DeepSecrets scans code, not a stream of bytes.
It bridges the gap between classic regex search and full-scale commercial SAST tools by combining pattern matching with deep semantic code analysis.
🚀 Key Features
True Code Understanding: Uses advanced lexing and parsing to analyze structure across 500+ languages and formats.
Semantic Filtering: Secret candidates are always semantically correct.
Context-Aware Confidence Scoring: Dynamically rates candidates using naming layouts, value entropy, and semantic "naturalness" to reduce noise.
Hybrid Strategy: Combines semantic analysis and dangerous variable detection with classic, regex-backed search.
Zero-Knowledge Secret Search (Experimental): Introduces the HashedSecret Engine. Provide hashed values of known secrets, and DeepSecrets will safely locate them in plain-text without storing the secrets themselves.
Performed using DeepSecrets v2.1.1
Recall (Valid Secrets Found): 93%
Precision (Fewer False Alarms): 69%
F1-Score (Overall Balance): 79%
The tool also discovered about 10K actionable findings beyond the scope of the SecretBench dataset.
Project Resources
Project Information
Slack