noAI (nodleAI) agent

noAI is a lightweight macOS app that turns Ollama-powered language models into a coding and creativity companion.
Think of it as a local-first AI agent where you can spin up different models (Llama, Gemma, code-specialized LLMs, etc.) and use them for tasks like:

• Coding help — write, debug, or refactor code with context-aware suggestions
• Model switching — easily swap between multiple Ollama models depending on your use case
• Creativity and brainstorming — not just for coding, but also writing, ideation, or general Q&A
• Local-first privacy — runs on your machine via Ollama, so your prompts and data stay private

It’s meant to feel like a “coding buddy” with a bit of personality — quick, local, and customizable completely offline.

Features

• Multiple Model Support: Run and swap between LLMs like Llama, Gemma, DeepSeek LLMs directly on macOS
• LoRA Fine-tuning: Load lightweight adapters to specialize a model on your own data
• RAG (Retrieval-Augmented Generation): Connect your agent to local files, docs, or databases for grounded answers
• MCP (Model Context Protocol): Extend your agent with tools, APIs, or external systems for custom workflows
• Thread Management: Create, delete, rename, and organize chat threads
• Export Options: Save chats as JSON or Markdown for later reference

Getting Started

Prerequisites

1. Install Ollama
   Follow the official installation guide: https://github.com/ollama/ollama

   On macOS, you can use Homebrew:

   brew install ollama

2. Run the Ollama server
   Start the Ollama daemon (it runs the models locally):

   ollama serve

   By default it listens on http://127.0.0.1:11434.

3. Pull a model
   Example pulling the Llama 3 model:

   ollama pull llama3

   You can also pull other models, e.g. gemma, codellama, or any from Ollama's model library.

---

Optional: Enable Image Generation (Stable Diffusion via Core ML)

noAI can also generate images in dedicated “Image Generation” threads using Apple’s Core ML Stable Diffusion models.

1. Download a Core ML Stable Diffusion model

Go to Apple’s Core ML Stable Diffusion collection on Hugging Face:

• https://huggingface.co/collections/apple/core-ml-stable-diffusion

Pick a model you like (e.g. Stable Diffusion 1.5 or 2.1) and download / clone it.

For each model, you’ll end up with a folder structure similar to:

coreml-stable-diffusion-*-*/        # model root
  original/
    compiled/   <-- ✅ this is the folder you need
    packages/
  split_einsum/
    compiled/
    packages/

For noAI you only need the original/compiled folder for the model you want to use.

• ✅ Required: original/compiled/ (this is what the app points at)
• ❌ Optional to keep/remove: packages/ folders, or the unused variant (split_einsum) if you want to save disk space

2. Point noAI at the model folder

1. Open noAI.
2. In the sidebar, expand Settings.
3. In the Image generation section:
   • Click “Choose folder…”.
   • Select the original/compiled folder of the Core ML SD model you downloaded.
     (For example: .../coreml-stable-diffusion-v1-5/coreml-stable-diffusion-v1-5/original/compiled.)
4. The status should change to something like “Stable Diffusion ready”.

The app will remember this folder using a security-scoped bookmark, so you don’t need to reselect it every launch.

3. Use image generation in the UI

• In the sidebar header, click New ▾ → New image generation.
• This creates a special image thread:
  • The composer shows an “Image generation” pill (no model selector).
  • Every prompt you send will:
    • Add your text message.
    • Generate an image locally with Stable Diffusion.
    • Display the image as a bubble in the chat.

Your local LLM chat threads and image-generation threads are completely separate, so the user doesn’t have to think about which LLM model is “linked” to the image generation — it always uses the selected Core ML Stable Diffusion model.

---

Optional: Use Ollama Cloud Models

Ollama Cloud models let you run large models (*:cloud) on Ollama’s hosted hardware while still using your local noAI app.

1. Sign in to Ollama Cloud

From your terminal:

ollama signin

Follow the instructions to sign in or create an account in your browser.

2. Run or pull a cloud model

Once signed in, you can pull a cloud model directly:

ollama pull gpt-oss:120b-cloud

Cloud models use the :cloud suffix (for example, gpt-oss:120b-cloud). They appear in your model list just like local models, but when you run them, the heavy lifting happens on Ollama’s servers.

---

Installation

---

You have two options to get started:

Option 1: Download Prebuilt App

• Go to the Releases page.
• Download the latest .dmg.
• Drag noAI.app into your Applications folder.
• On first launch, macOS may warn you since the app isn’t signed. Right-click → Open to approve.

Option 2: Build from Source

1. Clone this repository:

   git clone https://github.com/NodleCode/noai
   cd noai

2. Open the Xcode project and build the app:

   open noai.xcodeproj

3. Run the app in Xcode, or build and install it on your system.

The app will automatically try to connect to http://127.0.0.1:11434 (you can adjust this in settings).

Usage Examples

1. Basic Chat

Start a new thread and chat with a model. Messages stream in real-time.

1.1 Switching models

Swap between Ollama models like Llama, Gemma, or Code LLMs.

---

2. Create, Rename, Delete Threads

Organize your work with multiple chat threads. Easily rename or delete them.

---

3. Export Conversations

Save any thread as JSON or Markdown for later use. Export all threads in one go.

---

Contributions

• Open an issue if you find a bug or have a feature request.
• Submit a pull request if you’d like to add or improve functionality.

Related Projects

• Ollama – local-first LLM runtime

---

License

MIT License – see LICENSE for details.