Claude AI in Snipset: Optional, Per Feature, Off by Default

Snipset runs its AI on your own computer first, through Ollama and the local models you choose. Since the October 15 releases for Windows and Linux, you can also send specific features to Claude, Anthropic's cloud AI, through Snipset's server. Claude in Snipset is optional, you turn it on one feature at a time, and every switch starts off.

We explained why local processing matters in our earlier post, What Does "Local AI" Actually Mean for Your Privacy? This post covers the other side: what changes when you pick the cloud on purpose, and what does not.

Which Snipset Features Can Use Claude

Open Settings, go to AI & Models, and find the Claude AI (cloud) group. It has eleven switches, one per feature:

Switch	Switch
AI chatbot	Ask Anything
Chat page	Telegram voice notes
Note AI	Smart shorthand
Snippet titles and tags	Inline AI
Dictation polish	Daily journal
Translation

That list is the whole list. Snipset's code has one entry per switch, and a feature that is not on it has no path to Claude. In the chat overlay and on the Chat page, Claude's reply streams in as it is written.

Every Switch Starts Off

A feature only uses Claude when its own setting is stored as on. A fresh install has no such setting, so all eleven start off, and a missing or damaged value also counts as off. Turning on Translation does not turn on Note AI or anything else.

When at least one switch is on, Snipset shows a notice that names the features that now answer with Claude first. You can always see which parts of the app use the cloud, and switch any of them back.

What Goes to Claude and What Stays Local

When a switch is on, the text that feature needs (and, in the case below, a screenshot) goes to Anthropic's Claude through our server. The Settings group says so in plain words: Anthropic deletes that text within 30 days and does not use it for training by default. That matches Anthropic's published policy for API traffic, which lists a few exceptions.

A few details we built in on purpose:

• You do not need an Anthropic account or API key. The app talks to Snipset's server, and our server talks to Claude.
• Your device id, license code, and IP address stay with us. Our server uses them to check your licence and apply daily limits, and does not pass them to Anthropic.
• Claude never hears your voice. Dictation and Telegram voice notes are transcribed by your speech engine first (the local Whisper model, unless you chose Gemini Cloud). With Dictation polish or Telegram voice notes switched on, Claude cleans up the transcript, not the audio.
• Screenshots only with Visual Screen Context. If you turn on Visual Screen Context, Dictation polish and Translation can also send a screenshot of the active window along with the text, and Ask Anything a capture of the full screen. With it off, no screenshot goes along.
• Switched-off features stay local. If a switch is off, that feature goes straight to your local model through Ollama and never contacts our server for Claude.

When Claude Is Unavailable

If Claude cannot answer, the feature does not fail. It falls back to your local model and finishes the job there. Snipset tells you once per feature per session, in the app ("Claude is unavailable right now; Translation used your local model") and as a system notification, unless you turned desktop notifications off. If a streamed chat reply is cut halfway, you keep the part that was already written.

This is why the conversation model under AI & Models still matters with Claude on. It is the fallback for the features you sent to Claude, and it keeps answering every feature you left off.

Where Claude Works in Snipset Today

The Claude switches are in Snipset for Windows and Linux. The Android app does not have these switches and does not call Claude.

Choose the Cloud Only Where You Want It

Local stays the default, and the cloud is something you choose feature by feature. If that is how you want AI on your desktop, see Snipset plans and pricing.