# Witness.ai

> Verified, bring-your-own-cloud (BYOC) LLM fine-tuning for regulated enterprises. "Fine-tune inside your own cloud, with every claim about the result independently verified." Two tiers on a single honest axis — privacy: train in the open for free (public), or in a private enclave for money. You pay for privacy, not for the rigor; the rigor is standard.

## What it is
- **BYOC:** the control plane orchestrates; compute runs and data lives in the customer's own AWS/GCP/Azure account. Witness.ai never holds customer data or trained weights on the paid tier.
- **Verified rigor (the moat):** every run yields a signed, immutable audit record — what was checked, against what thresholds, with what result — and the platform actively hunts the failure modes that quietly produce a plausible-but-broken model. This is demonstrated by four real incidents, not asserted.
- **The four incidents:** (1) a GPU host reporting healthy while SSH was refused — caught by testing, not the status flag; (2) a train/eval prompt-format mismatch root-caused and its fix measured (bfcl 0.275→0.375); (3) a safety default silently overridden by a duplicate hardcoded default, caught at $1.50; (4) a compression technique empirically shown not to work before shipping.

## The two tiers
- **Open (free):** real SFT on 3 shared H200 slots, same signed audit record. The catch, stated plainly: the dataset and resulting weights are retained and may be published to a public open corpus. Not for confidential/regulated data.
- **Enclave (paid):** runs in the customer's own cloud; Witness.ai never sees data or weights; exportable compliance record (HIPAA/SOC2-ready); support. A regulated enterprise cannot use the free tier (their data can't be public) — so the tiers never cannibalize each other.

## Status
- Marketing site: live. Platform: in development (MVP = AWS, SFT stage, manual onboarding of 2–3 design partners). The live training feed shows a truthful "launching" state until the platform API exists.

## Key pages
- [Home](https://witnessai.com/)
- [How it works & tiers](https://witnessai.com/#tiers)
- [Proof / the four incidents](https://witnessai.com/#proof)
- [Guidance docs](https://witnessai.com/docs)
- [Build log](https://witnessai.com/blog)
- [Privacy Policy](https://witnessai.com/privacy)
- [Terms of Service](https://witnessai.com/terms)

## Contact
- Design partners & general: partners@witnessai.com
- Privacy: privacy@witnessai.com
- Operated by Peter Lodri EV (EU VAT HU69757200).