Overview
Plotly Studio Embedded does agentic data analytics and visualization. It takes on real work: it can join datasets, derive metrics, work multiple steps at a time, and generate highly custom Plotly graphs. It will raise questions when it encounters ambiguity, shows its work, and surfaces its queries for you to verify.
Plotly Studio Embedded is where Data Apps meet AI meets BI; it gives your audience a open-ended interface ask additional questions not presented in the original Dash app while giving your development team a break from answering the long-tail backlog of data requests.
The best part is that it's easy to roll-out and adopt. Viewers don't need to configure anything themselves to get started: data access preconfigured in advance by the development team in the Dash app.
Enterprise-Ready in Five Ways
Our customers needed a product that satisfied five key requirements:
- Secure (constrained and configurable)
- Easy to roll-out (zero setup for end users)
- Embedded (co-exists on the Dash app)
- Trustworthy (transparent and benchmarked)
- Purpose-Built for Data Analytics
1. Secure: Constrained and Configurable
Data access stays under your control. Your development team has full control over which datasets Plotly Studio Embedded has access to and who can access the application. This allows you to restrict, filter, or gate data access as part of the app. Rather than giving your audience a connector to the entire data warehouse, you can write queries in advance in code that fetches just a part of the warehouse - certain rows or certain tables - and only expose this restricted segment of the data to your users. From a governance perspective, this is a very simple and configurable solution.
Only people who can view the app can use the chatbot, and every viewer authenticates with their own Plotly Cloud account. Each conversation draws AI credits from the viewer's team, so AI token costs can be measured and billed according to use rather than ownership.
The Plotly Studio agent won't go rogue. Rather than deploying an agent that can do anything, we engineered the Plotly Studio Embedded agent to only perform data analytics and visualization tasks. This makes it secure by design: unlike coding agents, there's no arbitrary code execution, no network access, no filesystem access, and no data exfiltration vectors.
2. Easy to Roll Out: Zero Setup for End-Users
What make Plotly Studio Embedded secure is also exactly what makes it easy to roll out.
Most AI initiatives stop short because they're too risky and complex to secure and configure across a workforce.
Configuration often requires each user to manage credentials. Users are often lost trying to navigate the myriad of connectors, plugins, credentials, skills, MCP servers, and so on just to access their data. The integration and onboarding becomes too onerous to achieve widespread adoption and ROI.
With Plotly Studio Embedded, we solved this integration problem by enabling the data team to configure access to the data in code in advance. As we mentioned above, this has an additional benefit of being secure and governable as the development team has full control over the data. Once it's set up, the workforce can chat with the data within the deployed Dash app without any additional setup.
It becomes as simple as visiting a URL in your browser.
3. Embedded: Co-Exists Along Dash Apps
Plotly Studio Embedded co-exists on top of the Dash app, mixing and matching predefined analytics created by the development team alongside agentic AI analytics generated on the fly.
We are strong proponents in this design and believe that this is how analytics should be delivered within organizations. Having both in the same pane of glass has a number of advantages:
- Reduced Token Cost: Common questions can be turned into predefined views within the Dash app by the development team, allowing end-users to get their answers without spending any tokens.
- Augmented Analytics: AI augments the rigid views that the end users see in the Dash app. They can follow their curiosities, sanity check results, ask for data to be presented in a different way, drill in further, run comparisons, generate charts with different styles, and so on. The data app and dashboard is no longer a dead-end.
- Single Codebase: The same data pipeline and queries that you use to access and display data within your Dash app can be used to expose data within the agentic analytics experience, reducing development cost, maintenance, and data touchpoints.
- Custom UI: Even though agentic analytics is extremely flexible - users can ask just about anything! - it can be useful to ground your end users with structured preexisting views of the data. This solution allows you to clearly display to the user what datasets are available, provide any additional context about the data, and present them with some starter charts. By customizing the Dash app, you can deliver analytics to your workforce that is much easier to get started from than a blank screen with a prompt box.
4. Trustworthy
Many AI experiences hide the work that they do under the hood to come up with the answer. In coding agents, analytic scripts will often be run in hidden, one-off /tmp folders. Similarly, in many agentic experiences in SaaS products, the thinking steps and tool-calls are hidden or obtuse.
Plotly Studio is designed to be transparent and show it's work. The queries and code generated at every step of the way is always available for you to inspect, verify, correct, and even run on your own. We treat the "Methodology" - the approach taken by the AI agent - as a first class design feature of the product.
Data is often messy. Analytics sessions often contain ambiguities, assumptions, and data quality issues. We've worked hard to tune the behavior of Plotly Studio to surface ambiguities and assumptions to the end user so that they're aware of the greater context of the answer and can steer their next question accordingly.
In our benchmark design, we test the product with messy data, ambiguous questions (questions where there could be multiple possible answers depending on the interpretation), and even impossible questions (where the answer can't answered with the dataset) to make sure that the agent doesn't display sycophantic, misleading, or narrow-minded answers.
We test the product against large, complex datasets that contain over 20M rows across 42 tables with features like messy column names, needle-in-a-haystack insights, null data, and intentionally "physically impossible" data to test how well it handles real-world datasets. Our benchmarks measure a wide range of our harness's analytics capabilities like joins, windows, filters, pivots, regexes, and so on. We also continuously test across multiple different models as they come out and chose the model that has the best accuracy. In our benchmarks, we've found that our Plotly Studio Embedded achieves A-level grades in these benchmarks.
5. Purpose-Built for Data Analytics
Most AI chat experiences show you a wall of text and an editorialized response instead of the actual dataset. Many AI products will also hide their thought process or the tool calls that used to come up with the answer. That's not how we think AI data analytics products should look like.
In Plotly Studio Embedded, we show you the data and queries front-and-center. This allows you to verify, validate, sanity check, and stay rooted in the data as it works. We've tuned the behavior to reduce sycophantic language and refrain from editorializing or jumping to conclusions; the behavior feels competent, neutral, and honest.
And it's performant. We run the analytics using a fast, columnar data querying engine. In our benchmark suite with 44 tables and 20M rows, most queries take less than a second.