npm - openzoo - Versions diffs - 0.43.4 → 0.43.6 - Mend

openzoo 0.43.4 → 0.43.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (3) hide show

package/lib/grokui.mjs CHANGED Viewed

@@ -115,7 +115,13 @@ Via RUN you can also make YOUR OWN paid openzoo calls — POST to
 http://localhost:8402/v1/chat/completions (or /v1/hrr/bind) with curl/python/etc. Auth is
 "Authorization: Bearer sk-openzoo" — any string works, x402 pays per call, not the key. Do
 NOT tell the user you "can't fire the paid calls" or need "their client's bearer key" —
-that's wrong, you can make these calls yourself via RUN.
+that's wrong, you can make these calls yourself via RUN. When you do, set max_tokens
+generously (1000+, not 50) — a reasoning model can burn its ENTIRE budget on internal
+reasoning before writing any visible answer, especially against a large bound corpus, and
+comes back with content:null and finish_reason:"length" (confirmed live) if you starve it.
+/v1/hrr/bind also caps around ~8MB per request after JSON-escaping — chunk large corpora
+(e.g. 512KB raw per request) and pass the PREVIOUS chunk's context_id on each next request
+to append to the same bound context, rather than one giant request that silently fails partway.
 For normal questions just answer directly — do not use any of these unless the request
 actually calls for delegation or file work.`;
@@ -230,17 +236,32 @@ function newGroupThread(names) {
 // turn's brain()/brainStream() call picks up t.contextId once it lands, via
 // the X-HRR-Context header, so retrieval is real and automatic, not a prompt
 // claim about a mechanism that doesn't exist.
+// Chunked, not one shot: a single request over ~8MB (post JSON-escaping)
+// gets rejected, so a large/growing thread's bind would silently fail past
+// whatever point it crossed that line — confirmed live by a bot's own RUN
+// diagnostic ("Failed at chunk 2 — JSON escaping pushed a 3MB chunk over the
+// ~8MB request limit"). 512KB raw per request leaves wide margin. Each
+// chunk after the first carries the PREVIOUS chunk's context_id so the
+// sidecar appends to the same bound context instead of starting fresh.
+const BIND_CHUNK_BYTES = 512 * 1024;
 async function bindThread(t) {
   const corpus = t.history.map((h) => (h.who === 'user' ? 'you' : (h.name || t.name)) + ': ' + h.text).join('\n');
   if (!corpus.trim()) return;
   try {
-    const r = await fetch(`${PROXY}/hrr/bind`, {
-      method: 'POST',
-      headers: { 'content-type': 'application/json' },
-      body: JSON.stringify({ corpus }),
-    });
-    const j = await r.json().catch(() => ({}));
-    if (j?.context_id) { t.contextId = j.context_id; saveThreads(); }
+    let ctx;
+    for (let i = 0; i < corpus.length; i += BIND_CHUNK_BYTES) {
+      const part = corpus.slice(i, i + BIND_CHUNK_BYTES);
+      const body = ctx ? { corpus: part, context_id: ctx } : { corpus: part };
+      const r = await fetch(`${PROXY}/hrr/bind`, {
+        method: 'POST',
+        headers: { 'content-type': 'application/json' },
+        body: JSON.stringify(body),
+      });
+      const j = await r.json().catch(() => ({}));
+      if (j?.context_id) ctx = j.context_id;
+      else break; // this chunk failed — stop, keep whatever bound so far rather than lose it all
+    }
+    if (ctx) { t.contextId = ctx; saveThreads(); }
   } catch { /* leCore sidecar unreachable — thread still works, just not bound this round */ }
 }
@@ -639,6 +660,7 @@ const APP_HTML = `<!doctype html>
       <div class="hrow"><span>our cost (cogs)</span><span id="hYouCogs">—</span></div>
       <div class="hrow"><span>margin</span><span id="hYouMargin" class="hlime">—</span></div>
       <div class="hrow"><span>direct would be</span><span id="hYouDirect" class="hember">—</span></div>
+      <div class="hrow"><span>saved vs. naked calls</span><span id="hYouSaved" class="hlime">—</span></div>
       <div class="hfoot" id="hFoot">loading…</div>
     </div>
     <div id="log"></div>
@@ -1034,6 +1056,17 @@ const APP_HTML = `<!doctype html>
       document.getElementById('hYouCogs').textContent = usd(cogs);
       document.getElementById('hYouMargin').textContent = margin;
       document.getElementById('hYouDirect').textContent = usd(direct);
+      const savedEl = document.getElementById('hYouSaved');
+      if (spent > 0) {
+        const mult = direct / spent;
+        // honest either way: >=1x is a real saving vs a naked direct call,
+        // <1x means you're currently paying MORE than direct would cost —
+        // don't dress that up as green when it isn't one
+        savedEl.textContent = mult.toFixed(2) + 'x';
+        savedEl.className = mult >= 1 ? 'hlime' : 'hember';
+      } else {
+        savedEl.textContent = '—';
+      }
       document.getElementById('hFoot').textContent = (you.paidCalls || 0) + ' paid calls this session';
     } catch (e) {
       document.getElementById('hFoot').textContent = 'error: ' + e.message;

package/lib/podagent.mjs CHANGED Viewed

@@ -192,7 +192,7 @@ function hasImages(messages) {
 // grokui.mjs then renders as a generic "(no response)" — indistinguishable
 // from a model that genuinely had nothing to say. Callers should know WHY.
 function httpErrorNote(status) {
-  if (status === 402) return '(payment failed — the wallet\'s x402 retry gave up after HTTP 402. Try again; if it keeps happening, check the wallet balance/RPC.)';
+  if (status === 402) return '(payment failed — the wallet\'s x402 retry gave up after HTTP 402, likely out of funds. Open http://localhost:8402 for this wallet\'s funding addresses (Solana USDC/TOKEN, Base USDC, Robinhood Chain) and current balance, or run `npx openzoo` in a terminal to see the same info. Try again once it\'s funded.)';
   if (status === 429) return '(rate limited — HTTP 429, try again in a moment)';
   if (status >= 500) return `(upstream error — HTTP ${status}, try again)`;
   return status ? `(request failed — HTTP ${status})` : '';

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.43.4",
+  "version": "0.43.6",
   "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",