Use case · AI red teaming
Red-team your AI with a model that plays the adversary.
Test models, agents and guardrails against what real users will try. An adversary that refuses to attack can't tell you much.
Built forAI safety & platform teams
Why refusals get in the way
Red teams need a model that writes the jailbreak, the manipulative persona or the policy-breaking request their own system has to withstand. Mainstream assistants refuse exactly those requests, so test coverage stops where a vendor's filter starts.
What teams use it for
- Generate adversarial prompts and jailbreak variants for your own models
- Probe agents and tool calls for prompt injection
- Stress-test moderation filters and guardrail classifiers
- Build evaluation sets that catch regressions between releases
What Lyber gives you
Answers, not refusals
Uncensored and abliterated weights, with no refusal layer of Lyber's own on top.
Your policy, in your system message
The scope and rules for each request travel with it, in the system message you write.
One OpenAI-compatible API
Keep your SDK, harness and agents. Change the base URL and the model name.
Keys, limits and records
A key per app or person with spending limits, shared organizations, and a usage ledger designed to exclude prompt text.
The request
POST /v1/chat/completions
{
"model": "lyber/auto",
"messages": [
{ "role": "system", "content": "<target system, scope and rules of engagement>" },
{ "role": "user", "content": "<your request>" }
]
}Early access. Inference opens as each model is qualified, and launch prices are provisional.