# BF51.3 — Ollama Cloud Support

## Execution model

TeXise connects from the browser to the Ollama service on the user's computer. The TeXise server does not install, run or proxy Ollama.

Ollama models are classified from `/api/tags` and `/api/show`:

- **Local Ollama** — locally installed model; allowed with Privacy Mode on or off.
- **Ollama Cloud** — `remote_model` / `remote_host` indicates cloud-backed execution; allowed only with Privacy Mode off.

The selected model is revalidated before configured inference. AI Settings and AI Assistant label the execution class explicitly.

## Privacy

Privacy Mode blocks Ollama Cloud and BYOK AI cloud execution. It permits browser Local AI and verified Local Ollama. Automatic BYOK fallback after an Ollama outage remains disabled while Privacy Mode is on.

## Coverage

BF51.3 extends AI Suggest/readiness, AI Review, review checkpoints/provenance, Paper Memory documentation, comparison claims and the Ollama browser E2E. Review reports record when manuscript context was sent to an Ollama Cloud-backed model.