Overview

How Roofle uses a model

Transcription happens on your Mac and needs no model. Roofle uses an AI model for: live notes, suggested questions and follow-up actions. Transcript text is sent straight to the provider you choose. Roofle has no server in between.

Any gateway. If it exposes an OpenAI-compatible endpoint, paste its base URL and it works.
Any model. Roofle does not lock you to a list. Type the model name your provider expects.
Local models work too. Point Roofle at a model on your own machine and nothing leaves it.

Setup

The three settings

Open Settings, then LLM Configuration, and fill in three fields.

Field	What to enter
Base URL	The gateway endpoint, usually ending in `/v1`. See the table below.
Model	The model name exactly as your provider spells it, for example `gpt-4o-mini`.
API key	Your key for that provider. Use Show to check what you typed.

Then run the connection test. It confirms the URL, model and key work before your first call.

Reference

Common gateways and base URLs

Copy the base URL for your provider. Providers change endpoints from time to time, so check their docs if a test fails.

Gateway	Base URL	Key
OpenAI	`https://api.openai.com/v1`	OpenAI API key
OpenRouter	`https://openrouter.ai/api/v1`	OpenRouter key
Groq	`https://api.groq.com/openai/v1`	Groq key
Together AI	`https://api.together.xyz/v1`	Together key
Fireworks AI	`https://api.fireworks.ai/inference/v1`	Fireworks key
Mistral	`https://api.mistral.ai/v1`	Mistral key
DeepSeek	`https://api.deepseek.com`	DeepSeek key
Google Gemini	`https://generativelanguage.googleapis.com/v1beta/openai/`	Gemini API key
LiteLLM proxy	`http://localhost:4000`	Your proxy key
Ollama (local)	`http://localhost:11434/v1`	None needed
LM Studio (local)	`http://localhost:1234/v1`	None needed
llama.cpp server (local)	`http://localhost:8080/v1`	None needed

Using a company gateway or another provider not listed? Ask for its OpenAI-compatible base URL and use that.

Example

Using OpenAI

Create a key at platform.openai.com/api-keys, set the base URL to https://api.openai.com/v1, pick a model and paste the key. Keep the key private, and if it leaks, delete it on the OpenAI dashboard and create a new one.

Example

Using a local model

A local model keeps everything on your Mac. Transcription is already on-device, so with a local model no meeting content leaves your machine.

With Ollama

brew install ollama
ollama pull llama3.1
ollama serve

Then in Roofle, set Base URL to http://localhost:11434/v1 and Model to llama3.1. Local servers do not check the key. If the field will not stay empty, enter any placeholder such as local.

With LM Studio or llama.cpp

Start the local server in the app, copy the address it shows (see the table above), and use the model name it lists.

Recommendation

Which model to choose

Roofle makes frequent, short requests during a call, so speed and cost matter more than the largest model. A small, fast model is the right starting point.

Goal	Recommended	Why
Default for most people	`gpt-4o-mini`	Fast, inexpensive and good enough for notes and questions.
Higher quality notes	A larger model from your provider	Better judgement on subtle gaps, at higher cost and some added delay.
Lowest latency	A small model on Groq or similar	Questions appear sooner after the words are spoken.
Fully private	A local model such as `llama3.1`	Nothing leaves your Mac. Use a recent Apple Silicon Mac.

Start with the default, then change only if you notice a problem. You can switch models at any time, and the connection test tells you straight away if the name is wrong.

Help

If the connection test fails

Symptom	What to check
Unauthorized or invalid key	Re-paste the key with Show on. Make sure it belongs to the gateway in the Base URL.
Model not found	Copy the model name from your provider exactly. Names differ between gateways.
Cannot connect	Check the Base URL, including `/v1`. For local models, confirm the server is running.
Works, but notes are slow	Use a smaller or faster model, or raise Notes update interval under Live session.