MonkeyScore - AI Assurance Metrics Analyzer

Measure 15 AI Assurance Metrics from any LLM output. Inline annotations, mode presets, trend charts. 100% local.

sidemind.ai is a Chrome extension that measures 15 AI Assurance Metrics from any LLM output. 100% local. Zero data leaves your browser. What's New in v2.2.2 🔍 Inline Sentence Annotations Highlights problematic text directly inside LLM outputs — hallucinations (red), PII leaks (pink), security injections (dark red), bias/stereotypes (amber), speculative language (cyan), and vague attributions (indigo). Skips code blocks to avoid false positives. Toggle visibility with the eye icon in the results header. 📈 Metric Trend Charts New "Trends" view in the History tab. Filter by platform, see your overall score plotted over time as an SVG line chart, and browse a 15-card sparkline grid showing each metric's trajectory with ▲/▼/→ deltas from your previous session. 🎯 Analysis Mode Presets Six scoring modes — General, Research, Developer, Compliance, Healthcare, and Marketing — each with tailored weight profiles. Modes combine with platform weights using an additive model (capped at 1.60×). Persists across sessions, appears as selectable chips in both the popup and on-page panel, and shows as a badge in results and exported reports. Key features */Platform Auto-Detection The extension automatically recognizes which AI platform you're using. 11 major platforms supported- scoring adjusts accordingly. */Platform-Specific Weighted Scoring Not all AI models fail the same way. Each platform now has a tailored scoring profile based on known failure modes- hallucination, groundedness, policy compliance, security, and more are weighted differently depending on where you're analyzing. */Prompt-Response Alignment Provide the original prompt and the engine measures how well the output actually addresses it- term coverage, keyphrase matching, intent detection, and a 0-100 alignment score. It even lists the specific prompt terms the response missed. */Auto Prompt Extraction On supported platforms, the extension automatically detects your prompt from the page. No copy-paste needed. */Context-Aware Metrics 5 of the 15 metrics now factor in platform and prompt context for deeper, more meaningful scoring. */Smarter Performance Debounced DOM scanning means the extension stays fast even during streaming responses. This started as a simple idea: what if you could score any AI output the way you'd audit a model in production? 15 metrics. 6 categories. Accuracy, hallucination, groundedness, bias, robustness, drift, explainability, compliance, security, privacy, human override, reliability, latency, and business impact. All running locally in your browser.