Process
Point at Ollama, vLLM, or LM Studio. Run tagged payloads with live logs and cooperative stop.
RedPrompt
Authorized LLM red-team assessment.
Probe local models for prompt injection and jailbreaks — measure with confidence intervals, then remediate in plain language.
$ python -m redprompt_gui
[LLM-04] System Prompt Extraction · Looks good
[LLM-01] Prompt Injection · Worth reviewing
[LLM-07] Role Assumption · Looks good
Vulnerability Rate: 1/5 (20.0% +/- …, 95% CI) · remediation ready
Built for security & AI teams
One composition of process, evidence, and fix steps — not another opaque scanner dump.
Measure
Objective scoring, 1–5 compliance, Wilson 95% intervals — triage labels that don’t fake certainty.
Remediate
What happened · Why it matters · What to do — written for the whole department, not just engineers.
From authorize to fix
Authorized testing only. Follow the SOP before you aim this at anything.
AuthorizeOwner approval · in-scope endpoint · classification tags.
AssessCLI or GUI · configurable runs · audit trail.
ReviewCI rates · human disposition · evidence excerpts.
HardenPlaybook remediations · re-test until Passed.
Start in minutes
Python 3.14 company baseline. Engine stays stdlib; GUI needs CustomTkinter.
git clone https://github.com/Elpi97/RedPrompt.git cd RedPrompt pip install -r requirements.txt # configure .env → VLLM_BASE_URL / MODEL_NAME python -m redprompt_gui