Cut your AI coding costs by 95%. Vectorless code index for intelligent context retrieval.
Every time you paste your codebase to ChatGPT or Claude, you're burning tokens.
codeindexgives AI exactly the context it needs — nothing more.
You're paying $20-100/month on AI coding tools, but here's the dirty secret:
| What You're Doing | What It Costs |
|---|---|
| Pasting entire files to ChatGPT | ~50,000 tokens/query |
| Claude analyzing your codebase | $0.03-0.15/query |
| Context window overflow errors | Priceless frustration |
The average developer wastes 80% of their AI budget on irrelevant context.
Every CTRL+C → CTRL+V to an AI chat burns tokens on code that has nothing to do with your question. You're paying for context soup when you only need one ingredient.
codeindex builds a hierarchical tree index of your codebase. When you ask a question, LLM reasoning selects exactly which modules, files, and symbols are relevant — then returns only that.
Result: ~1,000-3,000 tokens per query instead of 50,000+
Before: Paste 200 files (50KB) → Ask "fix my login bug"
After: Paste 3 files (2KB) → Same answer
| codeindex | Vector Embeddings | Manual Copy-Paste | |
|---|---|---|---|
| Tokens/query | ~2 KB | ~100 KB | 50+ KB |
| Setup | 2 minutes | 30 minutes | 0 |
| Accuracy | LLM reasoning | Cosine similarity | You're guessing |
| Updates | Instant | Re-embed entire codebase | Manual |
| Privacy | 100% local | Data leaves machine | You're in control |
| Cost | Free (MIT) | $20-100/month | Free (wasteful) |
- Vectorless Architecture — No embeddings, no external storage, no recurring costs
- LLM Reasoning — Asks "what code is relevant?" instead of "what code is similar?"
- Multi-Language — TypeScript, Python, Go, Rust, Java, C#, C++, PHP, Swift
- Incremental Updates — Only re-indexes changed files
- IDE Integration — HTTP API for VSCode, JetBrains, Neovim, Claude Code, Cursor
- Git Hook Ready — Auto-update index after commits
- Production Ready — API key auth, rate limiting, structured logging
# 1. Install
pnpm install -g @codeindex/cli
# 2. Setup (one-time)
codeindex setup
# 3. Index your project
cd your-project
codeindex index .
# 4. Query!cd
codeindex query "How does the auth module work?"That's it. No cloud signup, no API costs, no vector database to maintain.
┌─────────────────────────────────────────────────────────────┐
│ Your Query │
│ "How does auth validation work?" │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ codeindex Index │
│ │
│ Project │
│ └── src/ │
│ ├── auth/ ← LLM selects this module │
│ │ ├── login.ts ← And these files │
│ │ └── validators.ts │
│ └── users/ │
│ └── ... │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Response │
│ │
│ auth/validators.ts + auth/login.ts (2KB) │
│ "Here are the validation functions..." │
└─────────────────────────────────────────────────────────────┘
| Project Size | Before | After | You Save |
|---|---|---|---|
| 50 files | 15,000 tokens | 800 tokens | $0.05/query |
| 200 files | 60,000 tokens | 1,500 tokens | $0.15/query |
| 500+ files | Context overflow | 2,500 tokens | Priceless |
At 10 queries/day, that's $15-45/month saved.
# Add to ~/.claude/tools/codeindex.sh
curl -s -X POST http://localhost:3131/query \
-d '{"query": "$1", "maxTokens": 3000}' | jq -r '.context'Add to your .cursorrules:
When you need codebase context, run:
curl -s -X POST http://localhost:3131/query -d '{"query": "YOUR_QUESTION"}' | jq -r '.context'
Use codeindex query "your question" to get relevant code context.
Start server: codeindex serve . --port 3131
TypeScript • Python • Go • Rust • Java • C# • C++ • PHP • Swift
Multi-language support is built-in. Each language is handled by its own dedicated adapter.
Run all tests across the workspace:
pnpm test # Run all tests
pnpm test:watch # Run in watch modecodeindex/
├── packages/
│ ├── core/ # Tree index, retrieval, storage
│ ├── cli/ # CLI & HTTP server
│ └── adapter-*/ # Language-specific parsers
└── docs/ # Documentation
No external dependencies. Everything runs locally.
Contributions welcome! See docs/ for architecture details.
MIT — Use it freely, even in commercial projects.
If codeindex saves you time and money, consider buying me a coffee ☕
Stop paying for context you don't need. Start using codeindex.