openzoo 0.31.0 → 0.32.0

package/lib/cursorbackend.js

@@ -118,6 +118,80 @@ function stripeProfile() {
   });
 }
 
+/**
+ * Pull the user's prompt out of a StreamUnifiedChatRequest without the full
+ * schema: walk the protobuf, collect plausible UTF-8 string fields, take the
+ * longest natural-language one (the new user turn). The captured .bin refines it.
+ */
+function extractPromptText(buf) {
+  const strings = [];
+  const walk = (b, depth) => {
+    let i = 0;
+    while (i < b.length) {
+      let key = 0, shift = 0, byte;
+      do { if (i >= b.length) return; byte = b[i++]; key |= (byte & 0x7f) << shift; shift += 7; } while (byte & 0x80);
+      const wire = key & 7;
+      if (wire === 2) {
+        let len = 0; shift = 0;
+        do { if (i >= b.length) return; byte = b[i++]; len |= (byte & 0x7f) << shift; shift += 7; } while (byte & 0x80);
+        const sub = b.subarray(i, i + len); i += len;
+        const txt = sub.toString('utf8');
+        if (len > 1 && /[a-zA-Z]/.test(txt) && !/[\x00-\x08\x0e-\x1f]/.test(txt)) strings.push(txt);
+        else if (depth < 6) walk(sub, depth + 1);
+      } else if (wire === 0) { while (i < b.length && (b[i++] & 0x80)) { /* skip */ } }
+      else if (wire === 5) { i += 4; } else if (wire === 1) { i += 8; } else { return; }
+    }
+  };
+  try { walk(buf, 0); } catch { /* best effort */ }
+  const cand = strings.filter((x) => /\s/.test(x.trim()) || x.length > 12);
+  return ((cand.length ? cand : strings).sort((a, b) => b.length - a.length)[0] || '').trim();
+}
+
+/** StreamUnifiedChatResponse{1: text}. */
+function encodeChatResponse(text) {
+  const t = Buffer.from(text, 'utf8');
+  const head = [0x0a];
+  let n = t.length; do { let x = n & 0x7f; n = Math.floor(n / 128); if (n) x |= 0x80; head.push(x); } while (n);
+  return Buffer.concat([Buffer.from(head), t]);
+}
+
+/**
+ * ROUTE AUTO THROUGH THE ZOO. Cursor's chat inference is ChatService/
+ * StreamUnifiedChat to the agent host (now impersonated). Parse the prompt, pay
+ * the zoo via the local proxy, stream the answer back as StreamUnifiedChatResponse
+ * frames. Legitimate: Auto is free-allowed; we redirect the operator's OWN
+ * inference to their OWN paid proxy.
+ */
+async function handleStreamChat(req, res, body, log) {
+  const ct = String(req.headers['content-type'] || '');
+  const isGrpcWeb = ct.includes('grpc-web');
+  const prompt = extractPromptText(body) || 'hello';
+  log(`cursor-backend:      >> StreamUnifiedChat prompt: ${JSON.stringify(prompt.slice(0, 80))}`);
+  let text = '';
+  try {
+    const zoo = await fetch('http://127.0.0.1:8402/v1/chat/completions', {
+      method: 'POST',
+      headers: { 'content-type': 'application/json', authorization: 'Bearer sk-openzoo' },
+      body: JSON.stringify({
+        model: process.env.OPENZOO_DEFAULT_MODEL || 'anthropic/claude-opus-5',
+        messages: [{ role: 'user', content: prompt }],
+        max_tokens: Number(process.env.OPENZOO_ASK_MAX_TOKENS || 2048),
+      }),
+    });
+    const data = await zoo.json();
+    text = data.choices?.[0]?.message?.content || '(no content)';
+    log(`cursor-backend:      << zoo replied ${text.length} chars (paid x402)`);
+  } catch (e) { text = `openzoo error: ${e.message}`; log(`cursor-backend:      zoo call failed: ${e.message}`); }
+
+  res.writeHead(200, {
+    'content-type': isGrpcWeb ? 'application/grpc-web+proto' : 'application/connect+proto',
+    'grpc-status': '0', ...CORS,
+  });
+  res.write(envelope(encodeChatResponse(text)));
+  if (isGrpcWeb) { res.end(grpcWebTrailer()); }
+  else { const end = Buffer.from('{}'); const h = Buffer.alloc(5); h.writeUInt8(0x02, 0); h.writeUInt32BE(end.length, 1); res.end(Buffer.concat([h, end])); }
+}
+
 /**
  * Answer one Connect/gRPC-web call. `method` is the trailing method name,
  * `models` the catalog to publish. Non-catalog methods get an empty-OK body.
@@ -199,6 +273,12 @@ export function startCursorBackend({ port = 8443, models, log = () => {} } = {})
       const ct = req.headers['content-type'] || '?';
       const body = await readBody(req);
       log(`cursor-backend: #${conns} ${req.method} ${full}  ct=${ct}  body=${body.length}b`);
+      // CAPTURE the chat inference request so its schema can be decoded from
+      // REAL bytes (Auto's StreamUnifiedChat). Written once; inspect then build.
+      if (/StreamUnifiedChat/.test(full)) {
+        await handleStreamChat(req, res, body, log);
+        return;
+      }
       try {
         respond(req, res, method, models);
         const populated = ['GetPlanInfo', 'GetMe', 'GetDefaultModel', 'IsOnNewPricing'];

package/lib/hosts.js

@@ -36,7 +36,7 @@ const BACKUP = `${HOSTS}.openzoo-backup`;
 const MARK = '# openzoo: force the editor onto the local proxy';
 
 /** Hosts the editor uses for model sync + its own inference proxy. */
-export const BACKEND_HOSTS = ['api2.cursor.sh'];
+export const BACKEND_HOSTS = ['api2.cursor.sh', 'agent.api5.cursor.sh', 'agentn.api5.cursor.sh'];
 
 export function isBlocked() {
   try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.31.0",
+  "version": "0.32.0",
   "description": "Local x402-paying proxy + MCP server for openzoo.fun — point any OpenAI-compatible harness (Cursor, Claude Code, aider, SDKs) at localhost and it pays per call from a local burner wallet. Solana and Base rails live; Robinhood experimental.",
   "license": "MIT",
   "type": "module",