STRAIGHT ANSWERS
Frequently asked.
Current answers for the private-beta service. Claims change only after the corresponding implementation and validation change.
Does AblitAI store prompts or responses?
AblitAI's application does not intentionally retain prompt or response content. Requests are processed transiently to generate a response. Infrastructure providers may process network and operational metadata; see Privacy.
What model is available?
The launch model is a text-only Qwen3.8-27B Abliterated community checkpoint, served using asymmetric AWQ W4A16 quantization on one RTX 3090 Ti. It remains in launch validation. See the model and serving disclosure.
What does “abliterated” mean?
It describes a community-modified checkpoint intended to reduce some refusal tendencies from the source model. It does not mean the service is ungoverned, infallible, or permitted for illegal use. The Acceptable Use Policy still applies.
Is an email address required?
No. Account creation returns a random account number and API key. The plaintext key is returned once and stored locally by the browser. The account number cannot recover the key or prove ownership.
How do credits work?
One credit pays for one prompt or generated token reported in usage.total_tokens. Packs and checkout availability come from the live API catalog. Paid checkout is currently closed.
Why Bitcoin first?
The planned initial payment beta uses self-hosted BTCPay Server and is restricted to attested Singapore customers while accounting, legal, capacity, and payment controls are validated. Bitcoin transactions remain public on the Bitcoin network.
Are coding harnesses supported?
Not publicly yet. OpenCode, Codex, and Claude Code have local/mock text and simulated tool protocol evidence, but standard tool-bearing requests are not release-supported on the text-only Release A GPU. Reference-only setup guides are published for review; exact-GPU parser validation, readable errors, exact accounting, and sustained load must pass first. Claude Code remains an unofficial non-Claude route. See the compatibility matrix.
Is there an uptime guarantee?
No. Private beta runs one dedicated launch GPU and has no uptime SLA. Checkout must fail closed when critical readiness checks fail.