$ localsutra privacy --endpoint https://api.example.com/ --format md (exit 0) # LocalSutra AI — Privacy and Offline-Boundary Check **Schema:** localsutra.privacy.v1 **Integrity:** SHA-256 seal verified (1efeb6d90692a2a9…) **Trust boundary:** Loopback only. No cloud calls, no telemetry. Unknown and not-run values are not successes. > **Verdict: ENDPOINT_BLOCKED** — Only http loopback is allowed ## Claims | Claim | Value | |---|---| | coreHasNoMandatoryCloudApiKey | true | | remoteEndpointBlockedByValidation | false | | modelRuntimeEgress | not_tested | | modelRuntimeTelemetry | not_tested | ## Data inventory | Field | Stored in | Retention | Shared | |---|---|---|---| | OS, architecture, CPU model/count, RAM total/free | report files you choose to write | until you delete the file | never automatically | | GPU name, VRAM, driver (only if nvidia-smi is present) | report files you choose to write | until you delete the file | never automatically | | Free disk space of the working-directory volume | report files you choose to write | until you delete the file | never automatically | | Model ids from your catalog or a local Ollama /api/tags listing | report files you choose to write | until you delete the file | never automatically | | Prompt hash (SHA-256) for opt-in inference | inference report | until you delete the file | never automatically | | Prompt and response text | not stored | none | never | ## Notes - External runtimes may use their own network features - For air-gap claims, use external network monitoring and disconnect tests - This audit does not establish compliance or end-to-end encryption - Hardware profile, model list and file names can be sensitive: review reports before sharing --- LocalSutra AI 0.2.1 · created by Vatsal Shah · Apache-2.0