Flue is a sandbox agent framework from the Astro team that provides a complete TypeScript harness for building autonomous agents. It gives models the context and environment needed for autonomous work — sessions, tools, skills, instructions, filesystem access, and a secure sandbox to run in. Agents built with Flue maintain context across conversations, run structured automations with code guiding agent reasoning, and operate in a secure environment where they can use tools, modify files, and complete real work. Flue is runtime-agnostic: write once, build, and deploy agents anywhere including Node.js, Cloudflare Workers, GitHub Actions, and GitLab CI/CD.
Flue is the Astro team’s sandbox agent framework — a runtime-agnostic harness for building agents that can safely take action, maintain continuity, and connect to systems where work already happens. Unlike point solutions that only handle chat or code generation, Flue provides the full agent lifecycle: sessions, tools, skills, filesystem access, and a secure sandbox. You write your agent once and deploy it to Node.js, Cloudflare Workers, GitHub Actions, or GitLab CI/CD.
Flue’s architecture separates the agent runtime from the deployment target, similar to how Astro and Next.js separate rendering from hosting. The framework handles session persistence, tool orchestration, and sandboxed execution out of the box. Agents can modify files, run code, and interact with external systems — all within a controlled environment that prevents unintended side effects. The TypeScript-first design means full type safety and IDE support for agent development.
Development teams use Flue for building coding agents with persistent memory, automating CI/CD workflows with agent-driven decision-making, creating deployment pipelines that adapt based on test results and code analysis, and building internal tools that combine agent reasoning with secure system access. The Cloudflare Workers integration makes it particularly suited for serverless agent deployments at scale.
Builder.io's open-source framework for building agent-native applications — shared actions, SQL-backed state, identity, tools, skills, jobs, observability, and UI surfaces that all work together.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.
Open-source AI pair programming tool that works in your terminal to edit code across your entire repository.