ollama-agent-cli

ollama-agent-cli is a command-line interface (CLI) tool that allows users to interact with a local Ollama server, leveraging the power of large language models to execute commands and provide intelligent responses.

Features

• Interactive Mode: Engage in a conversational chat with the Ollama model.
• Command Execution: The agent can propose and execute shell commands on your system with explicit user approval.
• Smart Approvals: Approve commands once, and the agent remembers your preference for future similar commands.
• Self-Correction: The agent attempts to fix failed commands by analyzing errors and proposing corrected solutions.
• File Manipulation: Includes a replace command to modify file content.
• Conversation History: Maintains context across interactions in interactive mode.
• Clear History: Use /clear to reset the conversation history.
• Cancel Operations: Press ESC to abort ongoing command executions or model responses.
• Debug Mode: Get detailed JSON output for debugging purposes.
• Non-interactive Approvals: Use --yes or --yes-all flags for automated command approval.
• Colorful UI: Enhanced user interface with ANSI colors for better readability.

Installation

To install ollama-agent-cli globally, ensure you have Node.js and npm installed, then run:

npm install -g ollama-agent-cli

Usage

Interactive Mode

To start an interactive chat session with the Ollama model:

ollama-agent --interactive

In interactive mode, you can type your prompts. The agent will respond conversationally or propose commands. You can also use special commands:

• /bye or /exit: Exit the interactive chat.
• /clear: Clear the conversation history.

Single Command Mode

To send a single prompt to the Ollama model and get a response:

ollama-agent "What is the largest file in the current directory?"

Options

You can use the following options with ollama-agent:

• -m, --model <name>: Specify the Ollama model to use (default: llama3).
• -H, --host <url>: Specify the Ollama host URL (default: http://localhost:11434).
• -i, --interactive: Start in interactive chat mode.
• -d, --debug: Enable debug mode, showing detailed JSON output and command execution details.
• --yes: Automatically approve the next proposed command.
• --yes-all: Automatically approve all proposed commands without asking for confirmation.

Command Approval

When the agent proposes a command, you will be asked for approval. You have three options:

• (y)es: Approve the current command for a single execution.
• (n)o: Deny the current command.
• (a)lways: Approve this type of command (based on its root command) for all future executions without asking again.

AI Command Format

When the Ollama model proposes a command, it should follow this strict format:

<explanation of the command>
@@COMMAND@@
<the command to execute>
@@COMMAND@@

The agent will parse this format to extract the explanation and the command for execution.

Development

Project Structure

/ollama-agent-cli
|-- .env                  # Local configuration file (ignored by git)
|-- .env.example          # Configuration template
|-- .gitignore            # Files to ignore by git
|-- GEMINI.md             # Project architecture documentation
|-- package.json          # Project definition, dependencies, and scripts
|-- index.js              # CLI entry point
|-- /lib
|   |-- ollama.js         # Module for communication with Ollama API and core agent logic
|   |-- tools.js          # Module for shell command execution and file manipulation
|-- README.md             # This file

Running Locally

1. Clone the repository:

git clone [email protected]:ericmora/ollama-agent-cli.git cd ollama-agent-cli 2. Install dependencies: bash npm install 3. Run the CLI: bash ./bin/ollama-agent --interactive ```

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

License

This project is licensed under the ISC License. See the LICENSE file for details.