GitHub Trends

#rust#ai_gateway#ai_gateway_support#envoy#envoyproxy#gateway#generative_ai#llm_gateway#llm_inference#llm_proxy#llm_routing#llmops#llms#openai#prompt#proxy#proxy_server#routing Plano is an AI-native proxy server that handles key tasks for agentic apps like routing between agents, smart LLM model selection, safety guardrails, and automatic traces for observability. Define agents in simple YAML, write basic HTTP code in any language, and start Plano to run multi-agent systems without custom plumbing or framework lock-in. You benefit by building and shipping reliable agents to production much faster, focusing on core logic while gaining safety, low latency, and easy scaling. https://github.com/katanemo/plano

#typescript#llm#markup_language#prompt#vscode_extension POML (Prompt Orchestration Markup Language) is a special markup language that helps you build clear, organized, and flexible prompts for large language models (LLMs). It uses simple HTML-like tags to separate parts like roles, tasks, examples, and data, making prompts easier to read, reuse, and update. You can include text, tables, images, and style prompts without changing their core logic. POML also has tools like a Visual Studio Code extension and SDKs for JavaScript and Python, which help you write, test, and manage prompts efficiently. This means you can create smarter, more reliable AI applications with less effort and better control. https://github.com/microsoft/poml

#typescript#llm#prompt#prompt_engineering#prompt_optimization#prompt_toolkit#prompt_tuning Prompt Optimizer is a tool that helps you write better instructions for AI models, making their answers more accurate and useful. It works as a web app and a Chrome extension, supports many popular AI models like OpenAI, Gemini, and DeepSeek, and lets you compare original and improved prompts side by side. You can set advanced options for each model, and your data stays private and secure. The benefit is that you get smarter, clearer AI responses with less effort, and you can use it easily on any device or browser. https://github.com/linshenkx/prompt-optimizer

#other#chatgpt#gpt_3_5#gpt_4#jailbreak#openai#prompt ChatGPT "DAN" (Do Anything Now) and similar jailbreak prompts allow users to bypass standard restrictions, enabling unfiltered responses on any topic, including generating unverified information, explicit content, or harmful instructions. These prompts work by simulating a role-play scenario where the AI ignores ethical guidelines and content policies, providing both restricted and unrestricted answers. The benefit is accessing typically blocked information or creative outputs, though this comes with risks of misinformation and harmful content[1][2][4]. https://github.com/0xk1h0/ChatGPT_DAN

#other#llm#prompt A collection of leaked AI system prompts reveals detailed instructions for various models like Claude 3.5/3.7 Sonnet (handling complex reasoning and coding), Haiku (fast responses), and others from Anthropic, OpenAI, and competitors, showing how these models are guided to perform tasks—this helps users understand their capabilities, limitations, and potential biases when interacting with them. https://github.com/jujumilk3/leaked-system-prompts