Ask AI

Chat with an AI assistant directly inside the Command Panel. Ask questions, search your sidebar, open links, and manage your workspace using natural language.

How to access

Open the Command Panel with Cmd Ctrl K, then activate AI mode in one of three ways:

Ctrl Return - Activates AI mode. If you have a search query with no results, it pre-fills that query.
“Ask AI” row - When a search returns no results, an “Ask AI” row appears in the results list. Press Return to activate it.

What the AI can do

The AI assistant can interact with your sidebar through natural language. It has access to these capabilities:

Category	Actions
Search and browse	Search saved links, list spaces, folders, tags, recent items, live tabs
Open and navigate	Open links in any browser, switch spaces, perform web searches
Create and organize	Save new links, create folders and spaces, move links between spaces
Settings	View current settings, shortcuts, and visibility options, open preferences
Sidebar control	Toggle sidebar visibility

Some actions (like opening links or creating folders) may ask for your permission before executing. You can choose to deny, allow once, or always allow each action type.

Slash commands

Type / in the AI input field to see all available commands:

Command	Description
`/guide`	Shows a summary of everything the AI can help with
`/model`	Opens the model picker to switch AI models
`/clear`	Clears the current chat history
`/help`	Lists all available slash commands

AI providers and models

Configure your AI provider in Preferences - AI under the “Features” section for “AI Chat.”

Provider	Models	Requires
Anthropic	Claude 4.5 Haiku, Claude 4.6 Sonnet, Claude 4.6 Opus	API key
OpenAI	GPT-5 Nano, GPT-5 Mini, GPT-5, GPT-5.2, GPT-5.2 Pro	API key
Google Gemini	Gemini 2.5 Flash, Gemini 2.5 Pro, and more	API key
OpenAI-Compatible	Any model the server offers	Base URL, optional API key
Ollama	Any locally pulled model	Ollama running locally
MLX	Gemma 2 2B, Llama 3.2 3B, Phi-4 Mini 3.8B, Qwen3 4B, Mistral 7B, Qwen 2.5 7B	Apple Silicon Mac, model downloaded

Use the /model slash command to switch between models for the current provider without leaving the chat.

OpenAI-Compatible provider

Many AI services and local servers use the same API format as OpenAI. The OpenAI-Compatible provider lets you connect to any of them - just provide a base URL and you’re ready to go.

Works with: Groq, Together AI, LM Studio, vLLM, Mistral, Fireworks, oMLX and other local MLX servers, and any other service that exposes an OpenAI-compatible /chat/completions endpoint.

Setup:

Go to Preferences - AI
Under the feature you want to configure (e.g. AI Chat), select OpenAI-Compatible as the provider
Enter the Base URL of your server, pointing at its /v1 root (e.g. http://localhost:8000/v1 for oMLX, http://localhost:1234/v1 for LM Studio, or https://api.groq.com/openai/v1 for Groq)
Enter an API key only if the service requires one. Local servers (oMLX, LM Studio, vLLM) usually don’t - just leave it blank
Click Save & Test. SupaSidebar checks the connection by listing models at <base URL>/models, so make sure your server is running first
Pick the model from the dropdown. After a successful test, SupaSidebar lists the models your server reports - choose a chat/instruct model (e.g. a Qwen Instruct model for oMLX). Embedding-only models such as all-MiniLM can’t chat. If the list is empty (server unreachable), you can still type the model name manually.

SupaSidebar sends chat requests to <base URL>/chat/completions in the standard OpenAI format. Tool calling (letting the AI search and open your links) is sent to the server too, but it only works if the underlying model supports OpenAI-style function calling - many local models don’t, in which case the AI still replies normally with text.

Ollama

Ollama runs open models locally on your Mac. SupaSidebar talks to it directly - no API key, and nothing leaves your machine.

Setup:

Install Ollama and pull a model, e.g. ollama pull qwen2.5
In Preferences - AI, select Ollama as the provider for AI Chat
The Endpoint defaults to http://localhost:11434. Change it only if Ollama runs on a custom port or another machine, then click Check to confirm the connection
Pick your model from the dropdown

Tool calling: Models that support function calling (Qwen 2.5, Llama 3.1+, and similar) can use SupaSidebar’s tools to search and open your links. Older models that don’t support tools still chat normally - SupaSidebar falls back to a plain reply automatically.

Keyboard shortcuts

Shortcut	Action
`Ctrl Return`	Activate AI mode from Command Panel
`Escape`	Exit AI mode and return to search
`/`	Show slash command menu

Copy a message

Hover over any AI or user message in the chat to reveal a copy button in the corner. Click it to copy the message text to your clipboard.

Tips

The AI remembers context within a conversation. Ask follow-up questions without repeating yourself.
Use /guide when you first try Ask AI to see everything it can help with.
Local models (MLX, Ollama) keep your conversations completely private - nothing leaves your Mac.
Combine Ask AI with Spaces to quickly organize links across workspaces using natural language.
See Command Panel for other search and navigation features.