Paraprompt is an open prompt registry — a community-curated library of AI prompts, prompt techniques and prompt templates for every major model (ChatGPT, Claude, Gemini, Deepseek, Llama, Mistral, and image/video/audio models). Browse, fork, share, and run prompts directly from the hub or via our Chrome extension.

What is a prompt registry?

A prompt registry is to AI prompts what npm is to JavaScript packages or GitHub is to source code: a central, searchable, versioned home where prompts are discovered, shared, attributed and improved by a community. Paraprompt is the open prompt registry for the LLM era.

Which AI models does Paraprompt support?

Paraprompt is model-agnostic. Prompts are tagged with the models they were designed for — including ChatGPT/GPT-4o, Claude (Opus, Sonnet, Haiku), Gemini Pro, Deepseek, Llama 3, Mistral/Mixtral, Qwen, Grok, Perplexity, and image/video models like Midjourney, Stable Diffusion, DALL·E, Sora and Runway.

What prompt-engineering techniques are covered?

Every prompt is tagged with one of five canonical techniques: Basic (zero-shot), Few-Shot, Chain-of-Thought (CoT), Tree-of-Thought (ToT), and Retrieval-Augmented Generation (RAG). Each technique has a dedicated guide on Paraprompt explaining when and how to use it.

Yes. Browsing, searching, downloading and sharing prompts is free. Account creation is free. The browser extension is free.

How is Paraprompt different from PromptBase or FlowGPT?

Paraprompt is open and community-first: prompts are tagged by technique, ranked by community signals (likes, power, funny), grouped into collections, and accessible from anywhere via our extension. There is no marketplace gate — every prompt is free to use and remix.

Can I save my own prompts to Paraprompt?

Yes — sign in, create prompts in the hub, organise them into collections, and access them from any browser tab via the Paraprompt Chrome extension. You decide whether each prompt is private or public.

What is the Paraprompt Chrome extension?

The Paraprompt Chrome extension is a side panel that surfaces your prompts on every page — including ChatGPT, Claude.ai, Gemini and any text input. Trigger any prompt by shortcut, fill in template variables on the fly, and paste straight into the model's UI.

What's the difference between zero-shot, few-shot, and chain-of-thought prompting?

Zero-shot gives the model a direct instruction with no examples. Few-shot provides 2–8 worked examples to anchor format and tone. Chain-of-Thought adds an explicit "think step by step" trigger so the model reasons before answering — best for math, logic and multi-step problems.

When should I use RAG instead of a longer prompt?

Use RAG (Retrieval-Augmented Generation) any time the answer depends on data the model wasn't trained on — your docs, tickets, code, or anything that changes after the model's cutoff. Pair RAG with explicit grounding instructions and citations for the highest factual accuracy.

Does Paraprompt support image, video and audio prompts?

Yes. Prompts on the hub support text, image, audio and video previews — so a Midjourney prompt shows the rendered image, a Suno prompt plays the generated audio, and a Sora prompt embeds the video output.

Can I use Paraprompt prompts commercially?

Most prompts on Paraprompt are shared by their authors for free reuse. Each prompt page shows attribution; check the author's note for any specific terms.

Prompt technique

Few-Shot Prompting

Show the model a handful of input/output pairs and it will mimic the pattern.

What it is

Few-shot prompting (also called in-context learning) gives the model 2–8 demonstrations of the task before asking it to perform on a new input. The model picks up format, tone, vocabulary and edge cases from the examples without any fine-tuning. This is the highest-leverage technique for structured output, custom formats, and domain-specific tone.

When to use it

✓You need a specific output format (JSON shape, table, custom DSL)
✓Tone, voice or vocabulary that the model wouldn't pick by default
✓Classification or extraction tasks with subtle label boundaries

Example

Classify the sentiment of each review as POSITIVE, NEGATIVE, or MIXED.

Review: "The screen is gorgeous but the battery dies in 3 hours."
Label: MIXED

Review: "Best laptop I've ever owned, no complaints."
Label: POSITIVE

Review: "{user_review}"
Label:

Why it works: Two demonstrations anchor the label set and the boundary case (MIXED) before the model is asked to classify a new review.

Pitfalls

!Biased examples bias the output — pick demonstrations that span the full label distribution.
!Token cost grows with the prompt; cache the few-shot block when possible.

Pairs well with

Chain-of-Thought →RAG →