Local-first processing
Log parsing, signal decoding and numeric measurements run locally. The engine filters relevant messages and builds compact observations rather than sending raw large logs to a model.
Live AI requires explicit selection of a configured provider. Relevant context may leave the machine in that mode; it is not an offline-only guarantee. See the privacy notes for that boundary.
Project-relative provenance
Exported evidence identifies artifacts relative to the selected project root. Inputs outside that root are rejected; an engineer must explicitly select a containing root when needed.
Reports do not serialize the absolute project root or runtime path mappings. Source text and user-authored content may still contain sensitive information. Review a report before sharing it.
Model credentials
The current engine reads configured model credentials locally. Application Programming Interface (API) keys are not embedded in reports or the non-secret model-call audit. Credentialed requests do not follow redirects.
This website does not collect API keys. Its Beta request form prepares an email draft and accepts no engineering file uploads.
What the current Beta can promise
- Evidence boundaries: measured facts, candidate inferences, unknowns and next investigation steps remain distinct.
- Time boundaries: separate logs, channels and clocks are not assumed to be synchronized.
- Knowledge boundaries: model output is derived knowledge, not independent engineering evidence; source checks do not replace full semantic review.
- Access boundaries: this website is an early access entry point, with no hosted analysis workspace or account system.
Enterprise identity controls, compliance certifications, penetration-test attestations and service-level commitments are not offered by this minimum website.
Report a concern
Send an initial, non-confidential description to beta@embedvera.com. Do not include credentials, customer data or proprietary files in an initial message.