Local models
With local processing, the request stays on the device or in the locally configured runtime. The setup flow can install Ollama and the configured Qwen model; internet access is needed for the initial download.
For teams · Privacy & operations
Data flows you can follow.
The selected mode determines where a request is processed. AYVI does not operate its own inference or model-training service.
18Only the selected excerpt is prepared.
19Local model
AYVIProcessing pathLocally on device or with the configured provider.
Example result− Local model+ Configured cloud provider
Processing paths
Local or with a consciously selected provider
AYVI does not send code to an AYVI-operated endpoint. Cloud inference sends the prompt and selection to the provider configured in the app.
Your provider
When a cloud provider is enabled, its terms, retention and region apply. Check whether the content may be sent to that provider before using it.
No AYVI training
AYVI does not operate its own training service for customer code. Any model use by an external provider is governed by that provider's terms.
Controls
The developer keeps the decision
Selection, not a whole project
AYVI reads the excerpt selected by the user. It does not analyse a full repository.
Diff before replacement
Standard replacements show a visible diff and wait for the user's decision. The separate Direct Improve option can be turned off in Settings.
Permissions per app
Accessibility permission is required to access selected text. Support and replacement depend on the target app.
Operational status
No central team management yet
There is currently no admin console, team management, SSO or central deployment. No public installer is approved; macOS is beta and Windows is in progress.