A powerful CLI tool that scans your codebase to detect LLM SDK usage, AI framework integrations, exposed API tokens, and hardcoded secrets — all in one command.

Zero dependencies. 145 detection patterns. Works with Node.js 18+.

Features

• LLM SDK Detection — OpenAI, Anthropic, Google Gemini, Cohere, Mistral, Groq, Replicate, Together AI, AWS Bedrock, Azure OpenAI, Ollama, LiteLLM, DeepSeek, and more
• AI Framework Detection — LangChain, LlamaIndex, Haystack, AutoGen, CrewAI, Vercel AI SDK, DSPy, Semantic Kernel, LangGraph, vLLM, and more
• AI Token Scanning — Detects hardcoded keys for OpenAI (sk-), Anthropic (sk-ant-), Hugging Face (hf_), Google (AIzaSy), AWS (AKIA), Groq (gsk_), Replicate (r8_), LangSmith (ls__), and more
• Generic Secret Scanning — Stripe, Twilio, SendGrid, GitHub, GitLab, Slack, Discord, Telegram, database URIs, private keys (RSA/SSH/PGP), JWTs, and 50+ more patterns
• Smart Filtering — Ignores .env files (they're meant to hold secrets) and filters out SDK/framework mentions in READMEs, docs, and example files
• Model Reference Detection — Spots references to GPT-4, Claude, Gemini, Llama, Mistral, and other models
• API Endpoint Detection — Finds direct API calls to LLM providers
• Multiple Output Formats — Rich console output, JSON, and SARIF (for CI/CD)
• Jupyter Notebook Support — Parses .ipynb files to scan code cells
• Token Masking — Automatically masks detected secrets in output for safety

Quick Start

# Run directly with npx (no install needed)
npx ai-scanner ./my-project

# Or install globally
npm install -g ai-scanner
ai-scanner ./my-project

Usage

# Scan current directory (AI + generic secrets)
ai-scanner

# Scan a specific directory
ai-scanner ./my-project

# Security-focused: only scan for exposed tokens & secrets
ai-scanner --tokens-only

# AI patterns only (skip Stripe, GitHub tokens, etc.)
ai-scanner --ai-only

# Include .env files in scan (skipped by default)
ai-scanner --scan-env

# Output as JSON
ai-scanner --json

# Save JSON report
ai-scanner -o report.json

# Save SARIF report (GitHub Actions, VS Code, etc.)
ai-scanner --sarif results.sarif

# CI mode: exit with code 1 if critical/high findings
ai-scanner --exit-code

# Skip endpoint/model detection for faster scan
ai-scanner --no-endpoints --no-models

# Combine options
ai-scanner ./src --tokens-only --exit-code --json

Smart Filtering

ai-scanner is context-aware and avoids noisy false positives:

| File type | SDK/Framework mentions | Exposed tokens & secrets | |---|---|---| | Source code (.js, .py, .go, etc.) | ✅ Reported | ✅ Reported | | README, docs, markdown | ❌ Ignored (just documentation) | ✅ Reported | | examples/, samples/, docs/ dirs | ❌ Ignored (just examples) | ✅ Reported | | .env files | ❌ Skipped by default | ❌ Skipped by default | | .env files with --scan-env | — | ✅ Reported |

This means scanning a project like an LLM gateway — which naturally references many SDKs in its README and examples — won't flood you with 100+ informational findings.

CI/CD Integration

GitHub Actions

- name: Scan for exposed tokens & secrets
  run: npx ai-scanner --tokens-only --exit-code --sarif results.sarif

- name: Upload SARIF
  uses: github/codeql-action/upload-sarif@v3
  with:
    sarif_file: results.sarif

Pre-commit Hook

# .husky/pre-commit
npx ai-scanner --tokens-only --exit-code

Severity Levels

| Level | Meaning | Example | |-------|---------|---------| | 🚨 CRITICAL | Exposed key with known prefix | sk-ant-abc123..., sk_live_..., ghp_... | | ⚠️ HIGH | Likely hardcoded credential | api_key = "...", JWT tokens, DB connection strings | | ℹ️ INFO | SDK/framework usage (awareness) | import openai |

Supported Detections

AI Tokens (20+)

OpenAI keys, Anthropic keys, Google AI keys, HuggingFace tokens, Cohere keys, Replicate tokens, Groq keys, Mistral keys, AWS access keys, LangSmith keys, Fireworks keys, W&B keys, Bearer tokens, Authorization headers

Generic Secrets (59 patterns)

| Category | Detections | |---|---| | Payment | Stripe (live, restricted, webhook), Square, PayPal Braintree | | Communication | Twilio, SendGrid, Mailgun, Mailchimp, Postmark | | Source Control | GitHub (PAT, fine-grained, OAuth, app), GitLab, Bitbucket, CircleCI | | Cloud | GCP service accounts, DigitalOcean, Heroku, Vercel, Netlify, Cloudflare | | Messaging | Slack (bot, user, webhook), Discord (bot, webhook), Telegram | | Database | Postgres/MySQL/MongoDB/Redis/AMQP URIs, Supabase, Firebase, PlanetScale | | Auth | Auth0, Okta, Clerk | | Monitoring | Datadog, Sentry DSN, New Relic, Segment, Mixpanel | | Crypto | RSA, EC, DSA, SSH, PGP private keys | | Generic | Passwords, client secrets, connection strings, JWTs |

LLM SDKs (23)

OpenAI, Anthropic, Google Generative AI, Vertex AI, Cohere, Mistral, Hugging Face, Replicate, Together AI, Groq, AWS Bedrock, Azure OpenAI, Ollama, LiteLLM, Fireworks AI, Perplexity, DeepSeek

AI Frameworks (24)

LangChain, LangGraph, LangSmith, LlamaIndex, Haystack, AutoGen, CrewAI, Semantic Kernel, Vercel AI SDK, DSPy, Guidance, Instructor, Chainlit, Flowise, Embedchain, Promptflow, Spring AI, vLLM, TensorRT-LLM, MLflow, Weights & Biases, Smolagents

Examples

Scan any public GitHub repo:

git clone --depth 1 https://github.com/user/repo /tmp/repo
npx ai-scanner /tmp/repo

Or use the helper scripts in examples/ — GitHub repo scanner, batch scanning, pre-commit hooks, GitHub Actions workflow, and using ai-scanner as a Node.js library.

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

The easiest way to contribute is adding new detection patterns — see the guide for the pattern format.

License

MIT