Overview

Flue is a groundbreaking, headless TypeScript framework for building the next generation of autonomous agents. Designed with a powerful built-in agent harness, Flue empowers developers to create sophisticated agents that operate independently, similar to the functionality of tools like Claude Code, but without any graphical interfaces or baked-in assumptions about human operators. It offers a runtime-agnostic approach, akin to modern web frameworks like Astro or Next.js, enabling agents to be built once and deployed across diverse environments, including Node.js, Cloudflare Workers, GitHub Actions, and GitLab CI/CD.

Key Features

• Agent Harness Framework: Provides a robust, programmable environment for agent development, abstracting away complex orchestration logic.
• Markdown-Driven Logic: Significant agent "intelligence" is defined using Markdown for skills, context, and configuration, allowing for rapid iteration and intuitive development.
• Runtime Agnostic: Build agents that deploy anywhere, supporting a wide range of execution environments from serverless functions to CI/CD pipelines.
• Programmable and Headless: Designed for full programmatic control with no inherent UI, making it ideal for backend and automated agent systems.
• Virtual Sandboxes: Offers a fast, cost-effective, and scalable virtual sandbox powered by just-bash by default, with options for more robust sandboxes like local() and Daytona for complex scenarios.
• Session Management: Robust session persistence, especially on platforms like Cloudflare, ensuring conversations and state are maintained across interactions.
• Task Delegation: Enables agents to spawn and manage focused, one-shot child agents (tasks) for parallel research and exploration.
• Extensible Connectors: Easily integrate third-party services, including sandbox providers and tools, through a flexible connector system.

Typical Use Cases

• Automated Task Execution: Develop agents that can autonomously complete complex tasks in CI/CD pipelines, such as issue triaging, automated testing, or code deployments.
• Intelligent Support Bots: Create highly responsive and stateful support agents that can access and search knowledge bases to provide accurate customer assistance, even across long periods.
• Headless AI Services: Build programmable AI services that can be integrated into any application or workflow, leveraging LLMs for sophisticated problem-solving without user interface dependencies.
• Coding and Development Agents: Construct agents that can interact with codebases, clone repositories, install dependencies, and even draft implementations, powered by environments like Daytona.
• Message-Driven Integrations: Design agents that communicate via HTTP or WebSocket, enabling seamless integration with external systems and real-time asynchronous input handling.