Dev-Crew

Code graph intelligence for AI-assisted development
_{Blast radius analysis, smart context selection, and 40 specialized agents. Free & open source.}

What it does

Dev-Crew builds a structural graph of your codebase — functions, classes, imports, call relationships — then uses it to:

1. Blast radius analysis — Change a file, see every function and file affected
2. Smart context selection — Send only relevant code to AI, saving tokens
3. Static analysis — TypeScript errors, ESLint issues, security patterns — before AI even runs
4. 40 specialized agents — Code review, testing, security, architecture, DevOps, and more

npx dev-crew scan              # Instant project analysis (no AI needed)
npx dev-crew review @src/api.ts   # AI code review with graph context
npx dev-crew security @src/       # OWASP security audit

Quick start

npm install -g dev-crew

# Check your setup
dev-crew doctor

# Scan your project (zero AI, runs instantly)
dev-crew scan

# Start interactive mode
dev-crew

AI providers (needed for agent commands, not for scan):

• Claude Code — npm install -g @anthropic-ai/claude-code
• Aider — pip install aider-chat
• Ollama — local models, no API key

The scan command

Run dev-crew scan on any project for instant, AI-free analysis:

  Dev-Crew Project Scan — zero AI, pure analysis
  ──────────────────────────────────────────────────
  Target: /home/user/my-app

  Project   my-app
  Language  TypeScript  Framework  Next.js
  Test fw   vitest

  Code Graph
    247 symbols across 38 files
    512 connections mapped
    Languages: TypeScript, JavaScript

  Hotspot Files (most connections = highest blast radius)
     47 connections  src/lib/database.ts
     31 connections  src/api/routes.ts
     28 connections  src/auth/session.ts

  Static Analysis
    2 errors
    5 warnings
    ✗ src/api/routes.ts:45 Missing return type
    ⚠ src/auth/session.ts:12 console.log in production code

  ──────────────────────────────────────────────────
  Completed in 340ms — no AI tokens used

Code graph intelligence

The code graph engine parses your source files and builds a structural map:

• Functions, classes, types, imports extracted via regex (no tree-sitter dependency)
• Call relationships traced across files
• 6 languages — TypeScript/JavaScript, Python, Go, Java/Kotlin, Rust
• Incremental updates — only re-indexes changed files

When you run an agent command, dev-crew uses the graph to:

1. Find the blast radius of files you're reviewing
2. Select only the relevant connected files as context
3. Include static analysis findings before sending to AI

This means the AI gets focused, relevant context instead of your entire codebase.

All agents

Interactive mode

Run dev-crew with no arguments to start the interactive REPL:

dev-crew> review the auth module for security issues
dev-crew> /graph src/api/routes.ts    # show blast radius
dev-crew> /analyze                      # run static analysis
dev-crew> /diff-review                  # review uncommitted changes

Natural language routing sends your request to the right agent automatically.

Configuration

dev-crew init                    # Create .dev-crew.yml config
dev-crew config set provider claude  # Set AI provider
dev-crew feedback review "always check for SQL injection"  # Teach agents

How it works

Your Code → Code Graph Engine → Blast Radius → Smart Context Selection
                                                        ↓
                                              Static Analyzer
                                                        ↓
                                              AI Provider (Claude/Aider/Ollama)
                                                        ↓
                                              Structured Response

Every agent follows the same pipeline:

1. Build code graph for target files
2. Calculate blast radius (what's affected)
3. Select smart context (only relevant files)
4. Run static analysis (real tsc/eslint errors)
5. Send enriched prompt to AI provider
6. Parse and format structured response

Contributing

Issues and PRs welcome: github.com/vasoyaprince14/dev-crew

License

MIT — Prince Vasoya