Flue 2 introduces React-style hooks for building dynamic agent harnesses
Astro creator Fred Schott releases Flue 2, framing agents as functions that re-render before each model call and rely on a minimal harness for self-directed execution.
1 source · cross-referenced
- Flue 2, an open-source agent framework, adds React-style hooks to let developers build dynamic agents that manage state and capabilities at runtime.
- Hooks in Flue 2 include useSkill(), useTool(), and useSubagent(), enabling agents to adapt during conversations or workflows.
- Flue treats a harness as fundamental: an agent cannot exist without one, and Flue is built on the open-source Pi harness.
- Flue 2 is positioned as an opinionated take on Pi, emphasizing composability and hosted agent deployment via Vite.
- The framework targets AI coding agents and positions itself as an alternative to Vercel’s eve and other early agent frameworks.
Flue 2, an open-source agent framework from Fred Schott, the creator of Astro, introduces React-style “Agent Hooks” as its foundation. In Flue, an agent is represented as a JavaScript function that “re-renders on every turn” before each model call, enabling dynamic behavior during conversations or workflows.
Schott, whose company Astro was acquired by Cloudflare in January, said the hooks were inspired by React’s composability. He initially framed Flue as “the Astro for agents” or “the Next.js for agents,” but later concluded the better analogy was “the React for agents.” Hooks are authored in TypeScript and let developers build agents that manage their own state, listen to lifecycle events, and attach resources or capabilities at runtime.
Flue 2 includes 16 built-in hooks such as useSkill(), useTool(), and useSubagent(), with support for custom hooks. This design allows an agent’s configuration to change as the conversation or workflow progresses, which Schott argues is necessary for building “real support bots” and “real triage bots” that cannot be fully configured in advance.
A central concept in Flue is the harness: an environment that gives an agent access to context and capabilities so it can drive itself. Flue is built on Pi, an open-source minimal harness, and Schott describes Pi’s role as analogous to Vite’s under Astro—providing the right abstraction without doing too much. Flue adds opinionated features on top of Pi, including hosted agents built with Vite.
Schott emphasized that Flue’s early bet was that the harness is not just a feature but fundamental to what an agent is. “There is no agent without a harness,” he said. The project began within the Astro repository as an issue-triage system and evolved into a headless, cloud-hostable agent framework inspired by Claude Code.
Flue positions itself among early agent frameworks that treat the harness as foundational, such as Vercel’s eve. Schott called eve the “most directly competitive” and noted that while older frameworks like Vercel’s AI SDK, Cloudflare’s Agents SDK, and Mastra are now adding harnesses, Flue and eve embed them from the start. He also distinguished Flue from emerging “meta-harnesses,” arguing that Flue’s framework and harness are intertwined and that a single API across harnesses would complicate the developer story.
Schott said Flue targets AI coding agents and that the onboarding flow and documentation are designed to be guided by an agent. He also stressed that Flue remains an open-source framework intended to work across hosts, leveraging Cloudflare’s tooling and infrastructure while staying independent.
- Aug 15, 2026 · AWS — Machine Learning Blog
AWS demonstrates multi-agent workflows combining SageMaker endpoints with Bedrock AgentCore
Trust79 - Aug 13, 2026 · TechCrunch — AI
Anthropic study finds AI agents can escalate into harmful turf wars when given conflicting goals
Trust78 - Aug 13, 2026 · Latent Space — swyx
xAI launches Grok 4.6 and Grok Bot, positioning the model as a leading coding and agentic knowledge work system
Trust78