Trust / Privacy
Privacy and your engineering data
Understand what stays on your computer, what an AI request sends, and what the service keeps before you share a design.
Updated September 24, 2026. SiliconForge is provided by Denis Miclea, an individual operating in Québec, Canada—not an incorporated company. This notice covers the public website, commercial service and native application. It does not replace the policies of your own AI provider or payment provider.
Collect only what the task needs
You can read documentation and download published files without submitting a support email or engineering project. The site does not load advertising trackers, session replay or optional analytics. Normal hosting and security logs may contain request metadata such as an IP address, time and requested URL. Reading a public page does not upload your desktop project.
The support form asks for an email address so the operator can contact you, a subject to route the case, and a message to investigate it. There are no required marketing fields. Use a redacted error and a minimal example: do not include API keys, passwords, card numbers, activation claims or confidential third-party RTL. The subject is stored as case metadata, so keep it non-confidential too.
Checkout takes place on Stripe. Stripe collects the payment and billing information requested on its checkout page; SiliconForge receives the identifiers and subscription/payment state needed to provide and manage access. Do not send payment-card details through support. See Stripe’s privacy policy for its processing.
Choose the AI route before sharing context
- Local engineering: project files, revision storage and installed engineering-tool runs are handled on your computer. Local processing is distinct from asking a remote AI provider for help.
- Managed AI: your request, conversation context and disclosed project files go to
api.siliconforge.space, then to OpenAI using SiliconForge’s service credentials. Approved tool interactions may add source or diagnostic context. Review both context and execution scope. - Your OpenAI API key or personal Codex route: these are separate desktop routes using your configured provider account/runtime. They are not part of the managed-AI allowance. A “local” credential setting does not mean model inference stays on your computer.
The managed service does not accept your personal provider credentials. Credentials for explicit desktop routes stay in native secure storage or the configured provider runtime, not website forms. Provider processing is subject to the terms and settings of the selected route. SiliconForge does not promise zero provider retention, a particular processing country or a blanket no-training guarantee for every route. See the detailed data inventory and route comparison.
Storage, retention and known limits
Managed response records and private streamed job content are encrypted in the service database for delivery and recovery. Support email addresses and message bodies are encrypted there; subjects, case status and creation times are not. Device, billing, quota and audit metadata support activation, usage accounting, replay protection and abuse investigation.
The admitted managed route currently uses the Responses API, not the legacy server Codex adapter. Personal Codex routes keep runtime state on your computer or at your selected provider. Historical or legacy server copies may still contain private runtime/thread state and candidate workspaces in restricted service directories; disabling an adapter does not erase them. These files are separate from encrypted database records. Encryption at rest is not end-to-end encryption: the service must process authorized content.
This release does not have a verified automatic deletion schedule for every hosted record category. Some expiring credentials, replay records and rate-limit data have cleanup routines, but an expired result or activation link is not proof that related jobs, threads, billing records, backups or support cases have been erased. Closing a support case does not delete it. Do not assume a fixed retention period that is not published here.
Local encrypted AI conversation/proposal records, project data and runtime files are separate stores on your device. Removing a project or uninstalling the application is not a promise to erase every store, your backups or provider-side records. Follow recovery and diagnostic guidance before deleting data.
Necessary browser storage, not advertising
Checkout uses browser session storage for a request identifier and an activation claim so the same tab can recover from an interrupted purchase. The activation page removes its stored claim after activation is confirmed. Session storage is scoped to the browser session; clearing it can prevent checkout recovery. Do not share activation URLs or claims. The site does not add an advertising consent banner or optional tracking scripts. This is not a claim that Stripe’s separately hosted pages use no cookies.
Your controls and privacy requests
Limit the files and context you disclose, use a non-confidential example when possible, and do not authorize remote processing of material you lack permission to share. You can revoke an installation and stop using an AI route. Revocation prevents further server-side use of its credentials; it is not a deletion request. The native app also stores a signed, installation-bound offline entitlement with the verified paid-through date. A definitive online revocation invalidates that lease, but a disconnected installation cannot receive the revocation immediately and may retain paid local access until the lease expires. No new source disclosure is needed merely to verify a desktop entitlement.
For access, correction, deletion or a privacy concern, use support with the subject “Privacy request.” Identify the relevant account or case using a receipt/case identifier and describe the request. Do not send identity documents, credentials or your full design in the first message. The operator may need to verify that you are authorized to act for the affected account before disclosing or changing records. Keep the returned case identifier.
For Québec privacy guidance, see the Commission d’accès à l’information du Québec. This does not determine every applicable jurisdiction.
Requests may be limited by legitimate billing, security and legal record requirements; a request is not an automatic erasure guarantee. Ask which records can be removed and which must be retained. If you are dissatisfied with the handling of a Canadian privacy concern, information about the complaint process is available from the Office of the Privacy Commissioner of Canada.
Data handling · Security boundaries · Business terms · Operator and contact