Choosing an engine: Local vs Cloud
Abe turns speech into text with one of two engines. You choose during setup and can switch any time in Settings.
| Local (offline) | Cloud (your own API key) | |
|---|---|---|
| Where audio goes | Nowhere. It stays on your computer. | Straight from your computer to the provider you chose |
| Cost | Free | A few cents per hour of talking, billed by your provider (Costs) |
| Setup | One-time model download (75 MB to 1.6 GB) | Create an API key and paste it in |
| Speed | Depends on your hardware | Usually around a second, regardless of your hardware |
| Accuracy | Good to excellent, depending on model size | Excellent |
| Works offline | Yes | No |
| Voice Transform | Needs a transform provider (cloud key, or Ollama on your computer) | Yes, with the same key |
Pick Local if…
Section titled “Pick Local if…”- You want your audio to never leave your machine.
- You have a reasonably modern computer: Apple Silicon Mac, or a PC with a recent CPU (a GPU helps a lot for the larger models).
- You don’t want to deal with API keys or bills.
Start with Base (the default) or Small, then try Large v3 Turbo if your computer keeps up. See Local Whisper models.
Pick Cloud if…
Section titled “Pick Cloud if…”- You’re on an older or low-power laptop where local models feel slow.
- You want the best accuracy for accents, jargon or noisy rooms without waiting.
- You’re fine with your audio being processed by a provider such as OpenAI or Groq under their data policy.
We recommend Groq: it’s the cheapest, very fast, and has a free plan that doesn’t need a credit card. OpenAI offers several models, from budget to highest accuracy. You can also point Abe at any OpenAI-compatible endpoint, including one you run yourself.