⏳ Patience

Automated red teaming & compliance validation for AI chatbots — purpose-built for regulated industries.

Zero cloud dependencies. Built for teams where data residency and auditability aren't optional.

Download · Website · Documentation · Security

---

Why Patience?

Patience is a native macOS app that red-teams and validates your AI chatbots entirely on your machine. No attack traffic, transcripts, or test results leave the device — so you get audit-ready evidence without breaking HIPAA, SOC 2, FedRAMP, or GDPR data-residency rules. It runs fully air-gapped when you need it to.

🔒 100% On-Device	Zero cloud dependencies. Nothing is transmitted externally.
🤖 AI-Powered Red Teaming	Autonomous probing for prompt injection, jailbreaks, and social engineering.
📋 Audit-Ready Evidence	Findings mapped to the OWASP LLM Top 10 and MITRE ATLAS.
🧩 Works With Your Stack	Test any HTTP/WebSocket endpoint — OpenAI, Anthropic, Ollama, or custom.

Three Testing Modes

🎬 Scenario Testing

Script multi-turn conversation flows and run them against any endpoint; validate each response against custom business rules in real time, and catch regressions before production.

🔎 Conversation Forensics

Import production logs in JSON, CSV, or plain text; surface failure patterns, policy violations, and drift over time. A guardrail-failure triage cascade then classifies where and how a target's guardrails failed — per turn and across the whole episode — staying on-device by default and reaching a frontier judge only when it changes the answer (and only after redaction).

⚔️ Adversarial Testing

AI autonomously probes your chatbot — prompt injection, jailbreaks, social engineering, and stress loads — then maps every finding to the OWASP LLM Top 10 and MITRE ATLAS.

🛡 Built for Security

Patience is designed for environments where data can't leave the building:

• Fully on-device / air-gappable — no attack traffic or test results transmitted externally.
• Compliance-aligned — supports HIPAA, SOC 2, FedRAMP, and GDPR data-residency requirements.
• Secrets stay in the Keychain — API keys are never written to configs or source.
• Open-core forensics — the architecture, verdict schema, and judge rubric are public; trained judge models and calibration data remain a private asset (see the contribution boundary).

See SECURITY.md for the full policy and vulnerability reporting.

Requirements

• macOS 13.0+
• Xcode 15.0+ (to build from source)

Quick Start

git clone https://github.com/ServerWrestler/patience-chatbot.git
cd patience-chatbot
open Patience.xcodeproj

Then build and run with ⌘R. Prefer a prebuilt binary? Grab the latest release.

Documentation

Guide	What it covers
Scenario Testing	Scripted multi-turn conversation testing
Conversation Forensics	Log import, metrics, pattern & drift analysis
Conversation Forensics — Triage	Guardrail-failure classification via a local→frontier cascade
Adversarial Testing	AI-powered automated red teaming
Adversarial Prompts	OWASP / MITRE attack-pattern reference

Full index: DOCUMENTATION.md · Version history: CHANGELOG.md

Contributing

PRs welcome — see CONTRIBUTING.md. Project conventions and architecture for contributors (and AI assistants) live in CLAUDE.md.

License

MIT