AYVI

Provider setup

Choose where your request goes.

AYVI never switches silently between local and cloud providers. Check provider and model before sending. Apple Foundation Models was smoke-tested locally with synthetic input; live cloud connections are not verified.

Apple Foundation Models · on device

On supported macOS 26+ systems this provider is preselected on a fresh install when macOS reports the system model as available. It runs locally and needs no cloud disclosure. The app distinguishes unsupported device, Apple Intelligence turned off, model still downloading and available. On older or unavailable systems Mock stays the default. The smaller system model can be less capable at complex debugging than larger models.

Mock · no AI request

The labelled demo provider is the fallback when Apple Foundation Models is unavailable. It returns prepared sample output and sends no selection to a model. Use it to test the UI and flow, not to judge real model quality.

Ollama · local endpoint

Install Ollama from the official Ollama site. When the service runs locally you can pull a model and check AYVI’ defaults:

ollama serve
ollama pull llama3.2

The local server usually listens on http://localhost:11434. In AYVI open Settings → AI → Ollama, check model and base URL and run the connection test. Only localhost, 127.0.0.1 and ::1 count as local. A LAN or remote address is treated as cloud, needs a disclosure and is blocked by “Only local models”.

Anthropic · cloud

  1. Create an API key in your Anthropic account and check cost and retention terms there.
  2. Open Settings → AI, choose Anthropic and enter the key in the provided field.
  3. AYVI stores the key in the macOS keychain. The key never belongs in a selection, question, website or log.
  4. On the first cloud request, read the notice about provider, model and data and confirm deliberately.
  5. Test with harmless, non-confidential text.

No real Anthropic endpoint with credentials has been tested for this version.

OpenAI-compatible APIs · cloud or your own local server

In Settings → AI you configure base URL, model and, if needed, an API key. LM Studio, vLLM or other compatible servers can use other local addresses and models. Follow the provider’s documentation rather than copying sample keys – this website intentionally shows no credentials. Loopback addresses such as localhost are treated as local; a server elsewhere on your network is not automatically safe or local. No live endpoint was verified here.

Before each request

  • Provider and model name are visible in the app.
  • Cloud needs an explicit confirmation the first time that provider is used.
  • “Only local models” blocks cloud providers and allows no fallback.
  • The local secret scanner masks detected keys by default before cloud requests; it is heuristic and can miss secrets.
  • Providers manage their own storage; read their current terms.