openzoo 0.48.38 → 0.48.40

package/lib/launch.js

@@ -143,6 +143,20 @@ export async function launchClaude(argv) {
   delete env.ANTHROPIC_API_KEY;
   env.ANTHROPIC_BASE_URL = base;
   env.ANTHROPIC_AUTH_TOKEN = process.env.ANTHROPIC_AUTH_TOKEN || 'sk-openzoo';
+  // RAISE THE CEILING, because the proxy already removed it.
+  //
+  // Claude Code counts the WHOLE local transcript against the model's window and
+  // hard-stops with "Context limit reached - /compact or /clear to continue". It
+  // cannot know the proxy binds the old prefix into leCore and forwards only the
+  // system block plus a bounded tail - MEASURED live, 8 of 166 turns actually
+  // sent. Disabling auto-compact WITHOUT raising this made it worse: it used to
+  // compact and carry on, and instead it just stopped.
+  //
+  // Safe only because spillTranscript is real: OPENZOO_TAIL_MAX_CHARS bounds what
+  // leaves this machine however long the conversation gets.
+  if (process.env.OPENZOO_KEEP_COMPACT !== '1' && !process.env.CLAUDE_CODE_MAX_CONTEXT_TOKENS) {
+    env.CLAUDE_CODE_MAX_CONTEXT_TOKENS = process.env.OPENZOO_CLAUDE_CONTEXT_TOKENS || '1000000';
+  }
 
   if (terminal) {
     const cli = resolveClaudeCli();
@@ -337,6 +351,20 @@ export async function launchHarness(cmd, args) {
   delete env.ANTHROPIC_API_KEY;   // conflicts with the gateway auth-token path
   env.ANTHROPIC_BASE_URL = base;
   env.ANTHROPIC_AUTH_TOKEN = process.env.ANTHROPIC_AUTH_TOKEN || 'sk-openzoo';
+  // RAISE THE CEILING, because the proxy already removed it.
+  //
+  // Claude Code counts the WHOLE local transcript against the model's window and
+  // hard-stops with "Context limit reached - /compact or /clear to continue". It
+  // cannot know the proxy binds the old prefix into leCore and forwards only the
+  // system block plus a bounded tail - MEASURED live, 8 of 166 turns actually
+  // sent. Disabling auto-compact WITHOUT raising this made it worse: it used to
+  // compact and carry on, and instead it just stopped.
+  //
+  // Safe only because spillTranscript is real: OPENZOO_TAIL_MAX_CHARS bounds what
+  // leaves this machine however long the conversation gets.
+  if (process.env.OPENZOO_KEEP_COMPACT !== '1' && !process.env.CLAUDE_CODE_MAX_CONTEXT_TOKENS) {
+    env.CLAUDE_CODE_MAX_CONTEXT_TOKENS = process.env.OPENZOO_CLAUDE_CONTEXT_TOKENS || '1000000';
+  }
   console.error(`openzoo: launching \`${cmd}\` on the zoo (ANTHROPIC_BASE_URL=${base}) — every turn pays x402`);
   const child = spawn(cmd, args, { stdio: 'inherit', env });
   child.on('exit', (code) => process.exit(code ?? 0));

package/lib/proxy.js

@@ -1106,6 +1106,12 @@ export async function startProxy({ silent = false, requireToken = null, sessionM
           sessionCogs += typeof x.cogsUsd === 'number' ? x.cogsUsd : x.billedUsd / MARKUP;
           sessionDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
           if (didSpill) {
+            // THE number that settles why a spilled call did or did not save:
+            // the gateway only prices a counterfactual when corpusTokens >
+            // promptTokens, so a tail that rivals the corpus silently falls
+            // back to markup and direct collapses onto billed.
+            const lc = x.lecore || {};
+            log(`spill priced: ${x.pricing} · corpus ${lc.corpusTokens ?? '?'} tok vs sent ${lc.tokensBefore ?? '?'} -> ${lc.tokensAfter ?? '?'} · billed ${(x.billedUsd ?? 0).toFixed(5)} direct ${(x.directUsd ?? 0).toFixed(5)}`);
             spillSpend += x.billedUsd;
             spillDirect += typeof x.directUsd === 'number' ? x.directUsd : x.billedUsd;
           }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openzoo",
-  "version": "0.48.38",
+  "version": "0.48.40",
   "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",