oli - Open Local Intelligent assistant

oli is an open-source alternative to Claude Code with powerful agentic capabilities for coding assistance. Features:

• A modern hybrid architecture:
  • Rust backend for performance and core functionality
  • React/Ink frontend for a beautiful, interactive terminal UI
• Support for both cloud APIs (Anthropic Claude Sonnet 3.7 and OpenAI GPT4o) and local LLMs (via Ollama)
• Strong agentic capabilities including file search, edit, and command execution
• Tool use support across all model providers (Anthropic, OpenAI, and Ollama)

⚠️ This project is in a very early stage and is prone to bugs and issues! Please post your issues as you encounter them.

Installation

Using npm

npm install -g oli-tui

This will install the latest version from npm and handle all dependencies automatically.

Using Homebrew (macOS)

brew install amrit110/oli/oli

Homebrew will download and install the latest version directly.

From Source

# Clone the repository
git clone https://github.com/amrit110/oli
cd oli

# Build both backend and frontend
./build.sh

# Run the hybrid application
./run.sh

Environment Setup

Development Setup

# Install Python dependencies (for pre-commit)
python -m pip install uv
uv venv
uv pip install -e .

# Install pre-commit hooks
pre-commit install

# Run Rust linting and formatting
cargo fmt
cargo clippy

# Run TypeScript checks in the UI directory
cd ui
npm run lint
npm run format

Cloud API Models

For API-based features, set up your environment variables:

# Create a .env file in the project root
echo "ANTHROPIC_API_KEY=your_key_here" > .env
# OR
echo "OPENAI_API_KEY=your_key_here" > .env

Using Anthropic Claude 3.7 Sonnet (Recommended)

Claude 3.7 Sonnet provides the most reliable and advanced agent capabilities:

1. Obtain an API key from Anthropic
2. Set the ANTHROPIC_API_KEY environment variable
3. Select the "Claude 3.7 Sonnet" model in the UI

This implementation includes:

• Optimized system prompts for Claude 3.7
• JSON schema output formatting for structured responses
• Improved error handling and retry mechanisms

Using Ollama Models

oli supports local models through Ollama:

1. Install Ollama if you haven't already
2. Start the Ollama server:

   ollama serve
   

3. Pull the model you want to use (we recommend models with tool use capabilities):

   # Examples of compatible models
   ollama pull qwen2.5-coder:14b
   ollama pull qwen2.5-coder:3b
   ollama pull llama3:8b
   

4. Start oli and select the Ollama model from the model selection menu

Note: For best results with tool use and agent capabilities, use models like Qwen 2.5 Coder which support function calling.

Usage

1. Start the application:

./run.sh

2. Select a model:

   • Cloud models (Claude 3 Sonnet, GPT-4o) for full agent capabilities
   • Local models via Ollama (Qwen, Llama, etc.)

3. Make your coding query in the chat interface:

   • Ask for file searches
   • Request code edits
   • Execute shell commands
   • Get explanations of code

Architecture

The application uses a hybrid architecture:

┌───────────────┐        ┌───────────────┐
│ React + Ink UI│◄───────┤ Rust Backend  │
│               │  JSON  │               │
│ - UI          │  RPC   │ - Agent       │
│ - Task Display│        │ - Tool Exec   │
│ - Loading     │        │ - Code Parse  │
└───────────────┘        └───────────────┘

• Rust Backend: Handles agent functionality, tool execution, and API calls
• React/Ink Frontend: Provides a modern, interactive terminal interface with smooth animations

Examples

Here are some example queries to try:

• "Explain the codebase and how to get started"
• "List all files in the project"
• "Summarize the Cargo.toml file"
• "Show me all files that import the 'anyhow' crate"

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.

Acknowledgments

• This project is inspired by Claude Code and similar AI assistants
• Uses Anthropic's Claude 3.7 Sonnet model for optimal agent capabilities
• Backend built with Rust for performance and reliability
• Frontend built with React and Ink for a modern terminal UI experience
• Special thanks to the Rust and React communities for excellent libraries and tools