FAQ

We do not keep logs or server-side history of your prompts, model responses, images, uploaded documents or extracted document text. Content is processed temporarily to fulfil your request; generated images are held briefly in memory for delivery. Persistent chat history and images stay on your client. We retain registration data, content-free billing and payment records, and anonymous website visit statistics.

Our own server collects website visit statistics, without third-party analytics SDKs. These include visit time and page, referral and campaign source, browser/device and language, country and network. Signed-in page visits and API traffic are excluded from website visit counts. Records are not linked to accounts and are retained for 90 days. Raw IP addresses are not stored in the statistics database; a monthly keyed hash counts repeat visitors. This is a pseudonymous identifier while its key exists; the key is held in memory and expires at the next UTC month.

The web chat uses IndexedDB in your browser, through Dexie. Conversations and image files are saved locally; image files are stored as Blobs. Signing in on another device does not download your history from us. Clearing browser site data can delete that local history, so export a backup before changing devices or clearing data.

No. To answer a request, our inference hardware must process the prompt and any image you submit. They are transmitted over TLS and processed transiently. Local-only describes persistent history and image storage, not on-device inference. Generated images are held briefly in bounded server memory for delivery, then saved by the web client.

We retain your registration/account identifiers, credit balance, credit ledger and payment records. Per-request usage metadata contains a timestamp, model and token counts, without prompts, responses or images. Anonymous website statistics are collected separately and are not linked to accounts. The Privacy Policy describes the retained records.

Local storage and TLS are different protections. TLS encrypts transport; the inference server processes plaintext to answer. Browser history is not advertised as an end-to-end encrypted vault. Protect your device and browser profile, and treat exported backups as private files.

Each model has a context window and output limit. Context includes the conversation and any supplied files or images, not just the last message. Rate limits, available credits and capacity limits also apply. See Models for context and pricing, and API docs for integration details.

Our server sends search queries to search engines and fetches requested pages; the browser communicates with our service. Running our own metasearch does not make those external services part of our inference infrastructure. Avoid sending secrets in search queries. History is still stored on your client; local storage does not prevent external services from receiving a search query.

No. Prompts, responses, images, documents and extracted document text are processed temporarily, without server-side content logs or history. Your saved history and images stay on your client. Registration data, content-free billing records and website visit statistics are retained separately.

Paid web chat and API requests have no general-purpose content filtering. Trial web chat applies a narrow illegal-content filter. All use must comply with our Acceptable Use Policy and applicable law. Model behaviour can still vary.

You can pay with cryptocurrency (including BTC, XMR, and USDT) or by card through Stripe. Card payments are available for PRO and MAX monthly subscriptions, which renew automatically. You can manage your subscription or cancel renewal from your account. Credits are spent as you use the models at the posted per-token rates.

Yes — the API is OpenAI-compatible. Point your client's base URL at https://api.ghostmetal.io/v1, use your key as the Bearer token, and set the model to glm-5.3. Works with SillyTavern, the OpenAI SDKs, and anything OpenAI-shaped.

glm-5.3, glm-5.3-long, and qwen-3.8. One balance covers all of them. See the Models page for context windows and prices.

Generate a recovery code in your account settings and save it somewhere safe. If you forget your password, that code plus your email resets it. Wallet and Google sign-ins don't need a password at all.

Refunds are available only where required by law. Buy a small pack first if you want to try it.

Your key is a bearer token — treat it like a password. You can rotate it anytime, and you can create named sub-tokens per app (disable or delete one without touching the others).

On hardware we operate, in a jurisdiction we choose. Your plaintext never touches a third-party GPU — a guarantee rented-cloud AI structurally cannot make.

Yes. The service is reachable as a Tor hidden service for users who want network-level privacy on top of no server-side logs.