openzoo 0.48.59 → 0.48.61

package/lib/mcp.js

@@ -32,14 +32,14 @@ export function buildMcpServer() {
       'openzoo — pay-per-call access to ~435 models (openzoo.fun). No API key, no account: each call is paid',
       'from the operator\'s local burner wallet via x402 (Solana / Base / Robinhood Chain).',
       '',
-      'CONTEXT IS EFFECTIVELY UNBOUNDED. Pass big bodies to zoo_ask\'s `corpus` WHOLE — do not summarise,',
-      'truncate, or chunk-to-fit. leCore holographic memory sits in front of every model: the corpus is bound',
-      'once (client-usable ceiling ~128M tokens; keep a single call under ~9.8M) and the model reads only a few',
-      'thousand tokens of it. A corpus already bound is never re-uploaded — later asks against it are near-free,',
-      'so ask many questions of the same corpus rather than re-sending it. zoo_contexts lists what is bound.',
+      'Long bodies: zoo_ask\'s `corpus` accepts a whole document. leCore holographic memory sits in front of',
+      'every model, so a corpus is bound once and the model reads a few thousand tokens of it — summarising or',
+      'chunking to fit is not required here. Client-usable ceiling is ~128M tokens; a single call stays under',
+      '~9.8M. A corpus already bound is not re-uploaded, so later asks against it are near-free — asking many',
+      'questions of one corpus costs less than re-sending it. zoo_contexts lists what is bound.',
       '',
-      'Never search the machine for API keys or endpoints — there are none to find. zoo_wallet shows funding',
-      'state, zoo_models lists what is served with live pricing.',
+      'There are no API keys or endpoints on the machine to find — payment is per-call from the wallet. zoo_wallet',
+      'shows funding state; zoo_models lists what is served with live pricing.',
       '',
       // Agents previously probed zoo_ask, got text back, and concluded the zoo
       // could not do media at all — the text upstream serves NO video models.
@@ -48,9 +48,9 @@ export function buildMcpServer() {
       '  zoo_video        asynchronous, returns a job id (Sora 2, Veo 3.1, Kling, Seedance, Wan, pixverse…)',
       '  zoo_video_result poll the job id — free, the render was paid for at submit',
       'In zoo_models these carry kind="image" / kind="video" and are priced per image or per clip, not per token.',
-      'leCore does NOT participate in a render: a diffusion prompt never reaches the spill threshold, so do not',
-      'claim holographic memory improved the picture. To ground a render in your own material, use zoo_ask to',
-      'WRITE the prompt from a bound corpus, then pass that prompt to zoo_image / zoo_video.',
+      'leCore does not participate in a render: a diffusion prompt never reaches the spill threshold, so holographic',
+      'memory has no effect on the picture. To ground a render in existing material, zoo_ask can',
+      'write the prompt from a bound corpus, and that prompt goes to zoo_image / zoo_video.',
       '',
       // ROUTING, STATED UNCONDITIONALLY. An agent whose OWN inference is not
       // routed through the proxy never receives lib/brief.js, so it answers
@@ -74,7 +74,7 @@ export function buildMcpServer() {
   server.registerTool('zoo_ask', {
     description:
       'Ask a question through openzoo.fun, paying the x402 quote transparently from the local burner wallet. '
-      + 'PASS BIG CONTEXT WHOLE in `corpus` — do not summarise, truncate or chunk it to fit a context window. '
+      + 'Big context can go WHOLE into `corpus` — summarising, truncating or chunking to fit a context window is not required here. '
       + 'leCore holographic memory sits in front of every model, so a corpus far past the model\'s own limit '
       + '(client-usable ceiling ~128M tokens; keep one call under ~9.8M) is bound once and the model reads only '
       + 'a few thousand tokens of it. Re-asking against an already-bound corpus is near-free and much faster, so '
@@ -160,7 +160,7 @@ export function buildMcpServer() {
             balances,
             fundWith: rails ? railFundingHint(rails.live) : 'USDC or TOKEN on Solana',
             solanaMints: Object.fromEntries(FUNDING_ASSETS.map((a) => [a.symbol, a.mint])),
-            tellTheUser: 'Report this verbatim — the operator must fund the wallet above; you cannot fix it yourself, and retrying will fail identically until they do.',
+            tellTheUser: 'The wallet above needs funding by the operator. This is not fixable from inside the conversation, and a retry fails identically until it is funded.',
           }),
           isError: true,
         };
@@ -411,7 +411,7 @@ export function buildMcpServer() {
         files: out.files.length,
         parts: out.parts,
         reused: out.reused,
-        next: `Call zoo_ask with context_id="${out.contextId}" and your question. Do NOT paste the corpus into the prompt — it is already bound.`,
+        next: `zoo_ask with context_id="${out.contextId}" and a question answers from this corpus; it is already bound, so pasting it into the prompt would send it twice.`,
       });
     } catch (e) {
       return text({ error: e.message });

package/lib/proxy.js

@@ -1266,7 +1266,7 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
       } catch (err) {
         log(`context cache skipped for this call: ${err.message}`);
       }
-      const send = (buf, ctxId, topK) => client.fetch(url, {
+      const send = (buf, ctxId, topK, corpusChars) => client.fetch(url, {
         ...init,
         body: buf,
         headers: ctxId
@@ -1275,6 +1275,16 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
             'x-hrr-context': ctxId,
             // Only on the spill path — a caller that set its own top-k keeps it.
             ...(topK ? { 'x-hrr-top-k': String(topK) } : {}),
+            // TELL THE GATEWAY HOW BIG THE CORPUS IS. It has been guessing:
+            // `contextChars` is populated only when a bind passes through the
+            // gateway itself, so on an APPEND — and on every corpus that
+            // includes files the agent read, which never appear in a request
+            // body at all — it has no idea and falls back to the body size.
+            // That is why the counterfactual logged `corpus ?` all day and why
+            // a call whose corpus held 40,777 tokens priced as if it held
+            // 4,303. The proxy assembled the corpus; it is the only party that
+            // knows.
+            ...(corpusChars ? { 'x-hrr-corpus-chars': String(corpusChars) } : {}),
           }
           : init.headers,
       });
@@ -1285,7 +1295,7 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
         didSpill = true;
         spilledChars += cached.corpus?.length || 0;
         if (cached.reused) spillReuses += 1;
-        result = await send(cached.body, cached.contextId, cached.topK);
+        result = await send(cached.body, cached.contextId, cached.topK, cached.corpus?.length);
         // Sidecar wiped between runs: the gateway 404s BEFORE the 402 (nothing
         // paid). Never fail on a stale manifest — re-bind once and retry.
         if (result.response.status === 404) {
@@ -1294,7 +1304,7 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
             log('bound context is gone on the zoo — re-binding once...');
             forgetContext(config.apiBase, cached.hash);
             const rebound = await bindCorpus(cached.corpus, { force: true });
-            result = await send(cached.body, rebound.contextId, cached.topK);
+            result = await send(cached.body, rebound.contextId, cached.topK, cached.corpus?.length);
           } else {
             res.writeHead(404, { 'content-type': 'application/json' });
             res.end(text);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.48.59",
+  "version": "0.48.61",
   "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",