Embedl agent

The Embedl agent works alongside you in the Hub. Ask it to investigate results, help document a decision, or update a project or run. It uses the current page and project as context and can consult these guides when you need help.

Give it a task#

Open the agent panel from the page you are working on. Describe the outcome you want and include any context it needs, such as the runs to compare or the reason for a change. It starts with your current project; name another project to work elsewhere.

For example:

• “Compare the accuracy and latency of these two runs and summarize the tradeoff.”
• “Add to this run’s description that we chose INT8 to meet the latency budget.”
• “Create a project called Classifier experiments.”
• “Help me add metric logging to my evaluation script.”

The agent can read projects and runs, create or rename projects, and create or update runs. For code tasks, it can use the documentation to help you write the code you will run.

Work through the result#

Continue in the same conversation to add context, refine the task, or ask for a follow-up change. For example, after comparing two runs, explain why a small accuracy loss is acceptable and ask the agent to record that rationale in the chosen run’s description.

Follow the links in its response to inspect the results and any updated records. Use Stop generating to stop a response and Clear conversation to start again.

Sharing information with the agent#

Before including sensitive information in a conversation, ask your administrator which model service processes it and what your team permits you to share.