Tangent is a canvas for exploring AI conversations, enabling users to branch, merge, and resurrect chat threads. It targets AI enthusiasts and researchers seeking a more visual and experimental approach to interacting with LLMs, moving beyond traditional chat interfaces.

How It Works

Tangent operates as an offline-first application, primarily leveraging local models via Ollama for both embeddings and LLM generation. It processes exported chat data from platforms like Claude and ChatGPT, allowing users to create conversational branches, resume interrupted threads, and cluster discussions by inferred topics. The architecture includes a modular backend API for managing chats, messages, states, and topics, with services for background processing, clustering, and embedding generation.

Quick Start & Requirements

• Install: Clone the repository and run ./install.sh.
• Prerequisites: Ollama (v0.4.4), Whisper.cpp (optional for voice features), exported chat data (Claude/ChatGPT).
• Models: Requires all-minilm for embeddings and qwen2.5-coder:7b for generation (or custom alternatives).
• Setup: The install.sh script handles Ollama setup, Whisper.cpp installation, Python environment setup, and frontend startup.
• Links: Discord Server

Highlighted Details

• Offline-first architecture using local LLMs via Ollama.
• Visual canvas for AI conversations, supporting branching and merging.
• Topic clustering for organizing and filtering discussions.
• Resurrects conversations hitting context limits.
• Supports Claude and ChatGPT data exports.

Maintenance & Community

• Active community with a Discord server for support and discussion.
• Open to contributions via pull requests and issues.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

The project is described as "kinda hardcoded for ollama" but with plans for generalization. While Whisper.cpp is optional, its setup is detailed as a prerequisite for voice features. The frontend setup may require manual installation of missing packages.