WordPrint

AI content detector that scores how likely a passage was written by a large language model.

Paste text → a chosen judge model analyzes stylistic red flags → you get:

• Confidence score (0–100)
• Traffic light — green (likely human), yellow (uncertain), red (likely AI)
• Findings list — concrete red flags with severity, category, and quoted evidence
• Human signals — aspects that look authentically human

npm package: wordprint
CLI: wordprint

npm install -g wordprint
wordprint analyze -p xai "your text here…"

Privacy

Analyzed text is sent to a third-party model provider. WordPrint does not run a local judge model: the passage you paste or pipe is transmitted to whichever backend you select (xAI / Grok, Anthropic / Claude, or OpenAI) so that provider can score it.

That matters if you are checking client documents, unpublished manuscripts, or other confidential material. Prefer a provider and account whose data policies you accept; do not paste secrets or private drafts unless you are comfortable with that provider receiving them. Credentials stay local (~/.wordprint, env vars, or grok login); only the text under analysis is sent for scoring.

Stack

• Next.js (App Router)
• Vercel AI SDK with multi-provider backends:
  • xAI / Grok (@ai-sdk/xai) — default (grok-4.5)
  • Anthropic / Claude (@ai-sdk/anthropic)
  • OpenAI (@ai-sdk/openai)
• Local SQLite credentials store (~/.wordprint/credentials.db) via Settings panel or CLI
• SpaceXAI OAuth for Grok (same session as grok login / X Premium+ / SuperGrok)
• Zod structured output
• Tailwind CSS

Prerequisites

• Node.js 22.5+ (uses built-in node:sqlite)
• At least one model backend:

| Provider | How to authenticate | |----------|---------------------| | Grok (default) | Settings / wordprint config, or grok login, or XAI_API_KEY | | Claude | Settings / wordprint config, or ANTHROPIC_API_KEY | | OpenAI | Settings / wordprint config, or OPENAI_API_KEY |

Claude Code / Claude.ai subscription login is not reused — use an API key (or the Settings panel).

Install

Option A — npm global CLI (recommended for day-to-day use)

npm install -g wordprint

Commands available:

| Command | Purpose | |---------|---------| | wordprint / wordprint | CLI entry (same binary) | | wordprint analyze … | Score text (no server needed) | | wordprint status | Auth + model catalog (no server needed) | | wordprint dev | Start web UI at http://localhost:3000 | | wordprint start | Production web UI (builds if needed) | | wordprint open | Open the UI in a browser |

One-shot without a global install:

npx wordprint analyze -p anthropic "In today's rapidly evolving landscape…"
npx wordprint status
npx wordprint dev

Option B — from source (development)

git clone <repo-url> wordprint
cd wordprint
npm install

Configure credentials

Keys can live in three places (highest priority first):

1. Environment — XAI_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY (or .env.local for the web app)
2. Local SQLite — ~/.wordprint/credentials.db (Settings panel or wordprint config)
3. Grok OAuth only — grok login → ~/.grok/auth.json

Web UI (Settings panel)

1. Start the app: wordprint dev or npm run dev
2. Open http://localhost:3000
3. Click API keys in the header
4. xAI OAuth: click Import from Grok login (after grok login), or paste a token
5. Optionally paste xAI / Anthropic / OpenAI API keys → Save keys
6. Toggle Prefer OAuth over xAI API key when you want SuperGrok OAuth first

Secrets are written to SQLite on disk (~/.wordprint/credentials.db, mode 0600). The API never returns full keys — only a masked hint.

CLI config (same SQLite store)

wordprint config                              # keys + OAuth status
wordprint config oauth import                 # from `grok login` → SQLite
wordprint config oauth refresh
wordprint config oauth prefer on              # use OAuth before API key
wordprint config oauth clear
wordprint config set --xai xai-...
wordprint config set --anthropic sk-ant-...
wordprint config set --openai sk-...
wordprint config clear --openai
wordprint config clear --all

Environment / .env.local

cp .env.example .env.local

# XAI_API_KEY=xai-...
# ANTHROPIC_API_KEY=sk-ant-...
# OPENAI_API_KEY=sk-...
# WORDPRINT_PROVIDER=xai
# WORDPRINT_MODEL=grok-4.5
# WORDPRINT_DATA_DIR=~/.wordprint

Or Grok OAuth only:

grok login

How to run

CLI only (no web server)

# Auth check
wordprint status

# Analyze (standalone — talks to the model provider directly)
wordprint analyze "paste at least 40 characters of text here…"
wordprint analyze -p anthropic -m claude-opus-5 "…"
wordprint analyze -p xai notes.txt
echo "…" | wordprint analyze -p anthropic -

JSON is printed to stdout (pipe to jq if you like).

Web UI (development)

# from a global install
wordprint dev

# or from a source checkout
npm run dev
# or
./bin/wordprint dev

Open http://localhost:3000.

1. Choose Provider and Model under the text box
2. Paste text (or use sample buttons)
3. Click Analyze

Production web UI

wordprint build
wordprint start
# or: npm run build && npm run start

Override port: PORT=4000 wordprint dev.

Inside Claude Code or Grok

# after: npm install -g wordprint
# and credentials in the environment / grok login

wordprint status
wordprint analyze -p xai "text of at least 40 characters…"
wordprint analyze -p anthropic "…"
wordprint dev    # optional web UI

Providers & model switch

| Provider | Default model | Auth | |----------|---------------|------| | xai | grok-4.5 | env → SQLite Settings → grok login OAuth | | anthropic | claude-opus-5 | env → SQLite Settings | | openai | gpt-4.1 | env → SQLite Settings |

• Catalog: config/providers.ts
• Forensic prompt / thresholds: config/analysis.ts

On each Grok analysis the server/CLI:

1. Loads the OAuth access token from ~/.grok/auth.json
2. Silently refreshes via https://auth.x.ai/oauth2/token when near expiry
3. Retries once on 401 after a forced refresh
4. Falls back to XAI_API_KEY only if no OAuth session exists

Publish to npm (maintainers)

Package name is wordprint (unscoped, public).

# 1. Log in (once)
npm login

# 2. Sanity-check the tarball
npm run pack:check
# or: npm pack --dry-run

# 3. Bump version when ready
npm version patch   # or minor / major

# 4. Publish
npm publish

After publish, anyone can:

npm install -g wordprint
wordprint --help

Notes

• Update repository / homepage / bugs in package.json to your real GitHub URL before publishing.
• CLI binary name is wordprint.
• Do not commit secrets; .env* is gitignored (.env.example is safe to ship).

How scoring works

The checker prompt lives in config/analysis.ts (analysisConfig.systemPrompt). The same forensic instructions are sent to whichever provider/model you select.

Default bias: skeptical toward “human” — structural patterns and absence of human traits outweigh single-word markers (which vary by language and model).

Seven forensic dimensions:

| # | Dimension | What it catches | |---|-----------|-----------------| | 1 | Lexical predictability | Generic collocations, institutional lexicon, prompt-shaped phrasing (language-adapted) | | 2 | Rhythm / burstiness | Simulated vs authentic sentence variation | | 3 | Structure / symmetry | Intro→N points→outro, tricolons, masked completism | | 4 | Tone / voice | Hedging, false assertiveness, missing lived experience | | 5 | Openings / closings | Over-polished hooks, punchlines, packing-peanut transitions | | 6 | Terminological density | Decorative vs functional jargon | | 7 | Perfection | Frictionless prose with zero micro-imperfections |

Verdicts: ai_generated · human_written · human_edited_ai

Thresholds:

| Score | Light | Meaning | |------:|:-----:|---------| | 0–34 | Green | Likely human | | 35–64 | Yellow | Uncertain / mixed / hybrid | | 65–100 | Red | Likely AI-generated |

This is a heuristic, not a watermark detector or proof of authorship. Short, technical, or heavily edited text is often ambiguous.

API (web server mode)

Base URL when running locally: http://localhost:3000.

GET /api/auth/status

Returns per-provider auth status and the model catalog.

POST /api/analyze

curl -s http://localhost:3000/api/analyze \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Your passage here (min 40 characters)…",
    "provider": "anthropic",
    "model": "claude-opus-5"
  }' | jq

provider and model are optional (fall back to WORDPRINT_* env or catalog defaults).

Success shape:

{
  "analysis": {
    "confidenceScore": 82,
    "riskLevel": "red",
    "verdict": "ai_generated",
    "summary": "…",
    "humanSignals": [],
    "findings": []
  },
  "provider": "anthropic",
  "model": "claude-opus-5",
  "auth": {
    "provider": "anthropic",
    "source": "api_key",
    "email": null
  }
}

Troubleshooting

| Symptom | Fix | |---------|-----| | “No xAI credentials” | Run grok login or set XAI_API_KEY, then retry | | “No Anthropic credentials” | Export ANTHROPIC_API_KEY (or put it in .env.local for the web app) | | CLI works, web UI does not see keys | Web app reads .env.local in the project/package root; restart after editing | | Port 3000 in use | PORT=3001 wordprint dev | | Global install cannot write | Use npx wordprint … or fix npm global prefix permissions |

New installs use ~/.wordprint only. If you previously used this tool under an older name and still have credentials under ~/.deslopper, WordPrint will fall back to that directory when ~/.wordprint does not exist yet (or set WORDPRINT_DATA_DIR).

License

MIT