Codixing
Website: codixing.com · Docs
Code retrieval engine that saves your AI agent 73% of its token budget. Replaces grep with ranked, AST-aware search — so models spend tokens reasoning, not reading.
Install
curl -fsSL https://codixing.com/install.sh | sh
Installs codixing to /usr/local/bin. macOS (Apple Silicon) and Linux (x86_64). Binaries also on the releases page.
Claude Code plugin (optional)
claude plugin marketplace add ferax564/codixing
claude plugin install codixing@codixing
Adds 5 slash commands: /codixing-setup, /codixing-explore, /codixing-review, /codixing-preflight, /codixing-release.
MCP server (optional — for Cursor, Windsurf, Continue.dev, Codex)
Add to your project's .mcp.json:
{
"mcpServers": {
"codixing": {
"command": "npx",
"args": ["-y", "codixing-mcp", "--root", ".", "--medium", "--no-daemon-fork"]
}
}
}
Or for OpenAI Codex CLI: codex mcp add codixing -- codixing-mcp --root .
Why Not Just Grep?
AI coding agents use grep, find, and cat for code navigation. These tools return everything, always — a single rg b2Vec2 on a real codebase returns 2,240 hits (225 KB), burning context before any reasoning happens.
Codixing returns the top 20 results in 1.3 KB — same signal, 99% less waste.
The cost of noise
Tested on 6 real-world repos (tokio, ripgrep, axum, django, fastapi, react — 9,493 files):
| Metric | grep/cat/find | Codixing | Savings |
|---|
| Tool calls per session | 58 | 26 | 55% fewer |
| Output tokens | ~84,600 | ~22,900 | 73% fewer |
| Est. cost (Opus @ $15/M) | $1.27 | $0.34 | $0.93/session |
At 50 agent sessions/day, that's $1,400/month back in your pocket — and the agent finds the right code more often.
What you get
| Capability | grep/rg | Codixing |
|---|
| Bounded, ranked output | No | Yes (BM25 + PageRank) |
| Symbol definitions (not just mentions) | No | Yes (AST-parsed symbol table) |
| Dependency graph queries | No | Yes (transitive imports, call graph) |
| Natural language search | No | Yes (BM25 + optional embeddings) |
| Token budget management | No | Yes (auto-truncation) |
Getting Started
60-second setup
# 1. Install
curl -fsSL https://codixing.com/install.sh | sh
# 2. Index your project
codixing init .
# ✓ Indexed 2,847 files, 14,203 chunks, 8,891 symbols in 1.2s
# 3. Search
codixing search "authentication handler"
# ► src/auth/handler.rs:42 [score: 0.94]
# pub fn handle_auth_request(req: Request) -> Result<Token>
That's it. Your agent now uses ranked search instead of grep.
CLI commands
# Search (natural language or symbol names)
codixing search "error handling middleware"
# Symbol lookup (definitions only, not mentions)
codixing symbols Engine
# Dependency graph
codixing callers src/engine.rs # who imports this file?
codixing callees src/engine.rs # what does this file import?
# Keep index fresh (re-indexes only changed files)
codixing sync
# Architecture map
codixing graph --token-budget 4000
Hybrid search (optional)
BM25-only works great for most queries. For natural-language queries ("how does the auth flow work?"), add semantic embeddings:
codixing init . --model bge-small-en # one-time, ~2 min
codixing search "how does auth work" --strategy fast
Requires ONNX Runtime (pip install onnxruntime or download from GitHub).
CLI Commands
19 commands for code intelligence:
codixing search "query" # Semantic code search
codixing symbols Widget # Find symbol definitions
codixing usages add_chunk # Find call sites and imports
codixing callers src/engine.rs # Who imports this file
codixing callees src/engine.rs # What this file imports
codixing graph --map # Architecture overview
codixing init . # Index a project
codixing sync # Incremental re-index
codixing audit # Find stale files
Full reference: codixing.com/docs
MCP server (optional)
For editors with MCP support, the codixing-mcp binary exposes 57 JSON-RPC 2.0 tools:
ferax564