RedCell gives vetted AI and security teams one OpenAI-compatible API for abliterated open-weight models, built for red-teaming, LLM jailbreak testing, safety evaluations and model-risk testing. Hosted in the EU, every request logged for your own audit, with strict acceptable-use rules.
Frontier and hosted models refuse a lot of legitimate safety work: jailbreak and refusal test cases, red-team probes, and model-risk and safety evaluations.
The uncensored tools that do exist are built for consumer and NSFW chat. They do not verify who the customer is, and they do not give you logs you can show an auditor.
So AI red teams either stand up their own GPUs or fight refusals all day, and neither leaves a clean paper trail for a client or a compliance review.
Point your existing client at a new base URL and keep your code. Drop it into your eval harness in minutes.
A small, curated set of refusal-removed open-weight models chosen for quality, not a 40,000-model dump.
Every request and response is logged and kept for you, so an audit or a client review has something concrete to read.
We publish how often each model refuses and how it benchmarks, and we update it as models change.
Data handled in the EU, with a data processing agreement available before launch.
Identity and company checks at sign-up, so you know every other customer on the platform was vetted too.
This is early access. The API is not live in production yet, and there is no payment to make today. Join the waitlist and tell us whether you would prepay for month one.
Join the waitlistWe log each request and response with a timestamp and your account ID. Logs are for your own audit and for abuse enforcement. At launch you will set your own retention window within legal limits, and you can export your logs. Exact defaults land in the DPA before launch.
A small set of strong abliterated open-weight models rather than a huge catalogue. We will name the exact models and publish their refusal rates and benchmark scores before launch. We are not committing to a list here because the good ones change month to month.
We take it seriously and we are being cautious rather than making promises we cannot keep yet. RedCell is an EU-hosted service and we intend to meet the obligations that apply to us, including transparency and logging, and we will take legal advice before launch. We are not claiming full compliance today. If the Act or its guidance rules out part of this, we will say so and change course. Verification, logging and the illegal-use ban are there partly so there is accountability for how the models are used.
Sign-up requires identity verification (KYC) and confirmation of your company or professional status. We expect to use a standard KYC provider plus a manual check for the first cohort. No access until both pass.
No. We host a small, curated set of abliterated models in the EU, vet every customer, log everything, and publish refusal and benchmark data. The consumer uncensored services (Venice.ai, Featherless, OpenUncensored, ArliAI, NanoGPT) do none of that verification or auditability, and they lean toward NSFW. Different product, different buyer.
Early access, launching soon. No payment today. We'll only use your details to contact you about RedCell.
When you join the waitlist we store the name, work email, company, role, use case, your prepay answer and any notes you enter, plus the time you submitted. We use it only to contact you about RedCell early access and to gauge demand. We do not sell it or share it with third parties. There is no payment and no card details are collected. To have your entry deleted, email fathindos.fd@gmail.com and we will remove it.