Agent Chat in imux provides a conversational interface to interact with AI models for coding, debugging, and system administration tasks. Unlike a standard terminal, Agent Chat can autonomously plan and execute multi-step operations while keeping you in the loop for critical actions.
In this tutorial, you will set up your first session by connecting to a Clavue model, referencing specific files in your workspace, and running a low-risk task. You will also learn how to read and approve tool confirmations, ensuring that every action taken by the agent is intentional and safe.
By the end of this guide, you will understand the basic workflow of imux's Agent Chat, including how to manage context, verify outputs, and prepare for more complex multi-agent workflows.
Prerequisites
- imux installed on macOS
- A Clavue account with active membership
- A local project directory with at least one text file
Steps
- 01
Open Agent Chat and Select a Model
Launch imux and navigate to the Agent Chat interface. Click the model selector in the top-right corner and choose a Clavue model suitable for your task. For general coding assistance, a standard model like
clavue-3.5is recommended. Ensure your membership is active to access these models. - 02
Set the Workspace Context
Agent Chat needs to know which files to work with. Open your project directory in imux's file explorer. If you haven't set a workspace yet, refer to the
first-workspace-setupguide. Click on the specific file you want the agent to reference, or use the@symbol in the chat input to explicitly add files to the context (e.g.,@main.py). - 03
Compose a Safe Task
Type a clear, concise instruction in the chat input. For your first session, choose a low-risk task. Example: "Read the content of main.py and summarize the key functions." Avoid commands that modify system configurations or delete files. This ensures you can observe the agent's behavior without risking data loss.
- 04
Review Tool Confirmations
Once you send the prompt, imux will display the agent's plan and any tool calls it intends to make. You will see a confirmation dialog for each tool execution (e.g.,
Read File,Write File). Carefully read the details of the action. If the action looks correct, click "Approve". If it seems unsafe, click "Reject - 05
Verify and Iterate
After approval, the agent executes the tool and returns the result. Review the output in the chat window. If the result is not what you expected, you can reply with a follow-up question or correction. The agent will use the new context to refine its response. This iterative process is key to effective agent-assisted development.
- 06
Explore Advanced Contexting
Now that you have completed a basic session, try adding multiple files to the context using
@or by selecting them in the sidebar. You can also use slash commands like/helpto see available options. For more complex tasks, consider exploring themulti-agent-workspaceguide to see how different agents can collaborate.
Understanding Tool Confirmations
Tool confirmations are a critical safety feature in imux. They prevent the agent from performing destructive actions without your explicit consent. Always review the file paths and content modifications before approving. If you are unsure about a tool call, you can ask the agent for clarification before proceeding.
Using @ for File Context
The @ symbol is a powerful way to provide context to the agent. By referencing specific files, you help the agent understand the scope of your request. You can reference multiple files by separating them with spaces or commas. This is especially useful for debugging or refactoring tasks where the agent needs to see related code.
Safety Best Practices
Always start with low-risk tasks when using Agent Chat. Avoid granting the agent access to sensitive system directories or production databases. Use the diff-application-workflow to review changes before applying them. Regularly check the agent's actions in the terminal log to ensure they align with your intentions.
Tips
- Use the `@` symbol to quickly add files to the chat context without navigating away.
- Always read the tool confirmation details before clicking "Approve".
- Start with simple read-only tasks to get familiar with the agent's behavior.
- Use the `/help` command to see available slash commands and options.
FAQ
How do I add multiple files to the chat context?
You can add multiple files by typing @ followed by the file names, separated by spaces or commas. For example: @main.py @utils.py. You can also select multiple files in the file explorer and right-click to add them to the chat.
Can the agent delete files without confirmation?
No. imux requires explicit confirmation for all tool calls, including file deletion. You will see a confirmation dialog before any destructive action is taken. You can reject the action if it seems unsafe.
What models are available in Agent Chat?
Agent Chat supports various Clavue models. You can select the model from the dropdown menu in the top-right corner. The availability of specific models may depend on your membership tier.
How do I stop the agent from executing a task?
You can stop the agent by clicking the "Stop
Is Agent Chat secure for production code?
Agent Chat is designed for development assistance. While it includes safety features like tool confirmations, it should not be used to directly modify production systems without rigorous review. Always test changes in a development environment first.