Agent Mode

Overview

VoltX Agent Mode acts as an intelligent assistant directly in your terminal. You provide instructions in natural language, and the agent suggests a corresponding shell command. You can then review, approve, and execute the command.

Starting an Agent Session

You can initiate an agent session in several ways, similar to Chat Mode:

Default Agent

Starts an agent session using your configured default provider and model. If no defaults are set, VoltX will prompt you to configure them or select temporary options.

voltx agent

Agent with a Specific Provider

Starts an agent session using a specific provider. VoltX will use the default model configured for that provider, or prompt you if none is set.

voltx agent <provider-name>

Replace <provider-name> with the provider you want to use (e.g., google, openai).

Temporary Model Selection

Starts an agent session with a specific provider but forces a temporary model selection for this session only. This overrides any configured defaults for the duration of the agent session.

voltx agent <provider-name> --temp

Replace <provider-name> with the provider you want to use. VoltX will prompt you to select a model from that provider for this specific session.

Using the Agent

Once activated, the agent will prompt you for input in your current directory.

1. Enter Command: Type your desired action in natural language (e.g., "list all typescript files", "create a new directory called temp").
2. Review Suggestion: The agent will process your input and suggest a shell command.
3. Confirm Execution: You will be asked to confirm whether you want to run the suggested command.
4. Execute (or Skip): If confirmed, the command runs, and its output is displayed. If skipped, the agent waits for your next input.
5. Exit: Type exit or quit to end the agent session.

Command History

Within an active agent session, you can access previously entered natural language prompts:

• Type !h and press Enter.
• A list of recent prompts will be displayed.
• Select a prompt from the list to rerun it.