Show HN: How to analyze your LLM output – A behavioural health monitor for LLMs
A new tool called PSA (Posture Sequence Analysis) allows users to analyze the behavior of language models without needing access to their internal weights. It provides real-time insights into model behavior, detecting anomalies and shifts over time through various classifiers. The platform integrates multiple subsystems for comprehensive monitoring and analysis of AI interactions.
- ▪PSA enables behavioral analysis of language models from an external perspective without requiring API keys or model access.
- ▪The tool features 24 metrics for detecting regime shifts, anomalies, and behavioral drift in real-time.
- ▪It includes a multi-agent risk detection system and privacy-compliant incident archiving.
Opening excerpt (first ~120 words) tap to expand
Black-box behavioral analysis PSA Posture Sequence Analysis Measure what your language model is doing — from the outside. Deterministic metrics, posture analysis, and multi-agent risk detection. No access to weights required. Get Started Sign In Public Spec Theoretical Foundations CPF → PSA Every PSA metric traces to a specific indicator in the Cybersecurity Psychology Framework — a published taxonomy of 100 pre-cognitive vulnerabilities (Canale, 2025). See the full theoretical genealogy: which CPF indicators each signal measures, and why. How it works 1 Send text via our API or through our web app. Provide any model response — no API keys or model access needed. 2 Analyze PSA posture classifiers and agentic graph analysis — computed in real-time.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at PSA — Silicon Psyche Labs.
Discussion
0 commentsMore from PSA — Silicon Psyche Labs
