Uncensored GLM-5.3 + Qwen-3.8 API
on self-hosted GPUs we own.
Uncensored open-weight models. Zero server-side logs. Pay in crypto. Built to run on hardware we own — not a rented cloud that quietly logs your prompts.
$ ask --private
you › write a raw socket port scanner in Rust
glm › Here's a complete async connect-scanner
async fn scan(ip: IpAddr, port: u16) …
nothing above was logged — it lives only in your browser
open the live terminalNo logs. Nothing to leak.
Your prompts travel over TLS, get computed, and are never written to our disks — the gateway keeps only a token counter for billing. Your history lives only in your browser, so there's nothing on our side to log, leak, or subpoena.
GHOSTMETAL · SERVERS
Owned hardware
No prompt logging, on hardware we run, in a jurisdiction we choose. Growing that fleet to the capacity the token plan assumes is a roadmap target — see the tokenomics.
No logs, by design
The gateway keeps only a token counter for billing. Prompts and outputs are never written to disk.
Local-first history
Conversations live only in your browser. One button burns them. Nothing to subpoena from us.
Buy compute, not a subscription
Pay-as-you-go credit packs, or a monthly plan for regular use — either way you're buying compute, never locked in.
Trial
20 free credits to try · we log nothing
- › Access to all models
- › 32K context
- › No API access
- › local-only history
Pro
2,000 credits/mo
- › Access to all models
- › 256K context
- › API access
- › Cheaper than comparable hosted Pro tiers
Max
6,000 credits/mo
- › Access to all models
- › Long context up to 1M
- › API access
- › early features
API
prepaid · priced by selected model
- › API access
- › Access to all models
- › OpenAI-compatible
- › SillyTavern ready
A token backed by hardware we own.
A claim on real inference, not vaporware yield. Stake $GM to mint $CMP and draw a daily compute allowance while the network operates; your staked $GM is returned on unstake.