This project provides an AI stack integrated directly into text editors, enabling features like voice-to-text, local LLM interaction, and document chat. It targets developers and power users seeking to augment their workflow with AI capabilities within their preferred coding environment, offering a portable and code-centric approach to LLM interaction.

How It Works

UniteAI functions as a Language Server Protocol (LSP) server, allowing it to integrate with any editor supporting the LSP standard. This architecture decouples AI logic from editor-specific code, promoting portability and ease of contribution. Core features leverage Python libraries for transcription, local LLM management (via a separate uniteai_llm server), and Retrieval Augmented Generation (RAG) for document querying.

Quick Start & Requirements

• Install: pip3 install --user "uniteai[all]"
• Prerequisites: Python 3, Pip. portaudio is required for speech-to-text (install via brew install portaudio on macOS, sudo apt install portaudio19-dev on Ubuntu/Debian). Ensure your Python bin directory is in your system PATH.
• Optional LLM Server: Run uniteai_llm to connect to local models.
• Editor Integration: Install the UniteAI extension for VSCode or configure lsp-mode for Emacs. VSCodium users may need to install a .vsix file.
• Configuration: A .uniteai.yaml file is generated upon first use, requiring user edits.
• Docs: screencast.webm, screencast_document_chat.webm

Highlighted Details

• Supports RAG for querying YouTube transcripts, Arxiv papers, PDFs, and Git repositories.
• Offers semantic search capabilities for document analysis.
• Includes keybindings for common actions like document search and LLM interaction.
• Designed for contributor-friendliness with a clear contribution model and well-documented code.

Maintenance & Community

The project is actively maintained, with recent updates in January 2025 noting compatibility with DeepSeek R1. Contributions are encouraged via issues and pull requests, with a structured approach outlined in the .todo/ directory.

Licensing & Compatibility

Licensed under the Apache-2.0 license. This license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

Initial setup for RAG on long documents may take a few minutes for embedding generation, though this is cached for subsequent use. Default keybindings may conflict with existing editor shortcuts on certain operating systems.