Configuring AI

Octarine can use cloud APIs, local model servers, and supported assistant accounts for the Writing Assistant and Ask Octarine.

Accounts and API Providers

The AI section in Settings separates two kinds of connections:

Settings page	Use it for
AI → Accounts	Local Codex, Claude Code, and OpenCode CLI sessions. Their official CLIs manage authentication, so credentials do not enter Octarine.
AI → API Providers	Cloud API keys, Ollama, LM Studio, and custom OpenAI- or Anthropic-compatible endpoints.

Account and provider choices are configured per workspace. You can copy them when creating a workspace with Use settings from an existing workspace.

Connect A CLI Account

Open Settings → AI → Accounts → Codex. Octarine uses the local Codex CLI session, so it does not ask you to paste a separate key. See Working with Codex for installation, sign-in, and troubleshooting.

The Accounts page also supports:

• Claude Code — install the Claude Code CLI and sign in with claude auth login. See Working with Claude Code.
• OpenCode — install OpenCode and connect one or more providers with opencode auth login. See Working with OpenCode.

Select an account to check whether its CLI is installed and authenticated. If it is not signed in, choose Open … login to launch the official login flow in your system terminal, then return to Octarine and click Refresh status. Connected tool-capable models appear in the Ask Octarine model selector, and the inline Writing Assistant uses the workspace's current AI model.

Octarine runs Claude Code and OpenCode requests in an isolated temporary environment. It does not load your CLI's project instructions, plugins, built-in tools, or saved sessions. Ask Octarine can still use the Octarine tools made available for the current request, subject to its normal workspace context and access controls.

Configure a Built-In API Provider

Open Settings → AI → API Providers, select a provider, enter the required API key or local server URL, and activate the connection. API-key providers validate the key and show a checkmark once connected.

Built-in API Providers include:

• OpenAI
• Anthropic
• DeepSeek
• Google Gemini
• Groq
• Mistral AI
• Perplexity
• Together AI
• xAI
• Ollama
• LM Studio

Ollama and LM Studio use a local server URL instead of a cloud API key. Follow the dedicated Ollama and LM Studio guides to start those servers.

Custom Providers

Choose New Custom Provider on the API Providers page when your service exposes an OpenAI- or Anthropic-compatible API.

1. Choose the OpenAI or Anthropic API format.
2. Give the provider a name.
3. Enter its base URL without a trailing slash.
4. Add an API key. A key is optional only when the URL begins with http://; HTTPS endpoints require one.
5. Save the provider. Octarine checks its models endpoint before adding it.

For the OpenAI format, enter the versioned API base that sits directly before /models and /chat/completions, such as https://openrouter.ai/api/v1. For the Anthropic format, enter the root before /v1/models and /v1/messages, such as https://api.anthropic.com.

Saved endpoints appear under Custom. Select one there to edit its name, URL, format, or key, or to delete it.

Selecting a Model

After a connection is configured, its tool-capable models appear in the Ask Octarine model selector. The selected provider, model, and reasoning choice are also used by internal AI features such as the inline Writing Assistant.

• Use the provider icons across the top of the picker to switch accounts or providers, then search that provider's models.
• You can change models between prompts or during a conversation.
• When the selected model exposes reasoning controls, use the bar icon beside the model name to choose its supported effort level or turn reasoning on or off.
• Octarine remembers the selected provider, model, and supported reasoning choice in the workspace settings. When a new model does not support the saved choice, Octarine selects the lowest effort that model offers.
• Capabilities such as tools and web search depend on the selected model and provider.

Skills and Slash Commands

AI Settings includes two ways to reuse instructions:

• Skills are Markdown instruction files that you explicitly attach to an Ask Octarine prompt.
• Slash Commands insert reusable prompt text in the Writing Assistant.

Privacy and Storage

API keys, provider URLs, and selected models are stored in the device-specific Octarine settings file. They are outside the workspace and are not synced by Git or cloud-folder sync. Codex, Claude Code, and OpenCode credentials remain in their respective CLI stores rather than Octarine. Custom Skills and Slash Commands are stored inside .octarine in the workspace and can sync with it.

When you use a cloud provider, Octarine sends the prompt and included note context to that provider. Use Ollama, LM Studio, or another local endpoint if you want model requests to remain on your own computer.

Quick Answers