lintsinghua/DeepAudit
DeepAudit
DeepAudit is a privacy-preserving code audit platform that combines multiple specialized AI agents to identify and verify security vulnerabilities in source code. It functions as a local LLM vulnerability scanner, an automated security report generator, and a sandboxed exploit verifier, all operating entirely within an internal network to keep sensitive code and data on premises.
What distinguishes DeepAudit is its multi-agent cooperative approach: teams of AI agents jointly plan, analyze, and cross-check findings across the codebase, moving beyond single-pass scanning. The platform also sandbox-confirms exploitability by running proof-of-concept scripts in an isolated environment, reducing false positives while validating real-world risk. All analysis runs locally using on-premises large language models, ensuring no data leaves the internal network.
DeepAudit complements these differentiators with configurable audit scopes — allowing teams to target specific directories and file patterns — and rule-based vulnerability matching against known patterns. After analysis, it automatically produces a structured report with severity levels and remediation steps.
Features
- Cooperative AI Auditors - Teams of specialized AI agents cooperatively plan, analyze, and verify vulnerabilities across the codebase.
- Security Audit Reports - Aggregates findings, severity levels, and remediation steps into a structured report automatically after analysis.
- Local Model Runtimes - Perform analysis locally using internal-network models to ensure data never leaves the premises.
- Local Inference Engines - Runs large language models entirely within the internal network for analysis, ensuring data never leaves the premises.
- Privacy-Preserving Code Auditors - Runs entirely within the internal network, ensuring source code and analysis remain on premises.
- Sandboxed Exploit Verifiers - Executes PoC exploit scripts in a controlled sandbox to confirm vulnerability exploitability without risk.
- Security and Vulnerability Scanning - Scan source code and configuration files for security vulnerabilities using rule-based analysis and known databases.
- Vulnerability Detection - Compare code against known vulnerability patterns to detect security flaws and misconfigurations.
- Security Report Generation - Producing a comprehensive report with findings, severity levels, and remediation steps automatically after the audit completes.
- LLM-Orchestrated Scanners - Automates vulnerability scanning by orchestrating local large language models for private code analysis.
- Source Code Vulnerability Scanning - Scanning source code and configuration files for known security vulnerabilities using rule-based pattern matching and vulnerability databases.
- Audit Scope Configurations - Specify which directories and file patterns the audit scans to focus analysis on relevant code.
- Vulnerability Matching - Scans code by comparing against a database of known vulnerability patterns and signature rules.
