Smart Tasks - Overview

Smart Tasks is the code execution capability available when using the Plan & Act mode. When a user asks the agent to work with data, produce a file, connect to an outside service, or visualize information as a dashboard, Smart Tasks writes & runs the necessary code in a secure environment - automatically, without any technical input from the user.

Smart Tasks are available only when using the Plan & Act mode.

Four capabilities

Each capability can be enabled or disabled independently.

Read Files

The agent reads and analyzes uploaded files in full. It processes the complete file - not a summary or sample - and responds with findings, calculations, or summaries based on what is there.

Supported input formats: pdf, docx, doc, odt, txt, jpg, jpeg, png, webp, xlsx, xls, csv, tsv, json, md, pptx.

Create Files

The agent generates a downloadable file from data in the conversation. When output is too large or structured for an in-chat response - a formatted report, a cleaned dataset, an export - the agent produces a file the user can download directly.

The wide variety of output formats is supported, including but not limited to md, txt, pdf, docx, csv, html, etc.

Make External Connections

The agent connects to outside services to fetch live data, send requests, or interact with an API. Use this when your agent needs information that is not in the knowledge base and changes over time.

Live Dashboards

The agent generates an interactive dashboard directly in the conversation. Instead of a text summary or a file download, the user gets a live view with charts, filters, and formulas they can explore in real time.

The agent chooses Live Dashboards when data is best understood visually- comparisons, trends, or datasets the user might want to filter and explore. Users can also request a dashboard directly.

Dashboard data can come from uploaded files, knowledge base results, code output, or any combination.

How it works

The agent decides which Smart Tasks capabilities to use based on the request. It runs them in sequence, in whichever order the task requires, and shows each step to the user in real time. If something fails, the agent analyzes what went wrong, adjusts its approach, and tries again - without stopping to ask unless it genuinely cannot continue.

Before responding, the agent reviews its own output for errors or poor formatting.

Security

Smart Tasks run in an isolated environment in the cloud, and only has access to your systems for which you explicitly provide a MCP server or API key.

Files uploaded or generated during a conversation are available to download within that conversation only.

Example use cases

Upload a sales spreadsheet and ask for a regional breakdown with month-over-month trends
Ask the agent to produce a formatted summary report and download it as a file
Fetch live currency rates from a public source and use them in a calculation
Upload a dataset and ask for an interactive dashboard with filters by category and date

Updated 23 minutes ago