Last updated 2026-11-20

Privacy policy

AbliteratedLLM is built to know as little about you as possible. Creating an account needs no KYC, name or email, and we don't store your prompts or outputs, with one narrow exception described below. This page lists everything we do keep, why, and for how long. Models labelled Official DeepSeek API are run by DeepSeek, which stores what you send them under its own privacy policy (details).

At a glance

What we store
  • A keyed hash (HMAC-SHA256) of your account number, never the number itself
  • Your balance, account status, referral code and account creation time
  • Keyed hashes of your API keys, with each key's name, 10-character prefix and spend caps
  • Request metadata for 30 days (time, key, model, token counts, cost, status, latency), then daily totals
  • Your credit ledger (deposits, bonuses, usage charges), top-up invoices (our UUID, provider resource ids, amount, status and amount credited) and crypto payment details (coin and amounts)
  • For card top-ups, your Stripe customer ID, each payment's method, card type and issuing country, and the results of the fraud checks
  • An HMAC of your network prefix: 30 days for a free-credit claim or a top-up invoice, 90 days after a deposit bonus
  • Your recovery email, hashed and encrypted, only if you add one
  • Your low-balance alert amount and when the last alert went out, only if you turn alerts on
  • Sealed safety incident records, as described below
  • Support tickets you open (subject and messages) until 180 days after they close
What we never store
  • Your prompts, including any images or video in them
  • Model outputs
  • Your IP address (it is held in memory only, and sent to Stripe once, for its risk check at your first card checkout)
  • Your name, ID documents, phone number or billing address (no KYC to create an account)
  • Card details, or the name, email address and billing address you enter on Stripe's checkout
  • Your wallet addresses or transaction hashes (our crypto payment processor has them; we don't)
  • Analytics, tracking or advertising data
  • Playground conversations (they live only in your browser tab, or in this browser's storage if you turn on saving)

Prompts, outputs and IP addresses have one exception, described below. These lists cover our own systems. Prompts and outputs sent to models labelled Official DeepSeek API are stored by DeepSeek, as described below. Card checkout details processed by Stripe are covered in Payments.

Your account

Your account is a random 16-digit account number, and that number is also your login. We keep only a keyed hash of it, so we can never show it to you again. If you lose it and haven't added a recovery email, we can't restore access.

API keys work the same way. We store a keyed hash of each key plus its first 10 characters, so you can tell your keys apart in the dashboard. The full key is shown once, when you create it.

Prompts and outputs

Your prompts and the model's outputs pass through our gateway in memory on their way to the model and back. They are not written to our database, our logs or disk, except in the sealed incident record described below. Images and video sent to models that accept them are handled the same way as text. Our abliterated models run on dedicated GPUs operated by Fireworks AI, which may run them in any of its regions, under a zero-data-retention configuration. Stock models run on shared GPUs operated by Fireworks AI, which may run them in any of its regions, under the same zero-data-retention configuration, as described below. Models labelled Official DeepSeek API are the exception: they run on DeepSeek's servers, as described below.

Traffic to this website and to the API reaches our servers through Cloudflare, which terminates TLS and protects the service against DDoS attacks. Prompts and outputs pass through Cloudflare only in transit, and it doesn't cache or keep them for us.

Except for designated reseller accounts described below, an automated in-memory check scans prompt text before forwarding and output text after completion (the beginning and end of a very long output), on every model. It reads text only: images and video are not screened. It refuses child sexual abuse material (CSAM): the request is blocked and a sealed incident record is kept, which we report as the law requires (see Sealed incident records). It also refuses clear, actionable instructions for mass-casualty weapons (CBRN weapons or mass-casualty explosives); those requests are not recorded. Neither carries automatic account action. The prompt check fails closed: if it can't run, the request is refused. The output check runs after the response has been delivered: if it can't run, that output isn't checked. No human reads prompts or outputs outside that sealed record, and the text the check reads is not stored. The acceptable use policy has the details.

Designated reseller accounts may be exempt from both automated checks when responsibility for screening their users is delegated to the reseller under its agreement with us. Requests on these accounts are not sent to our safety classifier and do not generate classifier-triggered incident records. The request-metadata and billing retention described in this policy still applies. This exemption does not waive legally required preservation or reporting.

Stock models

Models labelled Stock model run their vendor's original weights, unmodified: they are not abliterated and keep the vendor's own refusals. They run on shared GPUs operated by Fireworks AI, which may run them in any of its regions, under the same zero-data-retention configuration as our abliterated models. Our promise not to store your prompts or outputs covers them in full, and we record the same request metadata as for every other model.

Some stock models accept images, and some also accept video. Images and video pass through our gateway in memory like the rest of your request and are not stored, with the same single exception: the sealed incident record. The automated safety check reads only the text of these requests, so images and video are not screened. You remain responsible for everything you send, as the acceptable use policy sets out.

Models served through DeepSeek's official API

Models labelled Official DeepSeek API are DeepSeek's stock models. We don't run them: your request passes through our gateway as described above and is then sent to DeepSeek's official API, which generates the answer.

Prompts and outputs for these models are sent to DeepSeek, processed and stored in the People's Republic of China under DeepSeek's privacy policy, and may be used by DeepSeek to train its models. DeepSeek offers no zero-data-retention option. Our promise not to store prompts and outputs covers our own systems only; what DeepSeek keeps, and for how long, is governed by the DeepSeek Privacy Policy. If that matters to you, use one of our abliterated or stock models instead.

With each of these requests we send DeepSeek a pseudonymous identifier for your account: a keyed hash that is the same for all your keys and is not your account number, an API key or your IP address. DeepSeek uses it to keep its prompt cache and its content-safety handling separate per customer. Any user or user_id value you send is replaced by it. DeepSeek doesn't receive your account number, your API key, your IP address or your payment details from us.

Our in-memory safety check runs on these requests the same way as on every other model. DeepSeek may also decline requests on its side under its DeepSeek Terms of Use.

Request metadata and retention

For every API request we record:

  • the time, request id and endpoint;
  • the API key and model used;
  • input, cached input and output token counts;
  • the cost and the outcome (completed, cancelled by the client, or upstream error);
  • the latency and the time to first token, in milliseconds.

We keep these rows for 30 days so you can inspect recent usage and so we can investigate abuse reports. After 30 days they are rolled up into daily totals per account and model (request count, token counts and cost) and the individual rows are deleted. Your credit ledger and invoice history are kept as the record of your balance. Support tickets keep the subject and messages you and we write until the ticket is closed (by you, by us, or automatically 14 days after our last reply if you don't answer), and are deleted with all their messages 180 days after closing.

A request refused because the model had no serving capacity (it was unavailable, or every replica stayed busy) leaves no request-metadata row. We only count such refusals per account, model, reason and minute, so we can see which accounts an outage affected, and delete those counts after 30 days.

IP addresses

IP addresses are held in memory only. We use them while a request is being handled for rate limiting, login throttling and the free-trial and bonus checks, and we never write them to our database or logs. The only exception is the sealed incident record described below. Cloudflare, which carries the connection to us, also sees your IP address in transit. When you start your first card checkout, we send your IP address and your browser's user agent to Stripe for its risk check, as described in Payments.

When you have no NEXT_LOCALE cookie and your browser's language settings name none of the site's languages, we pick the page language from the country Cloudflare derives from your IP address (its CF-IPCountry header). The country is used only for that choice, and we never store or log it.

To allow one free trial per network, we keep an HMAC (a keyed one-way hash) of the IPv4 /24 or IPv6 /48 your address belongs to for 30 days, not linked to your account. Free credit isn't offered to datacenter, VPN or Tor networks, which we check in memory against public address lists.

At most 2 trial accounts per network can use the API within an hour. To enforce that, the API gateway keeps an HMAC of the IPv4 /24 or IPv6 /48 a trial account's request came from (and, for IPv6, of the enclosing /32) together with that account's id. This is held in our in-memory cache only, never in the database, and is deleted one hour after the last request a trial account sent from that network. Paid accounts are not recorded, and the address itself is never stored.

The deposit bonus is also once per network. Each invoice stores the HMAC of your network prefix (the IPv4 /24 or IPv6 /48) with your account for up to 30 days. When a bonus is granted, we keep a separate copy of that HMAC, not linked to your account, for 90 days. An HMAC can't be reversed into an address, and a daily job deletes these records once they expire. Login-attempt counters use the same kind of HMAC, of the /24 or /48 and, for IPv6, also of the enclosing /32, and expire within minutes.

Sealed incident records

When the text of a request or output is flagged as child sexual abuse material, we keep the flagged text, the first 256 KiB of the request and of any output, its source IP, time, account, key prefix and model in a record sealed with public-key encryption: the running service can't open it, only the operator's separately held key can. The request is kept as sent, so that part includes any images or video in it, as data or links. At most one such record per account per hour holds content; later flags in that hour are recorded without it. We delete it after one year and report it as the law requires, as set out in the acceptable use policy.

Payments

Crypto payments are processed by NOWPayments, a custodial crypto payment processor. You pay on NOWPayments' hosted checkout page, in the coin you pick there, and NOWPayments receives the funds and holds them in our balance with it. To do that, NOWPayments receives your transaction data: the sending address, the transaction hash and the amounts. The checkout page is NOWPayments' site, not ours, so it also sees your IP address, and our cookie and analytics promises below don't cover it. What NOWPayments collects and keeps is governed by the NOWPayments Privacy Policy.

NOWPayments runs its own anti-money-laundering (AML) program under its NOWPayments AML/KYC Policy. It may check incoming payments, and it may hold, delay, refuse or cancel a payment while it does or when it sees a risk of fraud or illegal activity. We don't ask you for identity documents.

Card payments, including Apple Pay, Google Pay and Link, and USDC stablecoin payments, are processed by Stripe, our payment processor. We sell you the API credit; Stripe takes the payment on its own hosted checkout page and handles the payment data under the Stripe Privacy Policy. To take the payment, Stripe receives what you enter on its checkout: your card details, the cardholder name, your email address and your billing address (for USDC, the wallet and transaction you pay with). With Apple Pay or Google Pay, your wallet sends Stripe a payment token instead of the card number. Stripe emails its receipt for the payment to the email address you enter. That checkout is Stripe's site, not ours, so our cookie and analytics promises below don't cover it. We don't send your prompts, outputs, API keys or account number to Stripe.

Stripe screens every payment for fraud with its Radar service, using the IP address and the browser and device data its checkout collects. So that Stripe can link your card payments for this screening, we create one Stripe customer record for your account, holding only your generated account id (a random identifier, not your login account number) and no name or email address. When you start your first card checkout, we also send Stripe your IP address and your browser's user agent, so that Stripe can assess the risk that one person is using several accounts. We store only Stripe's result (its ID, a risk level and a score) with your account, and an account Stripe rates highest risk can't top up by card. With every card checkout we also tell Stripe whether your account is established (at least 7 days old, with at least $50 of top-ups credited) or new, so that its fraud screening can be stricter for new accounts.

We don't send either payment provider your login account number or other customer account data, apart from the Stripe customer record and account tier described above. An invoice carries our invoice UUID, the amount in USD, a generic description and the links back to this site. The UUID maps the payment to your generated account id for API credit; it does not give the provider access to your account.

For each top-up we store our invoice UUID, provider checkout/payment/resource ids, status, amount, the payment processing cost a card price includes and amount credited, linked to your account. Card records contain no buyer identity or raw checkout payload. For each payment NOWPayments reports on an invoice, we also store the USD price, coin, crypto amount asked and amount received. We don't store your wallet address or transaction hashes. Payments in Bitcoin, Litecoin, Ethereum, USDT and other coins on transparent blockchains are public by nature. Monero hides the sender, recipient and amount on-chain, but NOWPayments still receives the transaction data described above. Every top-up invoice, including a card invoice, has the network-prefix HMAC retention described above.

For a card top-up, what we keep is only the checkout session and payment IDs, the price in USD and the payment processing cost it includes, the currency and amount Stripe charged when it wasn't USD, the payment's status, the payment method (card, Apple Pay, Google Pay, Link or USDC), the card's type (credit, debit or prepaid) and the country that issued it, the result of 3-D Secure verification, the number of declined payment attempts, and whether Stripe reported a refund, a dispute or a fraud warning from the card's issuer. These records are kept with your invoice history as the record of your balance (see Request metadata and retention). We use the payment method, the card type and the 3-D Secure result to decide whether a payment is credited, as the terms describe. To confirm a payment, our server reads it back from Stripe, and its reply also contains the cardholder name, the card's brand and last four digits, and the name, email address and billing address entered at checkout. We discard them without storing them: we never receive the full card number, and we never store card details, names, email addresses or billing addresses in our own systems. Stripe keeps them with the payment in our Stripe account, where its dashboard shows us the name, email address, billing address and the card's brand and last four digits, never the full card number.

Card and Pix payments we took through our former Stripe account, from 14 to 16 November 2026, were processed by Stripe, which keeps the records of those payments and handles their refunds and disputes. Its handling of that data is governed by the Stripe Privacy Policy. For each of those payments we still keep, with your invoice history, the checkout session and payment IDs, whether it was paid by card or Pix, the amount and currency, the exchange rate and processing fee it was priced with, its status, whether a card payment completed 3-D Secure verification, and whether Stripe reported a refund, a dispute or a fraud warning from the card's issuer. We never stored their card details, names, email addresses or billing addresses in our own systems; Stripe keeps them with the payment in that Stripe account, where its dashboard shows us the name, email address, billing details and the card's brand and last four digits, never the full card number.

Card payments we took through Creem, on 17 and 18 November 2026, were processed by Creem as merchant of record, which keeps the records of those payments and handles their refunds and disputes. Its handling of that data is governed by the Creem Privacy Notice. For each of those payments we still keep, with your invoice history, the Creem checkout, order and transaction IDs, the amount, the card processing fee and the sales tax or VAT charged, its status, and whether Creem reported a refund or a dispute. With each checkout we also stored the country Cloudflare derived from your IP address when you started it and the billing country Creem reported for the payment; we delete both 30 days after the checkout was started. We never stored card details, names, email addresses or billing addresses in our own systems; Creem keeps them with the payment in our Creem account, where its dashboard shows us the buyer's name and email address.

Recovery email (optional)

You can add a recovery email in your dashboard settings. It is saved only after you click a verification link (valid for one hour); until then the address is held in memory only. Once verified, we store a keyed hash of the address so we can look it up, plus an AES-256-GCM encrypted copy so we can send you a login link. Recovery links expire after 15 minutes. Besides login links, we use the address only for low-balance alerts you turn on in settings and for notices that we replied to your support ticket. Removing it (at any time) also turns low-balance alerts off.

These emails are delivered by Proton Mail, which receives the address and the message to send them. Proton's handling is governed by the Proton Privacy Policy.

Cookies and analytics

We set no cookies other than these three:

  • __Host-sid: your login session. It is httpOnly, sent only over HTTPS to this site, and expires after 30 days, and we store only a hash of it.
  • ref: the referral code from a ?ref= link, kept for 30 days so the referrer gets credit when you sign up.
  • NEXT_LOCALE: the language of the last page you opened, stored as a code such as ja and nothing else, and kept for one year. It is used only to choose the language of the site's pages and of its replies to your requests.

There are no analytics, tracking pixels or advertising scripts. Playground conversations stay in your browser's memory and are gone when you close or reload the tab.

Sharing and legal requests

We don't sell your data or share your customer account data with payment providers, apart from the Stripe customer record, account tier and risk check described above. Using a model labelled Official DeepSeek API sends that request to DeepSeek (see above), paying an invoice goes through NOWPayments or Stripe (see Payments), and email to and from our addresses goes through Proton Mail, which hosts them (see Proton Privacy Policy). If we receive a valid legal request, we can only provide what this page says we have, and that never includes your prompts or outputs (except for the sealed record described above).

Who we are and contact

This service is operated by Abliterated LLM LLC, organized under the laws of the State of Wyoming (registration number 2026-002136649), with its principal office at 30 N Gould St Ste R, Sheridan, WY 82801, United States. For privacy questions, write to legal@abliteratedllm.dev. To report abuse, see the report page. If this policy changes, we'll update the date at the top of this page.