Sessions
A session is a conversation with Plotly Studio's AI agent. Within a session you can upload or connect to data, ask ad-hoc questions, explore visualizations, request written summaries, and build shareable data apps.
How sessions work
When you send a prompt, the agent plans a series of steps to fulfill your request, then executes them one by one. Each step represents a discrete unit of work, such as loading data, running a calculation, generating a chart, or writing a summary.
The agent writes and runs Python code behind the scenes for each step. You never need to write code yourself, but you can inspect it at any time (see Viewing step details).
Managing sessions
You can delete a session from the home screen of the agentic workflow. Click the trash icon in the upper right corner of a session to delete it. Tools for organizing sessions are on the roadmap.
When the agent asks for guidance
If the agent encounters an ambiguous situation, it will pause and ask you a clarifying question before continuing. For example, if it's unclear how to interpret a column in your data, or two approaches seem equally valid, you can answer in the chat to let the agent proceed.
Viewing step details
Select any step in the session to open the side panel. The side panel gives you a closer look at what the agent did and what it found:
- Visualizations: Expand charts and tables to view them in detail.
- Summary: A plain-language description of the results, written by the agent. Key statistics and figures (shown in bold) are interpolated directly from the code output, ensuring that every number comes from executed code rather than the model's general knowledge. This approach eliminates fabricated statistics, a common source of AI hallucinations.
- Code: The Python code that produced the result. You can review it to understand exactly what the agent did at each step.
Data sources and credentials
If your data source requires credentials (for example, a database connection or API key), the agent opens the Credentials panel so you can enter them securely. Credentials are stored in your local keychain and, when you publish an app, are stored as secrets on Plotly Cloud or as environment variables on Dash Enterprise.
See Data Sources for full details on connecting to external data.
Artifacts
During a session, you can ask the agent to produce structured outputs called artifacts. Two artifact types are currently supported:
Reports
Ask the agent to produce a written report or summary of your findings. When building a report, the agent will ask you a series of questions to tailor the output, such as:
- Which findings to include
- The level of detail (executive summary vs. deep dive)
- How much editorial commentary to add alongside the data
Like all summaries in Studio, facts and figures in a report are interpolated from executed code, not generated from the model's world knowledge.
Data apps
Ask the agent to build an interactive data app. The agent creates a Dash app with charts, tables, and controls based on your session data. You can iterate on the app through follow-up prompts. Each iteration is saved as a new artifact attached to a new step, so previous versions are preserved.
Once you're happy with the result, you can publish the app to Plotly Cloud or Dash Enterprise. See Publishing for details.
