What WISPR is
A catalogue of uncensored open models that answer, run on sealed hardware, and the frontier models reached through sharding so they never see a whole request. One key, one base URL, one signed receipt per reply. Accounts by email or by recovery phrase. Provider price with enclave time included, one fee when money comes in.
What went
- The workspace. You already have tools. Cursor, Claude Code, VS Code, n8n and every OpenAI client plug straight in, so there is nothing of ours to install.
- Compute as a product. Phala confidential VMs are the only compute, their cost is folded into the per request price, and you never buy a box.
- The vault, sessions as a product, privacy levels. There is one mode and it is private. Everything else was a setting nobody should have to understand.
Step 1: sharding and the infrastructure under it
The enclave that pseudonymises, splits, routes, rebuilds and signs. The catalogue models served from confidential VMs. The gateway, keys with budgets, top ups, x402. Published enclave measurements and the SDKs that pin them. This is the product and it ships first.
Step 2: WISPR editions
Our own uncensored models, built the way Dolphin built theirs: refusals removed from the model, no filter in the weights, the limits in the Terms. Thirteen editions on current bases across text, image and video, each benchmarked against its base with the refusal rate and quality score on the card. Most uncensored shelves are one or two models from 2024 plus proxies. Ours will be current, measured, sealed and impossible to deprecate under us. The line up and method are in the docs.
Everything else
No logs, relay exits, deposit notes, staking, spend controls, portable memory. All real, all on the roadmap, none of them the pitch. The pitch is two sentences: models that answer, and frontier quality without the exposure.