Instrukt is a terminal-based AI environment for building, testing, and instructing modular AI agents. It targets developers and power users seeking an integrated, keyboard-centric workflow for AI-powered tasks like code assistance and document Q&A, offering a secure, sandboxed execution model.

How It Works

Instrukt leverages Langchain for AI orchestration and Textual for a rich terminal UI. Users create agents as Python packages, attach tools and document indexes (managed by Chroma), and instruct them via natural language. Agents run within Docker containers for isolation, with an option for gVisor for enhanced security. This approach provides a flexible, extensible platform for custom AI workflows directly within the terminal.

Quick Start & Requirements

• Install: pip install instrukt[all]
• Requirements: OpenAI API key, TERM=xterm-256color if color issues arise. Docker with CUDA support is recommended for headless/containerized use.
• Setup: Minimal, but requires API key configuration.
• Docs: https://github.com/blob42/Instrukt

Highlighted Details

• Keyboard-first terminal interface with rich output via Textual.
• Modular agent and tool architecture, shareable as Python packages.
• Document indexing with auto-detection of programming languages for optimized chunking.
• Integrated REPL-prompt console for rapid agent interaction and testing.
• Secure agent execution within Docker containers (early preview for Patreon supporters).

Maintenance & Community

• Active development by a single author.
• Discord server available for updates and support.
• Roadmap includes local LLM and embedding support.

Licensing & Compatibility

• AGPL license. Requires source code availability for any publicly accessible instance or modified distribution. Commercial use is permitted but requires adherence to the copyleft provisions.

Limitations & Caveats

The Docker agent functionality is an early preview exclusively for Patreon supporters. Currently, only OpenAI models are supported, with local model integration planned as a future milestone. The project is explicitly noted as a work in progress, with potential bugs and API changes.