dual-llm

AI-powered penetration testing regulatory intelligence system — LangGraph agents, dual LLM (OpenAI + local MLX), 20+ financial services jurisdictions

A Python implementation of the VETTING (Verification and Evaluation Tool for Targeting Invalid Narrative Generation) framework for LLM safety and educational applications.

Multi-terminal orchestration harness for Claude Code — Generator↔Evaluator loops, Phase Completion Gates, cross-model debate (/tk, /tkc), conflict-free work distribution (/tkm), and blind-evaluator verification (/ralph). v2: enforcement architecture from 31-run analysis.

Signed provenance labels and taint-tracking policy for LLM agent security. The core library behind AgentMesh.

Aider Configuration Setup This repository contains configuration and scripts to easily launch Aider, with dual LLM coding assistances running simultaneously on dual browsers. Required Files config.yaml This YAML configuration file stores your API keys and settings.

Setting benchmarking standards for the LLM security!

Runtime security layer that lets multi-app AI agents read Gmail, Calendar and Slack without letting injected instructions act using a dual-LLM plane split, deterministic policy engine, exfiltration gate, hash-chained audit log.

Advanced AI Assistant with Dual-LLM Architecture

NEEDS ALOT OF WORK FIRST-Unbiased, guardrailed academic chatbot — every answer drafted from scholarly literature and cross-verified by an independent auditor AI (Claude × Kimi K2). BYOK, deployable on Vercel.

Open n8n workflow (CC BY 4.0) that drafts FAQs from an approved fact base with a prompt-injection firewall (Simon Willison's dual LLM pattern). Privileged planner never sees page text; three deterministic gates; human review in Google Sheets.

A test-bench for prompt-injection attacks and defenses: run an attack corpus against a live Claude model and measure attack-success-rate per category, graded by a deterministic canary check + an LLM judge. Includes a dual-LLM quarantined-data defense.