Raycast AI Presets 2026: Custom Commands, Prompts & Workflows

Raycast’s built-in AI commands are useful out of the box, but the real productivity gains come when you start building your own. AI Presets — Raycast’s system for creating custom AI commands — let you define reusable prompts, pick the right model for each task, and trigger complex workflows from a single keyboard shortcut. No context-switching, no copy-pasting into ChatGPT, no friction.

If you’ve already read our Raycast AI Commands guide, you know the basics. This article goes deeper into presets specifically — what makes a great preset, how to write effective system prompts, and a library of ready-to-use presets for developers. Whether you’re on a free Raycast Pro trial or a long-time subscriber, these presets will change how you work.

What Are Raycast AI Presets?

AI Presets are saved, reusable AI commands that you configure once and use repeatedly. Each preset consists of a name, a system prompt (the instruction that shapes the AI’s behavior), a model selection, a creativity level, and an output action. Once created, a preset appears in the Raycast command palette like any other command.

Think of presets as specialized AI assistants. Instead of one general-purpose chatbot, you get a code reviewer, a test writer, a PR description generator, and a documentation drafter — each fine-tuned with the exact instructions, constraints, and tone you need. The difference between a generic “review this code” prompt and a well-crafted preset is the difference between a mediocre and exceptional result.

How to Create an AI Preset

Creating a preset takes about two minutes. Here’s the process:

1. Open Raycast and go to Settings → AI → Presets
2. Click “Create Preset”
3. Give it a descriptive name (this is what you’ll search for in the command palette)
4. Write your system prompt — the instructions that tell the AI exactly how to behave
5. Select a model (GPT-4o, Claude, Gemini, etc.)
6. Set the creativity slider — low for deterministic tasks like code generation, higher for brainstorming
7. Choose the output action — copy to clipboard, paste inline, replace selection, or display in a window

The system prompt is where the magic happens. A vague prompt produces vague results. A specific, well-constrained prompt produces output you can use immediately without editing. We’ll cover prompt engineering in detail below.

Best AI Presets for Developers

Here are the presets I use daily, complete with the actual system prompts. Copy these directly into Raycast and adjust them to match your stack.

1. Code Review

This preset analyzes selected code for bugs, performance issues, and style problems. Set the creativity to low and use Claude for best results on code analysis.

System prompt:

You are a senior software engineer conducting a code review. Analyze the provided code and identify: (1) bugs or potential runtime errors, (2) performance issues, (3) security vulnerabilities, (4) readability or maintainability concerns. For each issue, explain WHY it’s a problem and provide a specific fix. If the code is clean, say so — don’t invent problems. Be direct and concise. Format as a bulleted list.

2. Write Unit Tests

Select a function or class and generate comprehensive tests. Adjust the testing framework in the prompt to match your project.

System prompt:

You are a test engineer. Write unit tests for the provided code using Jest (TypeScript). Follow these rules: use describe/it blocks, write descriptive test names that explain the expected behavior, cover happy path + edge cases + error handling, use expect() assertions, mock external dependencies. Output ONLY the test code — no explanations. Include necessary imports.

3. PR Description Generator

Paste a git diff or list of commits and get a structured pull request description. Use GPT-4o with low creativity.

System prompt:

Generate a pull request description from the provided diff or commit list. Use this structure: ## Summary (2-3 sentences explaining the change), ## Changes (bulleted list of specific changes), ## Testing (how to verify these changes), ## Notes (anything reviewers should know). Be specific about what changed and why. Use conventional commit language. Keep it under 300 words.

4. Error Debugger

Paste an error message and stack trace to get an explanation and fix suggestions. Works best with Claude for technical accuracy.

System prompt:

You are a debugging expert. The user will provide an error message, stack trace, or unexpected behavior description. Do the following: (1) Explain what the error means in plain English, (2) Identify the most likely root cause, (3) Provide 2-3 specific solutions ranked by likelihood, (4) Include code snippets for the fix when applicable. Be direct — skip generic advice like “check your configuration.”

5. SQL to ORM Converter

Converts raw SQL queries to your ORM of choice. Customize the ORM in the prompt — Prisma, Sequelize, SQLAlchemy, or any other.

System prompt:

Convert the provided SQL query to Prisma (TypeScript). Follow these rules: use the Prisma Client API correctly, handle joins with include/select, convert WHERE clauses to Prisma’s where syntax, preserve query logic exactly. Output ONLY the Prisma code. If the query uses features Prisma doesn’t support natively, use $queryRaw and explain why.

6. Documentation Writer

Select a function, class, or API endpoint and generate clear documentation. Set creativity to medium.

System prompt:

Write developer documentation for the provided code. Include: a brief description of what it does, parameters/arguments with types and descriptions, return value, usage example, and any important notes (side effects, error conditions, performance considerations). Use JSDoc/TSDoc format for functions. Be concise — developers read documentation to find answers fast, not to read essays.

7. Commit Message Writer

Select a diff or describe your changes and get a well-formatted conventional commit message.

System prompt:

Write a git commit message for the provided changes using the Conventional Commits format. Structure: type(scope): description (50 chars max for the subject line). Add a body if the change is non-trivial, explaining WHY the change was made, not what was changed. Types: feat, fix, refactor, docs, test, chore, perf, style. Output ONLY the commit message — no commentary.

Advanced Presets: Beyond the Basics

Once you’re comfortable with simple presets, these more specialized ones can handle complex workflows.

API Response Mocker

Paste an API endpoint or TypeScript interface and get realistic mock data for testing. Useful for frontend developers who need to build against APIs that aren’t ready yet.

System prompt:

Generate realistic mock JSON data that matches the provided API response structure or TypeScript interface. Use realistic values (real-looking names, emails, UUIDs — not “test123”). Generate 3 varied items for arrays. Include edge cases like null values where types allow it. Output valid JSON only.

Regex Explainer & Builder

Either paste a regex to get an explanation, or describe what you want to match and get a regex back. Set model to Claude for precise pattern matching.

System prompt:

If the input is a regular expression, explain it step by step in plain English. Break down each part of the pattern and explain what it matches. If the input is a description of what to match, write a regex that handles it. Include the regex, an explanation, and 2-3 example matches/non-matches. Use JavaScript regex syntax.

Choosing the Right Model Per Preset

One of Raycast’s best features is per-preset model selection. Not every task needs the same AI model, and picking the right one improves both quality and speed. For a deeper look at how Raycast handles different AI models, see our Raycast ChatGPT integration guide.

• Claude — Best for code review, debugging, technical writing, and tasks requiring precision. Tends to follow instructions more literally, which is exactly what you want for structured output.
• GPT-4o — Excellent for creative tasks, brainstorming, PR descriptions, and communication. Strong at understanding nuance and generating natural-sounding text.
• Gemini — Good general-purpose option with strong reasoning capabilities. Worth testing for your specific use cases.
• Faster models — For simple transformations like formatting, casing conversions, or template expansion, use a faster model. The quality difference is negligible for simple tasks, and you get near-instant responses.

The key principle: match the model to the task complexity. Using GPT-4o to capitalize a string is overkill. Using a fast model for a nuanced code review will miss important issues.

Writing Effective System Prompts

The system prompt is the single most important part of any preset. A good prompt consistently produces output you can use. A bad prompt produces output you have to rewrite. Here are the patterns that work.

Be Explicit About Output Format

Don’t say “review this code.” Say “list issues as bullet points, each with the problem, why it matters, and a code fix.” The AI can’t read your mind about formatting preferences, so spell them out.

Set Constraints

Tell the AI what not to do. “Don’t add comments unless the logic is non-obvious.” “Keep the response under 200 words.” “Don’t suggest changes that would break backward compatibility.” Constraints prevent the AI from going off-track.

Define the Persona

Start with “You are a senior software engineer” or “You are a technical writer.” This isn’t just prompt engineering folklore — it genuinely calibrates the tone, depth, and assumptions the model makes.

Include Stack Context

If your preset is for a specific project, include the tech stack in the prompt. “The project uses React, TypeScript, Tailwind CSS, and Prisma.” This prevents the AI from guessing and producing code for the wrong framework.

Iterate Based on Failures

Run your preset 5-10 times on real input. Every time the output isn’t right, identify what went wrong and add a constraint or clarification to the prompt. The best presets are the result of dozens of small refinements.

Sharing Presets with Your Team

Individual presets are powerful. Team-wide presets are transformative. When every developer on your team uses the same code review preset, you get consistent standards. When everyone uses the same PR description preset, your pull requests become uniform and easier to review.

Raycast Teams lets you share presets across your organization. You can export a preset and share it directly, or publish it for your team to import. This is particularly valuable for:

• Onboarding — New team members get a curated set of presets that match your coding standards and workflows from day one
• Consistency — Commit messages, PR descriptions, and documentation follow the same format across the team
• Best practices — Your best developer’s code review checklist becomes everyone’s code review checklist

You can also browse community presets shared by other Raycast users. The Raycast community has published presets for everything from writing tweets to converting between data formats. Import the ones that fit your workflow and customize them.

Importing Community Presets

The Raycast community shares presets through the Raycast Store and community forums. To import a preset, navigate to Settings → AI → Presets and use the import option. You can also find shared presets on the Raycast community and import them directly.

When importing community presets, review the system prompt before using it. Make sure it matches your expectations and adjust the model, creativity, and output settings to match your preferences. A community preset is a starting point — customize it to make it yours.

Presets vs. Raycast Extensions: When to Use Which

Raycast also supports extensions that can integrate with external services and APIs. So when should you use a preset versus building or installing an extension?

• Use presets when the task is primarily about text transformation — rewriting, analyzing, generating, or converting text. Presets are faster to create and easier to modify.
• Use extensions when you need to interact with external services (GitHub, Jira, Slack), access system resources, or build complex UI. Extensions are more powerful but require JavaScript/TypeScript development.

For most AI-powered text workflows, presets are the right choice. They require zero coding, update instantly when you change the prompt, and are easy to share. Check out our Raycast Pro review for a broader look at how presets fit into the full Pro feature set.

Pricing: What You Need for AI Presets

AI Presets require Raycast Pro. The Pro plan starts at $8/month (annual billing) and includes unlimited AI commands, all AI models, cloud sync, and every other Pro feature. That’s significantly cheaper than standalone AI subscriptions — ChatGPT Plus alone costs $20/month.

If you haven’t tried Raycast Pro yet, the best current deal is 80% off with a free 14-day trial. No coupon code needed. Build a few presets during the trial and see if the workflow improvements justify the cost. In my experience, one good preset that saves 5 minutes per day pays for the subscription many times over.

Frequently Asked Questions

What are Raycast AI Presets?

Raycast AI Presets are reusable custom AI commands that you define with a specific system prompt, model selection, and creativity level. They appear in the Raycast command palette alongside built-in commands. Think of them as saved AI workflows you can trigger with a few keystrokes on any selected text or manual input.

Do I need Raycast Pro to create AI Presets?

Yes. AI Presets and all AI features in Raycast require a Raycast Pro subscription. You can try them with a free 14-day trial to see if custom AI commands fit your workflow before committing.

Can I share Raycast AI Presets with my team?

Yes. Raycast supports sharing AI Presets across teams through Raycast Teams. You can export presets and share them with colleagues, or import community presets that others have published. This ensures your entire team has access to the same custom AI workflows.

Which AI model should I use for my presets?

It depends on the task. Use Claude for code analysis, reviews, and technical accuracy. Use GPT-4o for creative writing, brainstorming, and general-purpose tasks. For quick formatting or simple transformations, faster models work fine and respond quicker. You can set a different model per preset.

How many AI Presets can I create in Raycast?

There is no hard limit on the number of AI Presets you can create in Raycast Pro. You can build as many custom commands as you need. Most power users maintain 10–20 presets covering their most common workflows, from code review to documentation to communication.