Core Features
Agent Chat Interaction
This is the core interaction surface of the product. For every prompt, you can fine-tune how the session runs, including the AI model and permission mode. The goal is to give you clearer control over each Agent action so the workflow can adapt to tasks with different levels of complexity and risk.
Permission Modes
Permission modes control how the AI Agent executes actions and when it asks for confirmation. You can quickly switch between them depending on task risk, complexity, and how much hands-on involvement you want.
Available Modes
The menu shown in the screenshot includes the following modes:
| Permission Mode | How It Works | Best For |
|---|---|---|
| Auto Mode | Automatically chooses a more suitable execution strategy based on the current task and balances efficiency with safety | The recommended default for everyday use |
| Default Mode | Uses the standard confirmation strategy, keeping necessary review points without interrupting too often | Stable day-to-day use with a familiar interaction flow |
| Auto Accept Edits | Lets the Agent modify files directly, while still asking for confirmation on more sensitive operations | Code writing, doc editing, styling, and other file-focused tasks |
| Plan Mode | The Agent first understands the request and produces a plan before execution begins | Complex tasks with multiple steps where you want to align on the approach first |
| Quiet Mode | Shows fewer interaction prompts and reduces interruptions during longer runs | When the task direction is already clear and you want smoother execution |
| Skip Permission Checks | Runs actions directly and minimizes manual intervention as much as possible | Only in trusted environments when you fully understand the risk |
Tip: For critical files, external commands, or network activity, use a more conservative mode first. For routine edits, a more efficient mode can reduce repeated confirmations.