Least privilege
Limit tool, filesystem and network access to what a workflow needs. Agent actions should remain visible and subject to deliberate approval.
Privacy & security
Control is a design choice.
Local execution is a useful boundary. Security comes from the decisions around it: provenance, permissions, network access and maintenance.
This preview keeps it simple.
The website is static. There are no accounts, payments, uploads or application backend. The chat streams canned text on your device.
- No cookies, analytics or third-party embeds.
- No chat messages sent from your browser.
- Only same-origin site assets and build-update checks.
The hosting server receives normal requests needed to serve a website. See the privacy notice for the boundary.
YOUR MACHINE
PRIVATE WORKSPACEPrompt Model ResponseDesigned to stay inside your boundary
OPEN MODELSLOCAL INFERENCE
Planned product principles
Make trust boundaries visible.
These are design intentions, not a certification or a claim about released software.
Model provenance
Review the source, licence and integrity of model files. An open model is not automatically suitable for every task or data type.
Operational ownership
Document patching, backup, recovery and access policies. A local deployment still needs people who can maintain it.
Output review
Models can produce incorrect or unsafe output. Keep a person responsible for reviewing consequential results and tool actions.
Report a security concern.
Contact hello@localai.best with a description and safe reproduction steps. Please do not send private user data, credentials or sensitive files.
There is no formal response-time commitment or reward program in this preview. We will review reports as part of maintaining the site.
Read privacy and preview terms