Private
Matters

Field notes / 05 · Privacy in practice

Local AI Privacy: Trace Text, Voice and Image Routes

A chat window running on your computer does not tell you where its model runs. Map each feature’s destination before deciding whether the configuration fits your privacy needs.

Updated September 18, 2026 · 3 min read · Adult readers 18+

Fictional adult woman with glasses at a warmly lit home computer desk.
AI-generated fictional adult character. Not a real user or an official app avatar.

Trace one request from start to finish

Interface → selected model endpoint → optional feature services → saved records

For each step, write the component name, where it runs, what it receives and where it stores results. Mark an unknown destination as unknown. A local text model with a cloud voice feature is a mixed configuration, so assess the voice route separately.

SillyTavern’s API connections documentation lists local and cloud backends. That makes the selected connection important: the name of the interface alone is not a privacy guarantee. This guide does not certify any particular installation.

A configuration worksheet that catches side routes

On a narrow screen, scroll the table sideways to read every column.

Fill these fields for the setup you actually use
ComponentRecord
Text generationEndpoint host; local process or remote operator; request contents
Speech recognitionWhere microphone audio goes; whether recordings are saved
Voice outputWhich service receives the text to speak
Image featuresUpload destination; prompt destination; saved outputs
Extensions / memoryEnabled components; search or embedding destinations; retained records
Storage / backupChat directory; exported files; device or cloud backup copies

A limited offline check

  1. Record the current configuration and save your work. Use a fictional test conversation.
  2. Disconnect networking temporarily and start a new request. Test text, voice and images individually.
  3. Record which features respond and which fail. Restore networking when finished.
  4. Compare those observations with the configured endpoints and documentation.

A feature working offline shows that this observed operation could complete without a live connection at that moment. It does not prove that the application never contacts a server, that it cannot queue data for later, or that its backups stay local. A failed feature is also not proof of data collection: licensing, downloads or other dependencies can require connectivity. Treat the check as a clue to investigate, not a security audit.

Local control creates local responsibilities

Decide who can use the computer and who can read its files. Review device access, backups and any remote-access configuration. A locally saved chat may still be copied by a sync folder. Keep API keys out of screenshots and public configuration files.

Choose a setup you can maintain. A fully local workflow may require hardware, model downloads and troubleshooting; a remote service shifts processing to an operator whose rules you must inspect. Neither label resolves every risk. The useful outcome is a documented route that matches what you are willing to send.

Revisit the worksheet after changing a model connection or enabling voice, image or memory extensions. If your next concern is removing existing records, use the deletion plan for both local copies and external providers.

Make it practical

Keep your own record

Download the blank worksheet and fill it in privately. No account or upload is needed; nothing you write in the downloaded file is sent to this site.

Download worksheet · CSV

Scope: editorial guidance and official documentation checked on September 18, 2026. No in-app deletion tests or infrastructure audit were performed. Illustrative scenarios are fictional. Evidence method · Website privacy

Your next conversation

Explore WantYou.ai

Take your next step with a clearer idea of what you want to share.

Continue to WantYou.ai ↗

18+ · External website, opens in a new tab. Review its current terms and privacy policy before signing up. WantYou.ai has not been assessed for privacy in this guide.