Jllama CLI

A command-line interface for Ollama LLM models, designed for seamless integration with shell scripts and command-line workflows.

Features

• Command-line interface for text generation and chat
• System roles for specialized tasks (shell commands, git commits, etc.)
• Chat history persistence
• Shell pipeline integration
• Multiple LLM provider support (currently Ollama)

Installation

Quick Install (Recommended)

Clone the repository and run the install script:

git clone https://github.com/owpk/jllama-cli.git
cd jllama-cli
./install.sh

The script will:

• Check for GraalVM Java 21 and native-image tool
• Build native image executable
• Install jllama binary to ~/.local/bin

Requirements: GraalVM Java 21 with native-image support

Manual Build

Option 1: Native Image (faster execution)

./gradlew clean nativeBuild
./build/native/nativeCompile/jllama "Hello"

Option 2: JAR File (slower but more compatible)

./gradlew clean build
java -jar build/libs/jllama-cli-*-all.jar "Hello"

Download Releases

Pre-built binaries available in Releases.

Setup and Configuration

Initial Setup

The application creates a default configuration file on first launch.

• Or you can find default jllama.yaml in the project root.

Configuration Path

~/.jllama/jllama.yaml

Default Configuration

providers:
    ollama:
        baseUrl: "http://localhost:11434"  # Ollama server URL
        modelName: null                # ⚠️ Set your installed model name

Required Configuration

1. modelName: Set your installed Ollama model Example for mistral:

   ollama pull mistral

   Then update config:

   modelName: "mistral"

2. baseUrl: Change if your Ollama server uses a different address

Verify Configuration

jllama "Hello, are you there?"

Troubleshooting

If you see "model not found":

1. Check if the model is installed (ollama list)
2. Verify the model name in config
3. Ensure Ollama server is running (ollama serve)

Usage

Basic usage:

jllama "write a simple program"  # chat is the default command

Use role:

jllama --role-name "role name" # or -r

Start new chat:

jllama --new-chat  # or -n

Generate text:

jllama generate "Your prompt here"

Chat mode:

jllama chat "Your message"

Predefined System Roles

The CLI comes with predefined roles:

1. default - General-purpose helpful assistant
2. shell - Generates shell commands for your OS
3. git-autocommit - Creates commit messages based on changes
4. describe-shell - Explains shell commands

Pipeline Integration Examples

Generate commit message based on git changes:

jllama -r git-autocommit generate "$(git diff)"

Get shell command explanation:

jllama -r describe-shell generate "ls -la | grep '^d'"

Configuration Options

Set LLM provider: Ollama is only supported and default by now

jllama -p ollama generate "Hello"

Default Configuration

• Default provider: Ollama (http://localhost:11434)
• Default role: general-purpose assistant
• Chat history is automatically preserved

Project Structure

The application is built using:

• Micronaut framework for dependency injection and HTTP client
• Picocli for command-line parsing
• Reactive programming with Project Reactor
• YAML for configuration

Roadmap

Planned LLM Provider Support

• OpenAI API integration
• Mistral AI support
• Maybe others :)

Upcoming Features

• Image generating/processing
• "Suffix" support
• Advanced parameter configuration (temperature, top_p, etc.)
• Model Context Protocol (MCP) support for standardized LLM interactions
• Context window management
• Embeddings API
• API key management

License

MIT