GitHub - puzzlet-ai/agentmark: Markdown for the AI era

AgentMark is a declarative, extensible, and composable framework for developing LLM applications using Markdown and JSX. It enhances readability by showing the exact inputs sent to the LLM and provides lightweight abstractions for developers. Built on TemplateDX and inspired by MDX, AgentMark supports features like Markdown, JSX components, unified model configuration, custom models, streaming, loops, conditionals, type safety, agents, and observability. It requires plugins to support model providers, with built-in plugins available for OpenAI, Anthropic, and Meta, among others. Language support includes TypeScript, with Python and Java coming soon.

AgentMark can be run using a VSCode extension, Node.js, or a Webpack loader. The framework allows developers to run `.prompt.mdx` files directly in VSCode, integrate with Node.js environments, or use a Webpack loader for workflow integration. The community is encouraged to contribute, with guidelines available, and can engage through Discord, issues, and discussions. The project is licensed under the MIT License.

Key takeaways

AgentMark is a declarative, extensible, and composable approach for developing LLM applications using Markdown and JSX.
AgentMark supports various features including Markdown, JSX components, unified model config, custom models, streaming, loops, conditionals, and more.
AgentMark does not support any model providers by default; support is added through plugins, with several built-in model plugins available.
AgentMark can be run using a VSCode extension, Node.js, or integrated with a webpack workflow.

GitHub - puzzlet-ai/agentmark: Markdown for the AI era

Key takeaways

Discussion (0)