Skip to main content

​
Overview

Use Poolside Assistant to complete complex, multi-step development tasks with more autonomy. It works in your development environment to gather context, use tools, and write code so you can move through extended workflows with less manual effort. You stay in control through the approval system.

​
How Poolside Assistant works

When you send a prompt to Poolside Assistant:
  1. Autonomous workflow: The agent plans and carries out steps toward your goal.
  2. Context gathering: It automatically collects relevant context from your codebase.
  3. Tool usage: It uses various tools to read files, write code, run commands, and more.
  4. Iterative refinement: It continues working until it achieves the goal or you stop it.

​
Controlling the agent

  • Interrupt anytime: Press Escape to stop the agent.
  • Approval required: You control what the agent can access through the approval system.
  • Full oversight: Review each action before it runs, or configure auto-approval for trusted operations.

​
Approval and permission system

Poolside uses a comprehensive approval system to let you control what Poolside Assistant can do in your environment.

​
How approval works

At each step, the agent requests permission before using a tool. You have three options:

Approve once

Allow this tool to run for this step. You’ll be asked again the next time this tool is needed.

Always approve

Allow this tool to run automatically in the future. This creates or updates an allow rule in your personal project-specific .poolside/settings.local.yaml file.

Deny

Do not allow this tool to run. The agent pauses and waits for new instructions.

​
Managing permissions

Poolside Assistant saves tool approval preferences in your project’s .poolside/settings.local.yaml file. You can use this file to:
  • Configure trusted tools to run without prompting
  • Maintain different permission levels per project
  • Review and modify approved tools at any time
For more information about configuring permissions, including how to run commands without approval prompts, see Tool permissions.
Start with individual approvals to understand what the agent is doing, then gradually enable auto-approval for the tools you trust.
Approvals reduce confirmation prompts but do not enforce runtime boundaries. To restrict what agents can do at runtime, use sandboxes.

​
Use Poolside Assistant

​
Modes

Choose between multiple modes depending on how you want to work. Agent mode supports autonomous workflows, while Chat mode provides conversational guidance without modifying files. See Modes for details.

​
Prompting

When you start a session, type a prompt to tell Poolside Assistant what you want it to do. While the model is responding, you can click the stop icon to interrupt it. Based on the initial answer, you can follow up with additional prompts to refine your request and provide feedback. If you do not like the output, you can regenerate the response.

​
Versioning

In a multi-turn conversation, each response that includes a file update creates a new version for the task. To restore a previous version of a file, click Restore next to the version number in the chat panel, or click Restore this version in the Review Changes tab. When you’re finished with your task, click Done to start a new conversation.

​
Workspace context

Poolside Assistant uses information from your workspace to understand your request and help complete it. In Agent mode, Poolside Assistant builds context dynamically during the workflow. In Chat mode, responses rely on the context you reference and the selected editor-derived sources. For more information, see Context.