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

1. Manual (preferred)

Option 1: Native image executable

⚠️ Important: GraalVM native image build may have some limitations:

• Build process might fail on some platforms (especially Windows) due to missing development tools
• GraalVM environment variables and paths must be properly configured
• Reflection and dynamic class loading may require additional configuration
• Build times vary significantly across machines

Use the latest GraalVM version and ensure all dependencies are installed.

⚠️ REQUIRES JAVA 21+ GRAAL VM SDK

• Build native image:

./gradlew clean nativeBuild

• The executable will be created at ${projectRoot}/build/native/nativeCompile/jllama

• Run with:

./jllama chat "Hello, help me please"

Option 2: Plain JAR file (slower but more stable):

⚠️ REQUIRES JAVA 21+

• Build project:

./gradlew clean build

• The JAR file will be at ${projectRoot}/build/libs/jllama-cli-*-all.jar

• Run with:

java -jar jllama-cli-*-all.jar chat "Hello, help me please"

2. Download releases

• Native executable or JAR file (platform testing in progress)

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