Keeps your graphify knowledge graph fresh without burning tokens.

◆ watching . — ignoring graphify-out/ node_modules/
◆ debounce 2000ms

~ changed src/auth/session.ts
↻ running graphify .
✓ graph rebuilt in 1.4s

~ changed docs/architecture.md
⚠  LLM rebuild triggered — these files will consume tokens:
  docs/architecture.md
  run graphify? [Y/n]

---

Why

graphify builds a knowledge graph from your codebase. But re-running it manually after every change is friction. graphify-chokidar handles all of that — including the very first build.

No graph yet? Save any file and it builds one from scratch.
Graph already exists? It updates only what changed.
Code change (.ts, .py, .go, …) → rebuilds instantly, no prompt, no LLM call.
Doc/media change (.md, .pdf, .png, …) → asks first, then rebuilds with LLM.

If a new change arrives mid-build, the previous build is cancelled and a fresh one starts immediately — but repeated LLM rebuilds are collapsed so you never get spammed with confirmation prompts.

---

Install

npm install -g graphify-chokidar

Or run without installing:

npx graphify-chokidar

Requires graphify to be installed and on your $PATH.

---

Usage

# Watch current directory
graphify-chokidar

# Watch a specific path
graphify-chokidar ./src

# Custom debounce (wait 5000ms after last change)
graphify-chokidar --debounce 5000

# Suppress all output except errors
graphify-chokidar --silent

# Run once on next change, then exit (useful in CI)
graphify-chokidar --once

All options

Flag	Default	Description
[path]	.	Directory to watch
-d, --debounce <ms>	2000	Wait after last change before rebuilding
-s, --silent	false	Only log errors
--once	false	Exit after the first rebuild (CI mode)

---

How it works

File change detected
      │
      ▼
  classifyFile()
      │
      ├─ ignore  →  do nothing
      │
      ├─ ast     →  cancel any running build → run graphify update <path>
      │              (no LLM, instant, safe to interrupt)
      │
      └─ llm     →  if build running: skip
                    else: prompt → run graphify update <path>

AST files (code): .ts .tsx .js .jsx .py .go .rs .java .c .cpp .cs .rb .swift .kt .scala .php .lua .zig .ex .vue .svelte .dart

LLM files (docs/media): .md .mdx .txt .rst .pdf .png .jpg .svg .mp4 .mp3 and more

Ignored always: graphify-out/ node_modules/ .git/ dist/ __pycache__/

---

Debouncing

Changes are batched. If you save 10 files in quick succession, one rebuild fires — not ten.

save fileA  →  start 2s timer
save fileB  →  reset timer
save fileC  →  reset timer
              ... 2s of silence ...
              → rebuild([fileA, fileB, fileC])

Adjust with --debounce <ms> to match how fast your editor flushes writes.

---

Agent skill

graphify-chokidar ships with a SKILL.md that lets AI agents (Claude, Codex, etc.) invoke the watcher as a slash command.

Install for Claude Code:

# Install globally first (if not already)
npm install -g graphify-chokidar

# Copy SKILL.md into Claude's skills directory
mkdir -p ~/.claude/skills/graphify-chokidar
cp "$(npm root -g)/graphify-chokidar/SKILL.md" ~/.claude/skills/graphify-chokidar/SKILL.md

Then in any Claude Code session:

/graphify-chokidar ./src

The agent will start the watcher, explain the log symbols, and handle start/stop for you.

---

Build from source

git clone https://github.com/yetanotheraryan/graphify-chokidar
cd graphify-chokidar
npm install
npm run build        # outputs to dist/

---

License

MIT