# Introducing prompt Playground 2.0: A New Prompt Engineering IDE Streamling your prompt engineering with Playground 2.0. An integrated LLM playground for testing and comparing prompts and models. Canonical HTML page: - Category: Article - Published: Feb 6, 2025 - Author: Mahmoud Mabrouk - Tags: Product Updates Prompt engineering is the foundation of any reliable LLM application. Yet most teams struggle with a fragmented workflow - testing prompts in one place, managing versions in another, and deploying somewhere else. Today, we're introducing Playground 2.0, a complete prompt engineering IDE that brings everything together. ## Why We Built a New Kind of Prompt Playground The original [OpenAI playground](https://platform.openai.com/) changed how we interact with LLMs. But as applications grew more complex, its limitations became clear. You couldn't save test cases, compare models side-by-side, or manage prompts across environments. We watched hundreds of teams build LLM applications and saw that success depends on rapid iteration - testing prompts, comparing models, and finding what works reliably. So we rebuilt our prompt engineering workflow from the ground up. ## What Makes Playground 2.0 Different ### Multi-Message Templates That Work Modern LLM applications need more than single prompts. Now you can: - Create templates with system and user messages in one view - Add variables using \{\{variable\}\} syntax with built-in validation - See exactly what your LLM will receive, eliminating surprises in production ### Real Model Comparison Stop guessing which model works best. Our playground lets you: - Compare outputs from different models side-by-side - Test across 50+ models including [GPT-4](https://openai.com), [Claude](https://claude.ai/), [Gemini](https://gemini.google.com/), [Mistral](https://mistral.ai), and [DeepSeek](https://deepseek.com) - Adjust parameters like temperature, top-k, and presence penalty to find optimal settings - See cost and latency differences to make informed decisions ### Testing Built In We've made testing a core part of the workflow: - Load test sets directly into the playground - Save working examples as new test cases - Import production data from traces for testing - Build benchmark suites to evaluate model performance ### An Integrated Platform Everything you need in one place: - [Prompt management](https://docs.agenta.ai/prompt-management/overview) with version control and instant rollback - [Observability](/blog/open-source-llm-observability) to track every model call in production - [Evaluation framework](https://docs.agenta.ai/evaluation/overview) to measure and improve performance - Deploy to different environments with one click ## For Engineering Teams We've built tools that make production deployment easier: - [Programmatic access through our API](https://docs.agenta.ai/prompt-management/prompt-management-sdk) - Works with any framework ([LangChain](https://langchain.com), [LlamaIndex](http://llamaindex.ai/), [CrewAI](https://www.crewai.com/)) - Version control for prompts and configurations - Deploy to different environments without code changes Creating prompts now happens instantly, and your whole team can collaborate without touching code. ## Getting Started Ready to improve your prompt engineering workflow? Here's how: 1. [Create a free account](https://cloud.agenta.ai) 2. Create a new prompt 3. Load your test data 4. Start comparing models Or [book a demo](https://cal.com/mahmoud-mabrouk-ogzgey/demo) to see how it fits your use case. p.s. The [new playground](https://docs.agenta.ai/prompt-management/using-the-playground) is available now. It's open source, so you can [self-host](https://github.com/Agenta-AI/agenta) or use our [cloud version](https://cloud.agenta.ai). --- ## Machine-readable - [Sitemap](https://agenta.ai/sitemap-index.xml) - [llms.txt](https://agenta.ai/llms.txt) - [OpenAPI specification](https://agenta.ai/openapi.json) - [Documentation](https://docs.agenta.ai) Every page on https://agenta.ai serves this representation when requested with `Accept: text/markdown`.