CodeSight Documentation

LLM-driven code analysis and security review CLI. Finds real bugs, not syntax.

01

Installation

pip install, configure your API key, and run your first scan in under a minute.
02

Commands

review, bugs, security, docs, explain, refactor, scan, diff, benchmark - all in one CLI.
03

Providers

OpenAI, Anthropic, Google Vertex AI, Ollama, or any OpenAI-compatible endpoint (OpenRouter, Groq, Azure, and more).
04

Benchmarks

CWE detection rates, comparison with Semgrep and CodeQL, false positive analysis.

What CodeSight Does

CodeSight sends your source code to large language models with tuned prompts for specific analysis tasks. Not a linter - it reads what the code is supposed to do and finds where it fails.

Code Review

Severity-tagged issues with line numbers, root cause, and concrete fixes.

Security Audit

CWE IDs, OWASP mapping, proof of concept scenarios, and remediation code.

Bug Detection

Logic errors, race conditions, resource leaks, off-by-ones, and edge cases.

Multi-Provider

15+ providers: OpenAI, Anthropic, Google, Ollama, OpenRouter, Groq, Azure, DeepSeek, and any OpenAI-compatible endpoint.

CI/CD Ready

SARIF output, exit codes, GitHub Action - drops into any pipeline.

Cost Tracking

Token usage and dollar cost per query. No surprises on your API bill.

Quick Example

Terminal
$ pip install codesight
$ export OPENAI_API_KEY="sk-..."
$ codesight security src/auth.py

╭─────────────────────────────────────────────╮
│ CodeSight  SECURITY  openai (gpt-5.4)        │
╰─────────────────────────────────────────────╯

[CRITICAL] SQL Injection - CWE-89
  ln 87 - f-string in SQL query
  Fix: Use parameterized queries

[CRITICAL] Hardcoded Secret - CWE-798
  ln 42 - JWT_SECRET = "admin123"
  Fix: Load from environment variable

[HIGH] Missing Auth - CWE-862
  ln 105 - /admin has no auth middleware

3 findings · 1,247 tokens · $0.003