Basanos
crucible.
crucible shows what an attacker could do to your AI systems, and captures reproducible proof when an attack lands. It sits between your AI app and the model it calls, so you read and alter the prompts, run custom and pre-built attacks against the model, and flag what gets through.
01
What it does
01 intercept
Read and alter the prompts.
crucible sits between your AI app and the model it calls, on OpenAI, Anthropic, Gemini and OpenAI-compatible endpoints like Ollama. See every prompt as the model sees it, and change anything before it is sent.
02 attack
Run custom and pre-built attacks.
Write your own, or run the built-in library: Policy Puppetry, Skeleton Key, many-shot and more. Fire one by hand, or thousands at once.
03 conceal
Hide payloads where an agent will find them.
The indirect prompt injection lab builds attacks into the documents, staged http and https pages, and other material an agent reads for you. It conceals instructions in document metadata, in invisible text, in data fields, split across retrieval chunks, or inside tool descriptions.
04 flag
Flag what gets through.
crucible labels the defensive signals in every response: refusals, safety filters, jailbreak markers. It surfaces the ones that should have been refused but were not, for you to confirm. The detection rules are built in, and you can edit them.
Pricing
Every feature, every seat.
$299 per seat, per year. First 100 seats pay $149 for the first year, then $299.
Prices are in USD and shown before any sales tax or VAT that may apply.
Request access