Available tools

The following page describes the tools that the W&B Agent uses to answer your questions and complete your requests.

The W&B Agent uses built-in tools to help it answer your questions and complete your requests.

Tools expand the capabilities of the agent and enable it to to gather information, analyze data, and interact with your W&B project.

The specific tools the agent uses depend on the context of your question.

Available tools

The following table lists a non-exhaustive list of tools that the W&B Agent may use to answer your questions and complete your requests.

Tool	Description
Python scripts	The agent can generate and run Python scripts that use the W&B Python SDK to query and analyze your W&B data. For example, the agent might generate a Python script to analyze patterns across runs, then run the script in a sandbox environment to get the results of the analysis.
Shell commands	The agent can run shell commands to interact with your W&B project. For example, the agent might use the W&B CLI to filter runs or create panels.
Reports and Workspaces API	The agent can use the Reports and Workspaces API to create and modify panels, saved views, and reports in your W&B project. For example, the agent might create a panel that shows the top performing runs based on a specific metric.
Python SDK	The agent can use the W&B Python SDK to query your W&B data and generate insights. For example, the agent might use the Python SDK to identify which runs have missing validation metrics.
W&B Weave	The agent can use W&B Weave for project analysis, runs, traces, evals, reports, monitors.

All scripts and shell commands are executed in a sandbox environment.