When you trigger a run, the agent operates in its own sandboxed cloud environment. You can observe the agent’s thought process and interact with it through the runs page. The main components of the run page are: the chat, the docs, the code, and the terminal.

​
Chat

  1. Send a follow-up message to the agent at any point to add clarification, steer it in the right direction, or request modifications.
Due to the asynchronous nature of Palmier, follow-ups are registered with the agent after its current step completes. In some cases, you may see an additional tool call executed before your follow-up is processed.
  1. Use the Pause/Resume button to stop or resume the agent’s work at any point during the run. If the agent has completed a run, you’ll need to send a follow-up message to reactivate it.
To conserve resources, we shut down the sandbox when an agent completes its run. When you send a follow-up, the environment is restored to where it left off, which takes approximately 20 seconds.
  1. Expand each tool call box to view the output. The contents are truncated for improved performance, but the agent receives the full content.
  1. If the ask_question tool is enabled for the agent, it may ask clarifying questions about the task to better understand how to help you. Questions will time out after 30 minutes if unanswered, but you can always follow up afterwards. For more information on agent tools, visit the Agents page.
Tool calls have a default 60-second timeout. This is often reached during large package installations or non-terminating commands (e.g., git log). See the Terminal section below for running terminal commands in the agent’s environment.

​
Docs

If the design_doc tool is enabled for the agent, it can output detailed design documentation in the Docs tab. This is particularly useful for:
  1. Planning tasks before executing them
  2. Visually separating agent thoughts from answers
  3. Displaying pull request reviews and other in-depth analysis
The documentation supports full Markdown formatting, including Mermaid diagrams for visualizing system architecture, workflows, and data flows. Diagrams are rendered interactively and can be clicked to view in an enlarged modal with zoom and pan capabilities. You can copy the raw Markdown contents of the document or create a project with this documentation for better context management. For more information, visit the Projects page. Docs tab showing design documentation
If you want the agent to modify the design document, explicitly ask the agent to call the design_doc tool. Otherwise, it may respond in chat instead.

​
Code

When the agent makes code changes, you can visualize them in the Code tab. From there, you can directly create a draft pull request with the changes or copy the raw diff (git format).
The default branch name will be created as palmier/[run-title-separated-with-hyphens]. You can push to your own specified branch instead.
The draft PR will be created with the “Palmier” label, and the description will include links to the run and the GitHub issue (if triggered from one).

​
Terminal

You can connect to the terminal and run commands directly in the agent’s environment, but only while the agent is actively running. This is particularly useful for installing packages, executing build commands that may have timed out for the agent, or running other arbitrary commands. Currently, the terminal can only be used while the agent is running since it requires an active environment. The best approach is to ask the agent to pose a question, then use the waiting period while the agent awaits your answer to run terminal commands. We’re working on a better solution for this limitation.