Message middleware

FluxyChat does not expose a client.use() message pipeline in the browser SDK. Middleware patterns split between Worker-side processing and LLM middleware for agents.

Worker-side

Worker-side validation, templates, moderation, and rate limits run before Durable Object broadcast. Declarative publish middleware lives in fluxy.config (onPublish allow/block/mask) via @fluxy-chat/config, bundled as apps/worker/fluxy.config.js. That is the FluxyChat equivalent of a one-file server rules deploy. You still ship it with the Worker, not a second CLI.

LLM middleware (agents)

The SDK exports LLM middleware helpers for composing provider calls on the server:

import { composeMiddlewares } from "@fluxy-chat/sdk";
// Used in custom agent/tool servers, not in browser clients

Agent tool execution flows through toolExecuteUrl on the agent profile.

Client patterns

For client-side filtering (analytics, UI transforms), subscribe to useChat messages:

const { messages } = useChat({ roomId, client });

const visible = messages.filter((m) => !m.deletedAt);

For send-time validation, validate in your UI before calling sendMessage.

Templates

Server-rendered template messages bypass free-text middleware:

await client.createMessage(roomId, "", null, undefined, clientMessageId, {
  templateId: "tpl_welcome",
  templateVars: { name: "Jane" },
});

See Message templates and Moderation.