Observed since 20 Oct, 20:12 UTC
No superseded metadata states collected. Unchanged observations do not create new revisions.
Local-first SAST engine with AST analysis, taint tracking, and AI-powered fix suggestions
Local-first static application security testing for Python, JavaScript, TypeScript, PHP, Java, Go, and basic C/C++ scanning. Aegis combines language parsers, security rules, taint tracking, suppression workflows, and CI-ready reports behind one production analysis path.
pip install aegis-ai-core
For repository development:
pip install -e ".[dev]"
aegis /path/to/project --format json
aegis /path/to/project --format html --output aegis-report.html
aegis /path/to/project --format sarif --output aegis-results.sarif
The equivalent module command is:
python -m src.scanner.cli /path/to/project --format json
Supported report formats are JSON, HTML, Markdown, and SARIF. Project scans can also use baselines, incremental mode, custom YAML rules, and optional AI-assisted remediation.
| Level | Languages | Analysis path |
|---|---|---|
| Full | Python, JavaScript/TypeScript, PHP, Java, Go | Language parser plus maintained AST/taint/DSL rules |
| Basic | C/C++ | Lightweight contextual rules for memory, string, pointer, and concurrency risks |
Unsupported languages return no findings rather than falling back to a generic cross-language regex scanner.
aegis rules init --language python --type sqli
aegis rules test .aegis/rules
aegis /path/to/project --rules-dir /path/to/project/.aegis/rules
See the rule authoring guide for schema, embedded TP/TN tests, loading boundaries, and contribution rules.
The scanner works without an AI provider. Install the optional OpenAI-compatible client only when generated remediation is needed:
pip install "aegis-ai-core[ai]"
Supported provider modes include DeepSeek, OpenAI, Ollama, and custom OpenAI-compatible endpoints. Source code is sent to a remote provider only after an explicit AI remediation action; local scans do not require a provider.
The 2026-07-12 clean-worktree baselines report:
| Target | Recall | Precision | F1 |
|---|---|---|---|
| DVWA | 100.0% | 44.2% | 0.61 |
| NodeGoat | 100.0% | 85.7% | 0.92 |
Reports record scanner/target revisions, ground-truth hashes, TP/FP/FN/TN,
duration, and peak RSS. These numbers are not generalized claims for every
framework or vulnerability category. See the
detection quality guide
and versioned reports in scripts/reports/ for the exact scope.
MIT
Download counts and popularity will appear here as the collector receives them.
| Version | Published | Direct dependencies | Open version |
|---|---|---|---|
| 1.5.0 | 27 Jul 2026 | 25 |
This describes our collected data; it is not a security assessment.
Observed since 20 Oct, 20:12 UTC
No superseded metadata states collected. Unchanged observations do not create new revisions.
package:58baf8eb7c93547ed6171a591a3bb7842f8c72654c2c6b6a3f28921b78fe4d5c