Connecting your own providersElite

Pick a provider based on what you need. SingularityApp supports several AI providers, and each has its own strengths:

• OpenAI — provider with GPT-4, GPT-4o, and other models. Requires paid API keys. Functionality may be limited depending on your region. Available in the web version and the desktop app.
• Anthropic — provider with Claude 3, Claude 3.5, and other models. Requires a paid API key. Functionality may be limited depending on your region. Available in the web version and the desktop app.
• Ollama — for running local models (LLama, Mistral, Phi). Available in the web version and the desktop app.

Access to some providers is subject to regional restrictions. If you run into connection problems, make sure your network connection meets the requirements for the provider’s region. Connecting by API key works under the same conditions as using that provider’s service from a browser.

When you pick a provider and a model, it helps to understand their capabilities and limits, since different models are built for different tasks. Read the model’s specs carefully in the provider’s official documentation. Pay attention to these key specs:

• Context Window — the maximum context size the model can process (measured in tokens). A large context lets you work with longer texts.
• Reasoning — support for extended reasoning. Matters for complex analysis and logic tasks.
• Tool Calling — the model’s ability to call functions and tools to extend what it can do.

Connecting a provider is straightforward, but it takes a few things ready in advance. Each provider needs specific data to authorize and set up the connection. To connect an external provider, prepare:

• API key — the access key for the provider’s API. Get it in your provider account.
• Base URL — the provider’s API address. For most providers, it’s pre-filled by default in the matching connection field in SingularityApp.

OpenAI

1. Go to openai.com/index/openai-api/ and sign in.
2. In the top-right corner, choose “Start building”.
3. Create an organization and give it a name.
4. Skip adding other users (or invite them if needed).
5. Enter a name and create an API key.
6. Copy the API key you get and save it.
7. Skip or top up your balance right away if needed.
8. Open the SingularityApp app (Settings → AI assistant → Custom models → OpenAI).
9. Enter the API key in the “API Key” field and save.

Anthropic (Claude)

1. Go to platform.claude.com/dashboard and sign in.
2. Choose “Get API Key”.
3. Click “Create Key”.
4. Choose a workspace (if you need a different one) and enter a name for the key.
5. Copy the API key you get and save it.
6. Open SingularityApp (Settings → AI assistant → Custom models → Anthropic).
7. Enter the API key in the “API Key” field and save.

Ollama

Ollama works with both local models and ollama-cloud. The connection steps are the same for both.

1. Go to ollama.com and download the Ollama app.
2. Install Ollama on your computer.
3. Start Ollama, either through the interface (via the app shortcut) or by running ollama serve in the terminal.
4. Download the model you want from ollama.com/library. Pick the model you need from the available catalog.
5. To add a model, run ollama run MODEL_NAME in the terminal (for example, ollama run llama2).
6. Open the SingularityApp app (Settings → AI assistant → Custom models → Ollama).
7. Enter the URL you get in the “Base Url” field (if Ollama is running at the same time, via ollama serve or the app, just click save).

1. Once a provider is connected, every model the provider makes available for your API key shows up automatically in the model list under Settings → AI. The list includes all models the provider exposes for your API key. They can be of different types, both for working with text and for other kinds of content.
2. Turn on the toggle for the provider’s models you want to use in the assistant chat. Once activated, the models appear in the list of available models in the chat.

Note: The AI assistant is built for text models. If you enable a model that isn’t meant for text, the request won’t be processed correctly.

3. Now you can pick the provider’s models in the chat. The models you activated in settings appear under their provider heading in the model list when you make a selection.

If you get a 403 error when connecting Ollama, set the environment variable OLLAMA_ORIGINS="*".

Linux

Temporary (current terminal session only):

1. Open the terminal.
2. Enter: export OLLAMA_ORIGINS="*".
3. Run ollama serve.

Permanent (the variable is set automatically):

For bash:

1. Open the terminal.
2. Enter: echo 'export OLLAMA_ORIGINS="*"' >> ~/.bashrc.
3. Enter: source ~/.bashrc.
4. Run ollama serve.

For zsh:

1. Open the terminal.
2. Enter: echo 'export OLLAMA_ORIGINS="*"' >> ~/.zshrc.
3. Enter: source ~/.zshrc.
4. Run ollama serve.

macOS

Temporary (current terminal session only):

1. Open the terminal.
2. Enter: export OLLAMA_ORIGINS="*".
3. Run ollama serve.

Permanent (macOS uses zsh by default):

1. Open the terminal.
2. Enter: echo 'export OLLAMA_ORIGINS="*"' >> ~/.zshrc.
3. Enter: source ~/.zshrc.
4. Run ollama serve.

Windows (PowerShell)

Temporary (current PowerShell only):

1. Open PowerShell.
2. Enter: $env:OLLAMA_ORIGINS="*".
3. Run ollama serve.

Permanent (saved to the user environment variables):

1. Open PowerShell.
2. Enter: setx OLLAMA_ORIGINS "*".
3. Close PowerShell.
4. Run ollama serve.