prompts.chat — The World's Largest Open-Source Prompt Library(f/prompts.chat)

One-line summary

prompts.chat — formerly the universally known Awesome ChatGPT Prompts — is no longer a CSV file. It's the world's largest open-source prompt engineering platform, packing thousands of community-validated prompts with semantic search, structured prompt types, MCP integration, and one-command self-hosting. Think of it as an "operating system" for prompts that works with every AI assistant.

Background and origins

The origin story has a bit of legend to it. It started in December 2022 as a single CSV file by Turkish developer Fatih Kadir Akn (@f), called Awesome ChatGPT Prompts, with one simple goal: collect role-settings for ChatGPT that people had verified worked well. It caught the exact takeoff of generative AI, climbing past 165K stars, getting covered by Forbes, cited 40+ times in Harvard and Columbia papers, and becoming the most-liked dataset on Hugging Face.

What's interesting is that it didn't stop at being a "list." As the model landscape exploded from a single ChatGPT into ChatGPT, Claude, Gemini, Llama, and Mistral, the author rebuilt it into a full Next.js application — swapping the flat file for PostgreSQL, adding semantic search, version control, an MCP server, and a CLI tool, and renaming it prompts.chat. The pivot from a static table to a platform where "prompts are versioned and composable like code" was genuinely well executed, and it's a big reason the project stayed relevant instead of becoming a historical curiosity.

The numbers tell the scale: the repo sits at 165K+ stars, was a GitHub Staff Pick, and is the first prompt library of its kind (dating to December 2022). Greg Brockman and other AI pioneers have publicly endorsed it. That kind of momentum is rare for what began as a text file — most "awesome lists" fade within a year, but prompts.chat kept evolving because the author treated it as a product with a roadmap, not a bookmark to abandon.

Core features in detail

• Massive prompt library: 1000+ community-validated prompts spanning writing, coding, learning, business, and creative scenarios. Each copies in one click, supports inline variable substitution, and drops straight into any AI assistant.
• Semantic search: not naive keyword matching, but vector-embedding search (HuggingFace-hosted embeddings) that understands intent. Searching "write unit tests" surfaces related prompts about TDD workflows and assertion patterns, so you find what you need far faster.
• Multi-type prompts: supports TEXT, IMAGE, VIDEO, AUDIO, STRUCTURED, SKILL, and TASTE types, each with its own schema. The standout is the variable placeholder syntax ${VariableName:DefaultValue}, which parameterizes a template for reuse — a code-review template with ${ProgrammingLanguage:Python} works across any language.
• Version control and change requests: prompt edits go through a PR-like change-request flow, merged only after community review. It's essentially a quality gate for prompts, keeping junk out of the main library.
• Multilingual and social layer: the interface ships in English, Spanish, Japanese, Turkish, and Chinese, with a voting and leaderboard system (PromptMasters) that surfaces the most proven prompts. Popular categories break down roughly into writing assistants (23%), coding helpers (18%), tutoring (15%), and business analysis (12%) — a useful map of where people actually lean on prompts day to day.
• Private prompts and personalization: you can keep private prompts, subscribe to categories for a curated feed, and vote on a leaderboard (PromptMasters). Organizations can build internal libraries too.
• Interactive learning: a free interactive "Book of Prompting" with 25+ chapters covering everything from basics to chain-of-thought, few-shot, and AI agents. There's even a gamified "Prompting for Kids" adventure for ages 8-14.

Architecture and implementation

The frontend is Next.js (14/16) with React and Tailwind, fully TypeScript, backed by PostgreSQL 17 with Prisma 6 for end-to-end type safety. The cleverest piece is "dual persistence": the production PostgreSQL database and the repo's prompts.csv stay in bidirectional sync. Community submissions on the web sync back into the Git repo through the generate-contributors.sh script, preserving both the database experience and Git version control. Auth uses NextAuth (GitHub/Google/Azure AD). The integration layer also wires in the OpenAI API for prompt testing and generation, and HuggingFace for embeddings.

Full comparison with competitors

To put it bluntly: PromptBase sells prompts as products, FlowGPT leans social play, and the official libraries are just example dumps. prompts.chat's unique value is "open-source + self-hostable + drops into your workflow via MCP/CLI," which makes it the most practical choice for teams and developers.

Who it's for

• Prompt engineering beginners: reading great prompts teaches structure and technique faster than any tutorial.
• Developers and teams: through the CLI (npx prompts.chat) and MCP Server, the prompt library feeds directly into Cursor and Claude Desktop, callable mid-coding.
• Enterprises/orgs: self-host a private internal library with white-label branding, themes, and authentication — data never leaves the building.
• Not for: extremely light users who just want one universal template. The web version is enough; don't bother with self-hosting.

Quick start guide

The web version is zero-setup — just search at prompts.chat. To self-host, one command gets you going:

# Option 1: one-command scaffold (recommended)
npx prompts.chat new my-prompt-library
cd my-prompt-library

# Option 2: manual clone
git clone https://github.com/f/prompts.chat.git
cd prompts.chat
npm install && npm run setup   # wizard configures branding/theme/auth

# Use it as an MCP Server inside AI tools (remote, recommended)
# add to Claude Desktop / Cursor mcp.json:
# { "mcpServers": { "prompts.chat": { "url": "https://prompts.chat/api/mcp" } } }

# CLI usage
npx prompts.chat            # browse/search prompts from the terminal

Configuration lives in .env (DATABASE_URL, AUTH_GITHUB_SECRET, optional OPENAI_API_KEY to enable AI search), and white-label branding is set in prompts.config.ts with useCloneBranding = true.

Community and ecosystem

From a one-person CSV to 190+ contributors and 165K+ stars, featured in Forbes with 40+ academic citations — the ecosystem is solid. It even ships a Claude Code plugin (/plugin marketplace add f/prompts.chat) and official MCP docs, so the developer integration path is complete. Updates are frequent; in 2026 it's still adding AI search, the Kids game, and more.

Verdict

In my view prompts.chat is the open-source project that has thought hardest about what "prompts" actually are — turning disposable text into versioned, composable, self-hostable infrastructure, a positioning nobody has beaten yet. The downside: self-hosting depends on PostgreSQL + Node 18+, which is a real barrier for non-technical folks; and while CC0 is liberating, it means anyone can commercialize the work, so the author essentially runs on sponsorships rather than a paid tier. For most individual users the free web version already covers everything, so the self-host route really pays off only when privacy or branding matters. If you're serious about prompt engineering or want a private team library, go for it without hesitation. If you just want to copy the occasional template, the web version is plenty.