I made a non-linear AI Chat Client for OpenAI and local Ollama models.

The concept is simple – to embrace a visual approach to organization and leverage every feature AI providers have to offer to achieve the right results for your particular workflow.

Instead of a linear chat interface, grafychat uses a canvas-based approach, where you create nodes (or "Chats") – simple editable text documents that serve as a pair of request-response, and visually connect/disconnect them to manage conversations.

These nodes are highly customizable, allowing you to change chat models and temperature on the go, apply custom instructions, hide or collapse certain areas, change node and font size, add color and labels.

You can have separate canvases (or "Contexts") for different workflows with its own defaults, and effortlessly navigate to specific contexts, conversations, or chats using full-text search functionality.

Need a classic chat? Simply switch to a linear chat view by picking a node or start chatting the Inspector. The Inspector also lets you export the current conversation to Markdown file.

Grafychat is designed as a local-first software, meaning your data remains completely private, stored locally on your device using IndexedDB. Data only leaves your device for non-local LLM requests, such as OpenAI.

For OpenAI users, you can access not only Chat models (GPT4, GPT-4 turbo etc.) but also voice input, speech, and image generation capabilities - all with your own API key. No active ChatGPT subscription is needed.

And for Ollama Users, enjoy free-to-use, locally running powerful models like Llama 3.

More AI providers coming soon.

Hope you find it useful! Try it out and I'd love to hear your thoughts!