WordingKit

WordingKit is a Raycast extension for rewriting selected text in any macOS app. It supports local Ollama models and optional cloud providers, so you can keep everyday rewriting local while still using OpenAI, Anthropic, or Groq when you need them.

Features

• Rewrite selected text from the active macOS app and paste the result back with one action.
• Use built-in language-preserving editing modes for grammar fixes, clarity, shortening, formal tone, work chat, social posts, and more.
• Create custom modes with a provider, model, system prompt, temperature, and token limit.
• Sort modes manually or by last use.
• Run against local Ollama by default, with optional OpenAI, Anthropic, and Groq API support.
• Keep provider credentials in Raycast Preferences instead of source files.

Requirements

• macOS with Raycast installed.
• Node.js 22 or newer for local development.
• Ollama running at http://127.0.0.1:11434 for the default local provider.
• Optional API keys for OpenAI, Anthropic, or Groq if you use cloud modes.

Installation from source

Clone the repository, install dependencies, and start Raycast development mode:

git clone https://github.com/suregoodru/wordingkit.git
cd wordingkit
npm ci
npm run dev

Raycast opens the extension in development mode. Configure provider keys and the Ollama URL in Raycast Preferences for WordingKit.

Development

npm ci
npm run dev

Commands

Command	Purpose
npm run dev	Start Raycast development mode.
npm test	Run Node.js tests for provider contracts, mode storage, manifest expectations, and UI flow guards.
npm run lint	Run Raycast lint checks.
npm run build	Build and validate the Raycast extension.
npm run eval:generate-modes	Regenerate manual evaluation modes from src/tones.ts.
npm run eval:rewrites	Run opt-in manual rewrite evaluation through configured providers.

Configuration

WordingKit reads provider configuration from Raycast Preferences:

Preference	Required	Used for
openaiApiKey	No	OpenAI modes.
anthropicApiKey	No	Anthropic modes.
groqApiKey	No	Groq modes.
ollamaUrl	No	Ollama endpoint, defaulting to http://127.0.0.1:11434.

The default built-in modes use Ollama model qwen3:14b. You can edit or replace modes from WordingKit Settings.

Privacy and security

• WordingKit has no backend service and does not store provider responses on a remote server.
• Selected text is sent only to the provider configured for the active mode.
• Ollama modes call your local Ollama server.
• Cloud modes call the selected provider API directly from the Raycast extension.
• API keys are configured as Raycast password preferences and must not be stored in repository files, logs, screenshots, or issue reports.
• Provider error messages are sanitized so configured API keys are replaced with [REDACTED] before they are shown.

Repository hygiene

Before publishing screenshots, evaluation messages, or issue examples, make sure they do not contain private text, customer data, access tokens, or internal URLs. Generated manual evaluation reports live in manual-eval/reports/ and are ignored by Git.

Manual rewrite evaluation

Manual evaluation is an opt-in workflow for comparing rewrite quality across the default WordingKit modes and selected Ollama models. It is not part of npm test, npm run build, npm run dev, lint, or the Raycast runtime.

Editable inputs

• manual-eval/messages.json — source messages to rewrite. Keep this as an array of strings. The current set is intentionally small and mixed by length: one short message, three medium messages, and three longer technical messages.
• manual-eval/models.json — Ollama model names to evaluate, for example qwen3:14b.

Generated inputs and reports

• manual-eval/modes.json — generated from manual-eval/models.json and src/tones.ts. Do not edit it by hand unless you are debugging the generator.
• manual-eval/reports/ — generated Markdown and JSON reports. This directory is ignored by Git.

Commands

Regenerate manual-eval/modes.json without calling any model:

npm run eval:generate-modes

Run the full rewrite evaluation:

npm run eval:rewrites

npm run eval:rewrites regenerates manual-eval/modes.json first, then runs every message through every generated mode sequentially. This keeps the payloads aligned with the real WordingKit prompts in src/tones.ts and reduces pressure on local Ollama.

Environment

For Ollama-only runs, make sure Ollama is running. The script uses OLLAMA_URL when it is set; otherwise the provider default is used.

Cloud providers are supported by the same provider path as the app, but the manual generator currently creates Ollama modes. If you add cloud modes manually, set the matching environment variables before running evaluation:

OPENAI_API_KEY=...
ANTHROPIC_API_KEY=...
GROQ_API_KEY=...
OLLAMA_URL=http://127.0.0.1:11434

The Markdown report is optimized for reading: original messages and rewritten outputs wrap as normal Markdown text, and the JSON report keeps raw structured results for later comparison.

Contributing

Bug reports, focused feature requests, and pull requests are welcome. Start with CONTRIBUTING.md, and use SECURITY.md for vulnerability reports.