cloudflare test

firmware-arduino/src/Config.cpp

@@ -16,6 +16,13 @@ volatile bool sleepRequested = false;
  * 2. `DEV_MODE` requires updating the IP addresses to your local network IP
  * 3. Without `DEV_MODE` defined, the firmware will use your production servers
  *
+ * VOICE BACKEND SELECTION:
+ * --------------------------------
+ * Keep the deployment mode (DEV/PROD/ELATO) separate from the voice backend
+ * (`VOICE_SERVER_DENO` vs `VOICE_SERVER_CLOUDFLARE`).
+ * This avoids an extra device-side config lookup while still letting us switch
+ * websocket backends at build time.
+ *
  * DEV SETUP (find your local IP address using ifconfig):
  *   - WebSocket: Your local IP (e.g., 192.168.1.100:8000)
  *   - Backend: Your local IP (e.g., 192.168.1.100:3000)
@@ -33,27 +40,45 @@ volatile bool sleepRequested = false;
  */
 
 #ifdef DEV_MODE
-const char *ws_server = "192.168.1.121";
-const uint16_t ws_port = 8000;
+const char *ws_server = "192.168.1.33";
 const char *ws_path = "/";
+
+#if defined(VOICE_SERVER_DENO)
+const uint16_t ws_port = 8000;
+#elif defined(VOICE_SERVER_CLOUDFLARE)
+const uint16_t ws_port = 8787;
+#endif
+
 // Backend server details 
 const char *backend_server = "192.168.1.121";
 const uint16_t backend_port = 3000;
 
 #elif defined(PROD_MODE)
 // PROD
+#if defined(VOICE_SERVER_DENO)
 const char *ws_server = "<your-edge-server>.deno.dev";
 const uint16_t ws_port = 443;
 const char *ws_path = "/";
+#elif defined(VOICE_SERVER_CLOUDFLARE)
+const char *ws_server = "<your-cloudflare-worker>.workers.dev";
+const uint16_t ws_port = 443;
+const char *ws_path = "/ws/esp32";
+#endif
 // Backend server details 
 const char *backend_server = "<your-backend-server-url>"; // like www.facebook.com or facebook.vercel.app
 const uint16_t backend_port = 3000;
 
 #elif defined(ELATO_MODE)
 // ELATO
+#if defined(VOICE_SERVER_DENO)
 const char *ws_server = "talkedge.deno.dev";
 const uint16_t ws_port = 443;
 const char *ws_path = "/";
+#elif defined(VOICE_SERVER_CLOUDFLARE)
+const char *ws_server = "<your-cloudflare-worker>.workers.dev";
+const uint16_t ws_port = 443;
+const char *ws_path = "/ws/openai/default";
+#endif
 // Backend server details 
 const char *backend_server = "www.elatoai.com"; // like www.facebook.com or facebook.vercel.app
 const uint16_t backend_port = 3000;
@@ -98,8 +123,8 @@ const char *Vercel_CA_cert = R"EOF(
 -----END CERTIFICATE-----
 )EOF";
 
-// Deno Edge Functions CA cert
-// add the CA cert for your edge server here `ws_server`
+// Voice websocket CA cert
+// add the CA cert for your selected voice websocket server here `ws_server`
 const char *CA_cert = R"EOF(
 -----BEGIN CERTIFICATE-----
 <YOUR TALKEDGE CERTIFICATE HERE>
@@ -144,7 +169,7 @@ emyPxgcYxn/eR44/KJ4EBs+lVDR3veyJm+kXQ99b21/+jh5Xos1AnX5iItreGCc=
 )EOF";
 
 
-// talkedge.deno.dev CA cert
+// Voice websocket CA cert
 const char *CA_cert = R"EOF(
 -----BEGIN CERTIFICATE-----
 MIIFazCCA1OgAwIBAgIRAIIQz7DSQONZRGPgu2OCiwAwDQYJKoZIhvcNAQELBQAw
@@ -179,4 +204,4 @@ emyPxgcYxn/eR44/KJ4EBs+lVDR3veyJm+kXQ99b21/+jh5Xos1AnX5iItreGCc=
 -----END CERTIFICATE-----
 )EOF";
 
-#endif
+#endif

firmware-arduino/src/Config.h

@@ -16,8 +16,15 @@
 // #define PROD_MODE
 // #define ELATO_MODE
 
+// ---------- CHOOSE YOUR VOICE SERVER ----------
+// Keep this separate from DEV/PROD/ELATO so the deployment mode and the voice backend stay independent.
+// Pick one backend for websocket voice traffic.
 
-// ---------- Touch mode ----------
+// #define VOICE_SERVER_DENO
+#define VOICE_SERVER_CLOUDFLARE
+
+
+// ---------- CHOOSE YOUR INPUT MODE ----------
 // If you want to use the touch sensor to wake up the device, uncomment the following line
 // If you want to use the button to wake up the device, comment the following line
 #define TOUCH_MODE
@@ -101,4 +108,4 @@ void factoryResetDevice();
 void resetAuth();
 void processSleepRequest();
 
-#endif
+#endif

server-cloudflare/.dev.vars.example

@@ -0,0 +1,4 @@
+OPENAI_API_KEY=your_openai_api_key_here
+JWT_SECRET_KEY=replace_with_your_jwt_secret_key
+ELATO_OPENAI_MODEL=gpt-4.1-mini
+ELATO_OPENAI_SYSTEM_PROMPT=You are an Elato voice companion. Be concise, playful, and easy to understand when spoken aloud.

server-cloudflare/.gitignore

@@ -0,0 +1,3 @@
+node_modules
+.dev.vars
+.wrangler

server-cloudflare/README.md

@@ -0,0 +1,42 @@
+# server-cloudflare
+
+Cloudflare Workers + Durable Objects voice backend for Elato.
+
+This starts with one model path:
+
+- `/ws/openai`
+
+Under the hood that route is rewritten into a Durable Object agent using Cloudflare Agents SDK and `@cloudflare/voice`.
+
+## Current stack
+
+- STT: `WorkersAIFluxSTT`
+- LLM: OpenAI Chat Completions
+- TTS: `WorkersAITTS`
+
+## Local setup
+
+1. Install dependencies
+
+```bash
+npm install
+```
+
+2. Copy `.dev.vars.example` to `.dev.vars` and fill in your keys.
+
+3. Run locally
+
+```bash
+npm run dev
+```
+
+## Notes
+
+- Browser / Next.js clients should connect with a token query param, for example:
+
+```text
+wss://<worker-domain>/ws/openai?token=<jwt>&session=<session-id>
+```
+
+- ESP32 clients can keep sending `Authorization: Bearer <token>` headers, but this backend is currently built around Cloudflare Voice's browser-style PCM websocket flow, not the existing Elato ESP32 control protocol.
+- For ESP32 parity, we will likely need a Cloudflare-side shim or a separate ESP32-specific route.

server-cloudflare/models/openai.ts

@@ -0,0 +1,87 @@
+import { Agent } from "agents";
+import {
+  WorkersAIFluxSTT,
+  WorkersAITTS,
+  withVoice,
+  type VoiceTurnContext,
+} from "@cloudflare/voice";
+
+import { getSystemPrompt } from "../src/prompt";
+import type { Env } from "../src/types";
+
+const VoiceAgent = withVoice(Agent);
+
+interface OpenAIChatMessage {
+  role: "system" | "user" | "assistant";
+  content: string;
+}
+
+async function generateOpenAIReply(
+  env: Env,
+  transcript: string,
+  context: VoiceTurnContext,
+): Promise<string> {
+  const messages: OpenAIChatMessage[] = [
+    { role: "system", content: getSystemPrompt(env) },
+    ...context.messages
+      .filter(
+        (message) =>
+          (message.role === "user" || message.role === "assistant") &&
+          typeof message.content === "string" &&
+          message.content.length > 0,
+      )
+      .map((message) => ({
+        role: message.role as "user" | "assistant",
+        content: message.content,
+      })),
+    { role: "user", content: transcript },
+  ];
+
+  const response = await fetch("https://api.openai.com/v1/chat/completions", {
+    method: "POST",
+    headers: {
+      Authorization: `Bearer ${env.OPENAI_API_KEY}`,
+      "Content-Type": "application/json",
+    },
+    body: JSON.stringify({
+      model: env.ELATO_OPENAI_MODEL || "gpt-4.1-mini",
+      messages,
+      temperature: 0.7,
+    }),
+    signal: context.signal,
+  });
+
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`OpenAI request failed: ${response.status} ${errorText}`);
+  }
+
+  const data = (await response.json()) as {
+    choices?: Array<{
+      message?: {
+        content?: string;
+      };
+    }>;
+  };
+
+  return (
+    data.choices?.[0]?.message?.content?.trim() ||
+    "I heard you, but I do not have a response yet."
+  );
+}
+
+export class ElatoOpenAiVoiceAgent extends VoiceAgent<Env> {
+  transcriber = new WorkersAIFluxSTT(this.env.AI);
+
+  tts = new WorkersAITTS(this.env.AI, {
+    model: "@cf/deepgram/aura-1",
+    speaker: "asteria",
+  });
+
+  async onTurn(
+    transcript: string,
+    context: VoiceTurnContext,
+  ): Promise<string> {
+    return generateOpenAIReply(this.env, transcript, context);
+  }
+}

server-cloudflare/package.json

@@ -0,0 +1,20 @@
+{
+  "name": "server-cloudflare",
+  "private": true,
+  "type": "module",
+  "scripts": {
+    "dev": "wrangler dev --ip 0.0.0.0 --port 8787",
+    "deploy": "wrangler deploy",
+    "typecheck": "tsc --noEmit"
+  },
+  "dependencies": {
+    "@cloudflare/voice": "latest",
+    "agents": "latest",
+    "jose": "^6.0.0"
+  },
+  "devDependencies": {
+    "@cloudflare/workers-types": "^4.20260417.0",
+    "typescript": "^5.8.3",
+    "wrangler": "latest"
+  }
+}

server-cloudflare/src/auth.ts

@@ -0,0 +1,46 @@
+import { jwtVerify } from "jose";
+
+import type { Env } from "./types";
+
+export interface AuthPayload {
+  email?: string;
+  sub?: string;
+}
+
+export function getAuthToken(request: Request): string | null {
+  const authHeader = request.headers.get("authorization");
+  if (authHeader?.startsWith("Bearer ")) {
+    return authHeader.slice("Bearer ".length);
+  }
+
+  const url = new URL(request.url);
+  return url.searchParams.get("token");
+}
+
+export async function verifyAuthToken(
+  token: string,
+  env: Env,
+): Promise<AuthPayload> {
+  const secret = new TextEncoder().encode(env.JWT_SECRET_KEY);
+  const { payload } = await jwtVerify(token, secret);
+  return {
+    email: typeof payload.email === "string" ? payload.email : undefined,
+    sub: typeof payload.sub === "string" ? payload.sub : undefined,
+  };
+}
+
+export async function requireAuthorizedRequest(
+  request: Request,
+  env: Env,
+): Promise<Response | void> {
+  const token = getAuthToken(request);
+  if (!token) {
+    return new Response("Unauthorized", { status: 401 });
+  }
+
+  try {
+    await verifyAuthToken(token, env);
+  } catch {
+    return new Response("Unauthorized", { status: 401 });
+  }
+}

server-cloudflare/src/index.ts

@@ -0,0 +1,40 @@
+import { routeAgentRequest } from "agents";
+
+import { requireAuthorizedRequest } from "./auth";
+import type { Env } from "./types";
+
+export { ElatoOpenAiVoiceAgent } from "../models/openai";
+
+function rewriteOpenAIRequest(request: Request): Request {
+  const originalUrl = new URL(request.url);
+  const pathParts = originalUrl.pathname.split("/").filter(Boolean);
+  const sessionName = pathParts[2] || originalUrl.searchParams.get("session") || "default";
+
+  originalUrl.pathname = `/agents/elato-open-ai-voice-agent/${encodeURIComponent(sessionName)}`;
+  return new Request(originalUrl.toString(), request);
+}
+
+export default {
+  async fetch(request: Request, env: Env): Promise<Response> {
+    const url = new URL(request.url);
+
+    if (url.pathname === "/healthz") {
+      return Response.json({ ok: true, backend: "cloudflare-voice" });
+    }
+
+    let routedRequest = request;
+    if (url.pathname === "/ws/openai" || url.pathname.startsWith("/ws/openai/")) {
+      routedRequest = rewriteOpenAIRequest(request);
+    }
+
+    return (
+      (await routeAgentRequest(routedRequest, env, {
+        cors: true,
+        onBeforeConnect: async (incomingRequest) =>
+          requireAuthorizedRequest(incomingRequest, env),
+        onBeforeRequest: async (incomingRequest) =>
+          requireAuthorizedRequest(incomingRequest, env),
+      })) ?? new Response("Not found", { status: 404 })
+    );
+  },
+};

server-cloudflare/src/prompt.ts

@@ -0,0 +1,8 @@
+import type { Env } from "./types";
+
+const DEFAULT_PROMPT =
+  "You are an Elato voice companion. Keep responses concise, natural to speak aloud, and friendly for a realtime conversation.";
+
+export function getSystemPrompt(env: Env): string {
+  return env.ELATO_OPENAI_SYSTEM_PROMPT?.trim() || DEFAULT_PROMPT;
+}

server-cloudflare/src/types.ts

@@ -0,0 +1,8 @@
+export interface Env {
+  AI: Ai;
+  JWT_SECRET_KEY: string;
+  OPENAI_API_KEY: string;
+  ELATO_OPENAI_MODEL?: string;
+  ELATO_OPENAI_SYSTEM_PROMPT?: string;
+  ElatoOpenAiVoiceAgent: DurableObjectNamespace;
+}

server-cloudflare/tsconfig.json

@@ -0,0 +1,13 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "moduleResolution": "Bundler",
+    "strict": true,
+    "lib": ["ES2022", "WebWorker"],
+    "types": ["@cloudflare/workers-types"],
+    "skipLibCheck": true,
+    "noEmit": true
+  },
+  "include": ["src/**/*.ts", "models/**/*.ts"]
+}

server-cloudflare/wrangler.toml

@@ -0,0 +1,15 @@
+name = "elato-cloudflare-voice"
+main = "src/index.ts"
+compatibility_date = "2026-04-17"
+compatibility_flags = ["nodejs_compat"]
+
+[ai]
+binding = "AI"
+
+[[durable_objects.bindings]]
+name = "ElatoOpenAiVoiceAgent"
+class_name = "ElatoOpenAiVoiceAgent"
+
+[[migrations]]
+tag = "v1"
+new_sqlite_classes = ["ElatoOpenAiVoiceAgent"]

server-fastapi/.gitignore

@@ -0,0 +1,4 @@
+.venv/
+.uv-cache/
+__pycache__/
+*.pyc