Any other questions? Feel free to reach out to us.

What’s the system requirement? Will RecurseChat be available on other Operating Systems?

We support Mac with Apple Silicon on macOS Ventura 13.5 or later.

We recommend RAM of 16GB or more. If your RAM size is 8GB, RecurseChat uses CPU inference for the default models.

There’s no immediate plan of supporting other OS, but if you would like to sign up for updates please subscribe here.

How do I start using RecurseChat?

• Step 1: Download the app.
• Step 2: Once installed, type in the chat box, the app will prompt you to download the model, and start a local AI server automatically.
• Step 3: There’s no step 3.

What AI models does RecurseChat support?

RecurseChat offers versatile support for AI models, including a default suite of models as well as the capability to create extensively customized models.

Default Models

We support Meta Llama 3, Mistral, Llava, WizardCoder Python, as well as OpenAI GPT-3 and GPT-4 models (If you bring your own API key) out of the box. Note that to use GPT-4, you would need to have made $5 payment to your OpenAI account (usage tier 1). We plan to keep default models up-to-date and enhance model / assistant customizability.

Local GGUF Models

We support custom GGUF models with your local GGUF files. You can create any number of local GGUF model and delete GGUF models.

RecurseChat is powered by llama.cpp under the hood. The app should work with most GGUF models that llama.cpp supports. (A couple of exceptions are Gemma and StarCoder2, we are working on supporting them.)

OpenAI Chat Completion Models

In addition, we support chat completion endpoints are compatible with OpenAI API endpoints. One example below is using Ollama with RecurseChat:

While RecurseChat doesn’t need Ollama to operate (It has a built-in llama.cpp executable), You can use local Ollama with RecurseChat through Ollama’s support of OpenAI compatible API.

The steps are:

• Serve an Ollama model with ollama serve
• Create a new OpenAI API model: go to top right of the model tab, and click the “New Model” button, and click “New OpenAI Chat Completion model”.
• To use Ollama model, set base url to http://127.0.0.1:11434/v1 and set model id to an Ollama model id like mistral.

How does RecurseChat compare to other similar applications?

RecurseChat aims to be your personal AI companion. We strive to let you use the power of local AI as your daily driver.

• You should feel at home chatting with local AI like you use any online AI providers. The UX should feel intuitive to starters, but not without customizability for advanced users.
• You should have ownership over your chat history. You should not get vendor-locked in, feel powerless when ChatGPT is down. In RecurseChat you can import your ChatGPT history, search over them, even continue chatting using local AI models. You can export your chat history anytime.
• You should chat with AI securely and privately. We’ve spent time meticulously sandboxing the app. If you use local AI models, the chats never leave your local machine. If you use OpenAI GPT models, your data would not be used for training. (Note that it’s however subject to OpenAI’s privacy policy.)

How private and secure is RecurseChat?

We do not collect any usage or telemetry within the macOS app. You can learn more about our privacy policy here.

The macOS app runs inside Mac App Sandbox. It’s a mandatory requirement to distribute in Mac App Store. (For apps distributed outside of Mac App Store, there’s no such requirements.) Apps running under Mac App Sandbox would only have access to user selected files and network access through entitlements, unauthorized access is blocked.