Choosing a backend

The options

Backend	Runs	Needs	Best for
None	—	Nothing	The default. Raw transcript, zero network.
Ollama	Your Mac	Ollama installed locally	Cleanup with no network at all.
OpenRouter	Cloud	OpenRouter key	One key that reaches many models.
Anthropic	Cloud	Anthropic key	Claude models directly.
OpenAI	Cloud	OpenAI key	GPT models directly.
Groq	Cloud	Groq key	Very low latency.

All cloud options are bring-your-own-key: you contract directly with the provider and pay them for tokens. OpenVoiceFlow takes no cut and has no server in the path.

Local cleanup with Ollama

Ollama runs a language model on your own Mac, so cleanup keeps the entire pipeline offline.

1. Install Ollama.
2. Pull a small instruct model — a 3B-class model is plenty for tidying text.
3. In Settings ▸ AI cleanup, choose Ollama.

OpenVoiceFlow talks to Ollama at localhost:11434, and allows a longer timeout (120 seconds) than for cloud providers, since local generation on a busy machine can be slower. If Ollama is not running, cleanup fails open and you get the raw transcript.

Cloud providers

Each backend uses that provider's standard API endpoint with a sensible default model. You can override the model per backend with the model override field in Settings if you prefer a specific one — leave it empty to use the default.

Since cleanup is a short, simple rewriting task, the cheapest small model from any provider generally does it well. Reaching for a frontier model here mostly buys you latency.

What to expect on speed

Cleanup adds a step between releasing the key and seeing text. Cloud providers are typically fast enough to feel immediate, with Groq marketed specifically on latency. Local Ollama depends entirely on your Mac and chosen model — a small model on Apple Silicon is comfortable; a large one on a busy Intel Mac is not.

If cleanup ever feels slow, remember the 10-second cloud timeout: past that, the app gives you the raw transcript rather than making you wait.

Switching backends

Change it any time from Settings ▸ AI cleanup or the menu-bar Cleanup submenu. Each backend keeps its own key in the Keychain, so switching back and forth does not require re-entering anything. Next steps: API keys & costs.