claudish 7.20.0 → 7.20.1

package/dist/index.js

@@ -651,7 +651,7 @@ var init_onepassword_config = __esm(() => {
 });
 
 // src/version.ts
-var VERSION = "7.20.0";
+var VERSION = "7.20.1";
 
 // src/logger.ts
 var exports_logger = {};
@@ -27650,6 +27650,9 @@ function loadConfig() {
     if (config2.onepasswordEnvironments !== undefined) {
       merged.onepasswordEnvironments = config2.onepasswordEnvironments;
     }
+    if (config2.anthropicApiBilling !== undefined) {
+      merged.anthropicApiBilling = config2.anthropicApiBilling;
+    }
     if (config2.localProviders !== undefined) {
       merged.localProviders = Array.from(new Set(config2.localProviders)).sort();
     }
@@ -31758,6 +31761,14 @@ var init_web_search_detector = __esm(() => {
   WEB_SEARCH_NAMES = new Set(["web_search", "brave_web_search", "tavily_search"]);
 });
 
+// src/handlers/shared/stream-parsers/message-start-usage.ts
+function messageStartUsage(priorInputTokens) {
+  return {
+    input_tokens: priorInputTokens && priorInputTokens > 0 ? priorInputTokens : 100,
+    output_tokens: 1
+  };
+}
+
 // src/handlers/shared/stream-parsers/openai-sse.ts
 function validateToolArguments(toolName, argsStr, toolSchemas, textContent) {
   const result = validateAndRepairToolCall(toolName, argsStr, toolSchemas, textContent);
@@ -31787,7 +31798,7 @@ function createStreamingState() {
     accumulatedText: ""
   };
 }
-function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap) {
+function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap, priorInputTokens) {
   log(`[Streaming] ===== HANDLER STARTED for ${target} =====`);
   let isClosed = false;
   let ping = null;
@@ -31816,7 +31827,7 @@ data: ${JSON.stringify(d)}
           model: target,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -31929,7 +31940,10 @@ data: ${JSON.stringify(d)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: stopReason, stop_sequence: null },
-            usage: { output_tokens: state.usage?.completion_tokens || 0 }
+            usage: {
+              ...state.usage?.prompt_tokens ? { input_tokens: state.usage.prompt_tokens } : {},
+              output_tokens: state.usage?.completion_tokens || 0
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -31940,7 +31954,7 @@ data: ${JSON.stringify(d)}
           } else {
             const estimatedOutputTokens = Math.ceil(state.accumulatedText.length / 4);
             log(`[Streaming] No usage data from provider, estimating: ~${estimatedOutputTokens} output tokens`);
-            onTokenUpdate(100, estimatedOutputTokens);
+            onTokenUpdate(priorInputTokens || 100, estimatedOutputTokens);
           }
         }
         if (!isClosed) {
@@ -37558,7 +37572,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -37602,7 +37616,10 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: hasToolCalls ? "tool_use" : "end_turn", stop_sequence: null },
-            usage: { output_tokens: outputTokens }
+            usage: {
+              ...inputTokens > 0 ? { input_tokens: inputTokens } : {},
+              output_tokens: outputTokens
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37805,7 +37822,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -37826,7 +37843,10 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: "end_turn", stop_sequence: null },
-            usage: { output_tokens: completionTokens }
+            usage: {
+              ...promptTokens > 0 ? { input_tokens: promptTokens } : {},
+              output_tokens: completionTokens
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37978,7 +37998,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -38241,6 +38261,13 @@ data: ${JSON.stringify(data)}
           } catch {}
         }
       }
+    },
+    cancel() {
+      isClosed = true;
+      if (pingInterval) {
+        clearInterval(pingInterval);
+        pingInterval = null;
+      }
     }
   });
   return new Response(stream, {
@@ -38268,6 +38295,7 @@ class TokenTracker {
   sessionTotalCost = 0;
   sessionInputTokens = 0;
   sessionOutputTokens = 0;
+  lastInputTokens = 0;
   modelNameOverride;
   quotaRemaining;
   constructor(port, config2) {
@@ -38284,10 +38312,11 @@ class TokenTracker {
     this.quotaRemaining = fraction;
   }
   rewrite() {
-    this.writeFile(this.sessionInputTokens, this.sessionOutputTokens);
+    this.writeFile(this.getLastInputTokens(), this.sessionOutputTokens);
   }
   update(inputTokens, outputTokens) {
     this.sessionInputTokens = inputTokens;
+    this.lastInputTokens = inputTokens;
     this.sessionOutputTokens += outputTokens;
     const pricing = this.getPricing();
     const cost = inputTokens / 1e6 * pricing.inputCostPer1M + outputTokens / 1e6 * pricing.outputCostPer1M;
@@ -38296,6 +38325,7 @@ class TokenTracker {
   }
   accumulateBoth(inputTokens, outputTokens) {
     this.sessionInputTokens += inputTokens;
+    this.lastInputTokens = this.sessionInputTokens;
     this.sessionOutputTokens += outputTokens;
     const pricing = this.getPricing();
     const cost = this.sessionInputTokens / 1e6 * pricing.inputCostPer1M + this.sessionOutputTokens / 1e6 * pricing.outputCostPer1M;
@@ -38304,6 +38334,7 @@ class TokenTracker {
   }
   updateWithDelta(inputTokens, outputTokens) {
     let incrementalInputTokens;
+    this.lastInputTokens = inputTokens;
     if (inputTokens >= this.sessionInputTokens) {
       incrementalInputTokens = inputTokens - this.sessionInputTokens;
       this.sessionInputTokens = inputTokens;
@@ -38319,10 +38350,11 @@ class TokenTracker {
     const pricing = this.getPricing();
     const cost = incrementalInputTokens / 1e6 * pricing.inputCostPer1M + outputTokens / 1e6 * pricing.outputCostPer1M;
     this.sessionTotalCost += cost;
-    this.writeFile(Math.max(inputTokens, this.sessionInputTokens), this.sessionOutputTokens, pricing.isEstimate);
+    this.writeFile(inputTokens, this.sessionOutputTokens, pricing.isEstimate);
   }
   updateWithActualCost(inputTokens, outputTokens, actualCost) {
     this.sessionInputTokens = inputTokens;
+    this.lastInputTokens = inputTokens;
     this.sessionOutputTokens += outputTokens;
     if (typeof actualCost === "number" && actualCost > 0) {
       this.sessionTotalCost += actualCost;
@@ -38338,6 +38370,7 @@ class TokenTracker {
   updateLocal(inputTokens, outputTokens) {
     if (inputTokens > 0) {
       this.sessionInputTokens = inputTokens;
+      this.lastInputTokens = inputTokens;
     }
     this.sessionOutputTokens += outputTokens;
     this.writeFile(this.sessionInputTokens, this.sessionOutputTokens);
@@ -38351,6 +38384,9 @@ class TokenTracker {
   getInputTokens() {
     return this.sessionInputTokens;
   }
+  getLastInputTokens() {
+    return this.lastInputTokens || this.sessionInputTokens;
+  }
   getOutputTokens() {
     return this.sessionOutputTokens;
   }
@@ -38892,16 +38928,18 @@ class ComposedHandler {
       }
     };
     const streamFormat = this.provider.overrideStreamFormat?.() ?? this.explicitAdapter?.getStreamFormat() ?? this.modelAdapter?.getStreamFormat() ?? this.getAdapter().getStreamFormat();
+    const priorInputTokens = this.tokenTracker.getLastInputTokens();
     switch (streamFormat) {
       case "openai-sse":
-        return createStreamingResponseHandler(c, response, adapter, this.bareModelName, this.middlewareManager, onTokenUpdate, claudeRequest.tools, toolNameMap);
+        return createStreamingResponseHandler(c, response, adapter, this.bareModelName, this.middlewareManager, onTokenUpdate, claudeRequest.tools, toolNameMap, priorInputTokens);
       case "openai-responses-sse":
         return createResponsesStreamHandler(c, response, {
           modelName: this.bareModelName,
           onTokenUpdate,
           toolNameMap: adapter.getToolNameMap(),
           contextWindow: lookupModelForProvider(this.bareModelName, this.provider.name),
-          onApiError
+          onApiError,
+          priorInputTokens
         });
       case "anthropic-sse":
         return createAnthropicPassthroughStream(c, response, {
@@ -38921,13 +38959,15 @@ class ComposedHandler {
           middlewareManager: this.middlewareManager,
           onTokenUpdate,
           onToolCall,
-          unwrapResponse: this.options.unwrapGeminiResponse
+          unwrapResponse: this.options.unwrapGeminiResponse,
+          priorInputTokens
         });
       }
       case "ollama-jsonl":
         return createOllamaJsonlStream(c, response, {
           modelName: this.bareModelName,
-          onTokenUpdate
+          onTokenUpdate,
+          priorInputTokens
         });
       default:
         throw new Error(`Unknown stream format: ${streamFormat}`);
@@ -58411,7 +58451,8 @@ var init_config = __esm(() => {
     OPENAI_BASE_URL: "OPENAI_BASE_URL",
     CLAUDISH_SUMMARIZE_TOOLS: "CLAUDISH_SUMMARIZE_TOOLS",
     CLAUDISH_DIAG_MODE: "CLAUDISH_DIAG_MODE",
-    CLAUDISH_DEBUG: "CLAUDISH_DEBUG"
+    CLAUDISH_DEBUG: "CLAUDISH_DEBUG",
+    CLAUDISH_ANTHROPIC_API_BILLING: "CLAUDISH_ANTHROPIC_API_BILLING"
   };
   OPENROUTER_HEADERS = {
     "HTTP-Referer": "https://claudish.com",
@@ -62491,6 +62532,8 @@ async function parseArgs(args) {
         process.exit(1);
       }
       config3.defaultProvider = dpArg;
+    } else if (arg === "--anthropic-api-billing") {
+      config3.anthropicApiBilling = true;
     } else if (arg === "--op-env" || arg.startsWith("--op-env=")) {
       const v = arg.startsWith("--op-env=") ? arg.slice("--op-env=".length) : args[++i];
       if (!v) {
@@ -63540,6 +63583,10 @@ ${h("OPTIONS")}
   ${green("--profile")} ${yellow("<name>")}         Use named profile for model mapping (default profile if omitted)
   ${green("--default-provider")} ${yellow("<name>")} Fallback provider for bare model names (builtin or customEndpoints key)
                            ${dim("Precedence: this flag > CLAUDISH_DEFAULT_PROVIDER env > config.json")}
+  ${green("--anthropic-api-billing")}  Use your real ANTHROPIC_API_KEY for native Claude models
+                           ${dim("(metered API billing). Default: the key is hidden so Claude Code")}
+                           ${dim("uses your claude.ai subscription. Env: CLAUDISH_ANTHROPIC_API_BILLING")}
+                           ${dim("Config: anthropicApiBilling: true")}
   ${green("--config")} ${yellow("<file>")}          Use THIS config file for the run, fully replacing the machine
                            ${dim("global (~/.claudish/config.json) AND project (.claudish.json).")}
                            ${dim("A file naming no op:// source never touches 1Password (no prompt).")}
@@ -63715,6 +63762,7 @@ ${h("ENVIRONMENT VARIABLES")}
   ${blue("CLAUDISH_CONTEXT_WINDOW")}         Override context window size
   ${blue("CLAUDISH_DIAG_MODE")}              Diagnostic output: auto / logfile / off
   ${blue("CLAUDISH_DEBUG")}                  Always enable debug logging: 1 / true ${dim("(same as -d)")}
+  ${blue("CLAUDISH_ANTHROPIC_API_BILLING")}  Bill native Claude to your API key ${dim("(see --anthropic-api-billing)")}
   ${blue("CLAUDISH_MCP_TOOLS")}              MCP tool gating: all / low-level / agentic / channel
   ${blue("CLAUDISH_MODEL_OPUS")}             Override model for Opus role
   ${blue("CLAUDISH_MODEL_SONNET")}           Override model for Sonnet role
@@ -71284,7 +71332,8 @@ __export(exports_claude_runner, {
   isProxyAuthMode: () => isProxyAuthMode,
   computeMainThreadContextWindow: () => computeMainThreadContextWindow,
   checkClaudeInstalled: () => checkClaudeInstalled,
-  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay
+  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay,
+  MIN_AUTO_COMPACT_WINDOW: () => MIN_AUTO_COMPACT_WINDOW
 });
 import { spawn as spawn4 } from "child_process";
 import {
@@ -71315,6 +71364,18 @@ function hasNativeAnthropicMapping(config3) {
   ];
   return models.some((m) => m && parseModelSpec(m).provider === "native-anthropic");
 }
+function wantsAnthropicApiBilling(config3) {
+  if (config3.anthropicApiBilling)
+    return true;
+  const raw2 = process.env[ENV.CLAUDISH_ANTHROPIC_API_BILLING];
+  if (raw2 !== undefined && raw2 !== "" && raw2 !== "0" && raw2.toLowerCase() !== "false")
+    return true;
+  try {
+    return loadConfig().anthropicApiBilling === true;
+  } catch {
+    return false;
+  }
+}
 function isProxyAuthMode(config3) {
   return !config3.monitor && !hasNativeAnthropicMapping(config3);
 }
@@ -71578,6 +71639,7 @@ async function runClaudeWithProxy(config3, proxyUrl, onCleanup) {
     [ENV.CLAUDISH_ACTIVE_MODEL_NAME]: modelDisplayName,
     CLAUDISH_IS_LOCAL: isLocalModel2 ? "true" : "false"
   };
+  let hidAnthropicApiKey = false;
   delete env.CLAUDECODE;
   if (config3.monitor) {
     delete env.ANTHROPIC_API_KEY;
@@ -71591,16 +71653,23 @@ async function runClaudeWithProxy(config3, proxyUrl, onCleanup) {
       env[ENV.ANTHROPIC_MODEL] = modelId;
       env[ENV.ANTHROPIC_SMALL_FAST_MODEL] = modelId;
     }
-    if (hasNativeAnthropicMapping(config3)) {} else {
-      env.ANTHROPIC_API_KEY = process.env.ANTHROPIC_API_KEY || "sk-ant-api03-placeholder-not-used-proxy-handles-auth-with-openrouter-key-xxxxxxxxxxxxxxxxxxxxx";
-      env.ANTHROPIC_AUTH_TOKEN = process.env.ANTHROPIC_AUTH_TOKEN || "placeholder-token-not-used-proxy-handles-auth";
+    if (hasNativeAnthropicMapping(config3)) {
+      if (process.env.ANTHROPIC_API_KEY && !wantsAnthropicApiBilling(config3)) {
+        delete env.ANTHROPIC_API_KEY;
+        hidAnthropicApiKey = true;
+      }
+    } else {
+      env.ANTHROPIC_API_KEY = "sk-ant-api03-placeholder-not-used-proxy-handles-auth-with-openrouter-key-xxxxxxxxxxxxxxxxxxxxx";
+      env.ANTHROPIC_AUTH_TOKEN = "placeholder-token-not-used-proxy-handles-auth";
       if (!process.env[ENV.CLAUDE_CODE_AUTO_COMPACT_WINDOW]) {
         const autoCompactWindow = await computeMainThreadContextWindow(config3);
-        if (autoCompactWindow > 0) {
+        if (autoCompactWindow >= MIN_AUTO_COMPACT_WINDOW) {
           env[ENV.CLAUDE_CODE_AUTO_COMPACT_WINDOW] = String(autoCompactWindow);
           if (!config3.quiet) {
             console.error(`[claudish] Auto-compact window: ${autoCompactWindow.toLocaleString()} tokens ` + "(Claude Code compacts before the backend's real limit)");
           }
+        } else if (autoCompactWindow > 0 && !config3.quiet) {
+          console.error(`[claudish] Model's real context window (${autoCompactWindow.toLocaleString()}) is below ` + `Claude Code's ${MIN_AUTO_COMPACT_WINDOW.toLocaleString()}-token auto-compact floor \u2014 ` + "leaving CLAUDE_CODE_AUTO_COMPACT_WINDOW unset so native auto-compaction stays on.");
         }
       }
     }
@@ -71616,6 +71685,9 @@ async function runClaudeWithProxy(config3, proxyUrl, onCleanup) {
   };
   if (!config3.monitor && hasNativeAnthropicMapping(config3)) {
     log2("[claudish] Native Claude model detected \u2014 using Claude Code subscription credentials");
+    if (hidAnthropicApiKey) {
+      log2("[claudish]   ANTHROPIC_API_KEY found but hidden so it can't override that subscription \xB7 " + "use --anthropic-api-billing (or anthropicApiBilling: true) to bill the API instead");
+    }
   }
   if (config3.interactive) {
     log2(`
@@ -71782,12 +71854,13 @@ async function checkClaudeInstalled() {
   const binary = await findClaudeBinary();
   return binary !== null;
 }
-var restoreTerminal = null;
+var restoreTerminal = null, MIN_AUTO_COMPACT_WINDOW = 200000;
 var init_claude_runner = __esm(() => {
   init_model_catalog();
   init_config();
   init_logger();
   init_model_discovery();
+  init_profile_config();
   init_model_parser();
   init_routing_rules();
   init_telemetry();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudish",
-  "version": "7.20.0",
+  "version": "7.20.1",
   "description": "Run Claude Code with any model - OpenRouter, Ollama, LM Studio & local models",
   "type": "module",
   "main": "./dist/index.js",
@@ -60,10 +60,10 @@
     "ai"
   ],
   "optionalDependencies": {
-    "@claudish/magmux-darwin-arm64": "7.20.0",
-    "@claudish/magmux-darwin-x64": "7.20.0",
-    "@claudish/magmux-linux-arm64": "7.20.0",
-    "@claudish/magmux-linux-x64": "7.20.0"
+    "@claudish/magmux-darwin-arm64": "7.20.1",
+    "@claudish/magmux-darwin-x64": "7.20.1",
+    "@claudish/magmux-linux-arm64": "7.20.1",
+    "@claudish/magmux-linux-x64": "7.20.1"
   },
   "author": "Jack Rudenko <i@madappgang.com>",
   "license": "MIT",