Set up Agent mode in Crade

Most Crade workflows are conversational: you ask, it answers. Agent mode adds a layer on top: the AI can use tools on your machine. File system, shell commands, web fetches. The hero examples on the homepage (building a landing page, generating a PDF quote, writing notes to a file) all use Agent mode.

What Agent mode does

When Agent mode is on, Crade exposes a set of tools to the AI: Read (open a file), Write (create or replace a file), Edit (modify a file), Bash (run a shell command), Glob (find files by pattern), Grep (search file contents), WebFetch (fetch a URL). The AI decides which tools to use based on your prompt.

Example: you say "Save a summary of this PDF to ~/Desktop/summary.md". With Agent mode off, Crade gives you the summary in the chat window. With Agent mode on, Crade writes the file and confirms in the chat. You did not have to copy anything.

How to turn it on

1. 1

   Open Crade settings

   Gear icon or menu bar.

2. 2

   Find "Agent mode"

   Usually under the AI Provider section.

3. 3

   Toggle it on

   A switch or checkbox. Crade confirms the mode is active.

4. 4

   Test with a low-stakes prompt

   Try something simple: "Make a file at ~/Desktop/hello.txt with the text 'Hello from Crade'." Crade should write the file and tell you so.

5. 5

   Verify the file exists

   Check your Desktop. The file should be there. If yes, Agent mode is working.

What changes after you enable it

• Crade can write and edit files on your machine without you copying from chat.
• Crade can run shell commands (after you confirm potentially risky ones).
• Tasks that touch the file system finish in one prompt instead of you handling the last step.
• The chat shows tool-use events as they happen ("Reading X.md", "Writing Y.pdf") so you can follow what is happening.

Safety

Agent mode runs with the same permissions as Crade itself: it can touch files you can touch, run commands you can run. Crade does not sandbox the AI, so think of Agent mode as similar to letting a smart assistant use your terminal. For low-stakes tasks (writing notes, generating PDFs, scaffolding projects), it is great. For destructive tasks (rm, sudo, anything mass-editing), be deliberate about the prompts you give.

• Crade asks for confirmation before running destructive shell commands.
• You can stop a running task with the cancel button.
• Files written to your filesystem stay there until you delete them. Crade does not undo on its own.

When Agent mode shines

• Scaffolding projects ("Make a Next.js app with the structure from this brief")
• File generation ("Save this quote as a PDF")
• Running diagnostic commands ("Check why my Wi-Fi is slow")
• Refactoring across files ("Rename this function everywhere in the project")
• Quick data tasks ("Pull the totals out of this CSV and append to summary.md")

When to turn it off

• You are using Crade on someone else's machine.
• You are working with code or data where you absolutely do not want unintended file changes.
• You prefer to keep all output in the chat window where you can read before acting.

You can flip the toggle off any time. Crade falls back to read-only chat mode immediately.

The summary

One toggle in settings. Behind the scenes, Crade hands the AI a set of file-system and shell tools. Agent mode is the difference between Crade as a chat partner and Crade as a hands-on assistant.