. We’re still figuring out the best ways for humans to work on shared context with agents!

For now, you don't have to worry about being taken out of your flow of working with the agent to see what it sees.

Have ideas for how you’d like to see this info? Let me know in a comment!

Note: Context manger will only appear if you have a Develop role or higher. 

What's the number one problem with analytics agents?

Setting up an AI analytics agent has been our #1 sales objection and the #1 reason that companies pump the brakes and say, “Maybe after our data engineering project.”

Messy data, missing context, no clear place to put any of it. We knew it. We've been heads down fixing it. Now, we're shipping the solution.

Today, we handed the reins over to Zoë.

She's incredibly capable, and instead of a human having to go back and forth reasoning about why the numbers aren't matching and how to prompt her differently, Zoë can now do that entire loop end-to-end.

In all of our testing she does a better job than any human (us included) at learning a new environment and making sense of messy data.

Customers are loving it so far, too! One Fortune 500 customer put it best: “This is now magic, and all the joins it created in a few seconds would have taken me hours to do manually.”

This feature was frankly just too exciting to keep to ourselves. That's why we're launching our self-service experience with Zoë, trained to set herself up for you.

Hop in. Try it. Upload one of your existing reports and tell Zoe to set herself up!

Zoë is our AI data analyst, and her artifacts (pptx, excel, and interactive dashboard results) have been a massive hit!

But nearly all of our customers complained about the same problem with artifacts when we launched them…

With such a rich visual result, it was really hard to see what data was actually used, and how the agent used it. It actually made the QA process to confirm the artifact was what you wanted much more difficult.

Today we're solving that! We've added citations to artifacts so that (in one click) for every beautiful output Zoë makes for you, you can see all the different queries that she used and how she's used them in the result.

I'm really excited about this and how we’re making it even easier to check which data is driving your artifacts →

I often find myself wishing that Zoë could find and iterate on my artifacts without me having to find the artifact first, and now she can!

You can ask Zoë about any artifact you have access to, or paste the link in and she can search across all your artifacts and find it for you.

She can show it to you, edit it, or update it with new or different data. Whatever you want.

I love this feature because it gives Zoë more and more agency (pun intended) over how she works with artifacts and it gives me fewer things I have to remember.

This is how that works →