claudish 7.19.1 → 7.20.0

package/dist/index.js

@@ -651,7 +651,7 @@ var init_onepassword_config = __esm(() => {
 });
 
 // src/version.ts
-var VERSION = "7.19.1";
+var VERSION = "7.20.0";
 
 // src/logger.ts
 var exports_logger = {};
@@ -27650,9 +27650,6 @@ 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();
     }
@@ -28326,8 +28323,8 @@ var init_provider_definitions = __esm(() => {
       shortcuts: ["kc"],
       shortestPrefix: "kc",
       legacyPrefixes: [{ prefix: "kc/", stripPrefix: true }],
-      nativeModelPatterns: [{ pattern: /^kimi-for-coding$/i }],
-      fixedModel: "kimi-for-coding",
+      nativeModelPatterns: [{ pattern: /^kimi-for-coding/i }, { pattern: /^k3(-|$)/i }],
+      modelDiscovery: { path: "/models", format: "openai-models-list" },
       isDirectApi: true,
       description: "Kimi Coding Plan (kc@)"
     },
@@ -30475,6 +30472,22 @@ function lookupModelForProvider(modelId, provider, cachePath) {
     return;
   return entry.aggregators?.find((a) => a.provider === provider)?.contextWindow ?? entry.contextWindow;
 }
+function resolveSubscriptionRouting(modelId, provider, cachePath) {
+  const entry = findCacheEntry(modelId, cachePath);
+  if (!entry)
+    return { kind: "unknown" };
+  if (entry.subscriptionPlans?.includes(provider)) {
+    const agg = entry.aggregators?.find((a) => a.provider === provider);
+    return agg?.externalId ? { kind: "serves", externalId: agg.externalId } : { kind: "unknown" };
+  }
+  return isSubscriptionPlan(provider, cachePath) ? { kind: "not-served" } : { kind: "unknown" };
+}
+function isSubscriptionPlan(provider, cachePath) {
+  const cache = readAllModelsCache(cachePath);
+  if (!cache)
+    return false;
+  return cache.entries.some((e) => e.subscriptionPlans?.includes(provider));
+}
 function findCacheEntry(modelId, cachePath) {
   if (modelId.includes("@")) {
     throw new Error(`model-catalog lookup received provider-routed ID "${modelId}" \u2014 callers must strip the "@" prefix before calling`);
@@ -31745,14 +31758,6 @@ 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);
@@ -31782,7 +31787,7 @@ function createStreamingState() {
     accumulatedText: ""
   };
 }
-function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap, priorInputTokens) {
+function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap) {
   log(`[Streaming] ===== HANDLER STARTED for ${target} =====`);
   let isClosed = false;
   let ping = null;
@@ -31811,7 +31816,7 @@ data: ${JSON.stringify(d)}
           model: target,
           stop_reason: null,
           stop_sequence: null,
-          usage: messageStartUsage(priorInputTokens)
+          usage: { input_tokens: 100, output_tokens: 1 }
         }
       });
       send("ping", { type: "ping" });
@@ -31924,10 +31929,7 @@ data: ${JSON.stringify(d)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: stopReason, stop_sequence: null },
-            usage: {
-              ...state.usage?.prompt_tokens ? { input_tokens: state.usage.prompt_tokens } : {},
-              output_tokens: state.usage?.completion_tokens || 0
-            }
+            usage: { output_tokens: state.usage?.completion_tokens || 0 }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -31938,7 +31940,7 @@ 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(priorInputTokens || 100, estimatedOutputTokens);
+            onTokenUpdate(100, estimatedOutputTokens);
           }
         }
         if (!isClosed) {
@@ -37556,7 +37558,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: messageStartUsage(opts.priorInputTokens)
+          usage: { input_tokens: 100, output_tokens: 1 }
         }
       });
       send("ping", { type: "ping" });
@@ -37600,10 +37602,7 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: hasToolCalls ? "tool_use" : "end_turn", stop_sequence: null },
-            usage: {
-              ...inputTokens > 0 ? { input_tokens: inputTokens } : {},
-              output_tokens: outputTokens
-            }
+            usage: { output_tokens: outputTokens }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37806,7 +37805,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: messageStartUsage(opts.priorInputTokens)
+          usage: { input_tokens: 100, output_tokens: 1 }
         }
       });
       send("ping", { type: "ping" });
@@ -37827,10 +37826,7 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: "end_turn", stop_sequence: null },
-            usage: {
-              ...promptTokens > 0 ? { input_tokens: promptTokens } : {},
-              output_tokens: completionTokens
-            }
+            usage: { output_tokens: completionTokens }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37982,7 +37978,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: messageStartUsage(opts.priorInputTokens)
+          usage: { input_tokens: 100, output_tokens: 1 }
         }
       });
       send("ping", { type: "ping" });
@@ -38272,7 +38268,6 @@ class TokenTracker {
   sessionTotalCost = 0;
   sessionInputTokens = 0;
   sessionOutputTokens = 0;
-  lastInputTokens = 0;
   modelNameOverride;
   quotaRemaining;
   constructor(port, config2) {
@@ -38289,11 +38284,10 @@ class TokenTracker {
     this.quotaRemaining = fraction;
   }
   rewrite() {
-    this.writeFile(this.getLastInputTokens(), this.sessionOutputTokens);
+    this.writeFile(this.sessionInputTokens, 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;
@@ -38302,7 +38296,6 @@ 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;
@@ -38311,7 +38304,6 @@ class TokenTracker {
   }
   updateWithDelta(inputTokens, outputTokens) {
     let incrementalInputTokens;
-    this.lastInputTokens = inputTokens;
     if (inputTokens >= this.sessionInputTokens) {
       incrementalInputTokens = inputTokens - this.sessionInputTokens;
       this.sessionInputTokens = inputTokens;
@@ -38327,11 +38319,10 @@ class TokenTracker {
     const pricing = this.getPricing();
     const cost = incrementalInputTokens / 1e6 * pricing.inputCostPer1M + outputTokens / 1e6 * pricing.outputCostPer1M;
     this.sessionTotalCost += cost;
-    this.writeFile(inputTokens, this.sessionOutputTokens, pricing.isEstimate);
+    this.writeFile(Math.max(inputTokens, this.sessionInputTokens), 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;
@@ -38347,7 +38338,6 @@ class TokenTracker {
   updateLocal(inputTokens, outputTokens) {
     if (inputTokens > 0) {
       this.sessionInputTokens = inputTokens;
-      this.lastInputTokens = inputTokens;
     }
     this.sessionOutputTokens += outputTokens;
     this.writeFile(this.sessionInputTokens, this.sessionOutputTokens);
@@ -38361,9 +38351,6 @@ class TokenTracker {
   getInputTokens() {
     return this.sessionInputTokens;
   }
-  getLastInputTokens() {
-    return this.lastInputTokens || this.sessionInputTokens;
-  }
   getOutputTokens() {
     return this.sessionOutputTokens;
   }
@@ -38905,18 +38892,16 @@ 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, priorInputTokens);
+        return createStreamingResponseHandler(c, response, adapter, this.bareModelName, this.middlewareManager, onTokenUpdate, claudeRequest.tools, toolNameMap);
       case "openai-responses-sse":
         return createResponsesStreamHandler(c, response, {
           modelName: this.bareModelName,
           onTokenUpdate,
           toolNameMap: adapter.getToolNameMap(),
           contextWindow: lookupModelForProvider(this.bareModelName, this.provider.name),
-          onApiError,
-          priorInputTokens
+          onApiError
         });
       case "anthropic-sse":
         return createAnthropicPassthroughStream(c, response, {
@@ -38936,15 +38921,13 @@ class ComposedHandler {
           middlewareManager: this.middlewareManager,
           onTokenUpdate,
           onToolCall,
-          unwrapResponse: this.options.unwrapGeminiResponse,
-          priorInputTokens
+          unwrapResponse: this.options.unwrapGeminiResponse
         });
       }
       case "ollama-jsonl":
         return createOllamaJsonlStream(c, response, {
           modelName: this.bareModelName,
-          onTokenUpdate,
-          priorInputTokens
+          onTokenUpdate
         });
       default:
         throw new Error(`Unknown stream format: ${streamFormat}`);
@@ -42584,7 +42567,8 @@ var init_default_routing_rules = __esm(() => {
     "o3-*": ["openai-codex", "openai", "openrouter"],
     "gemini-*": ["gemini-codeassist", "google", "openrouter"],
     "grok-*": ["x-ai", "openrouter"],
-    "kimi-*": ["kimi-coding@kimi-for-coding", "kimi", "openrouter"],
+    "kimi-*": ["kimi-coding", "kimi", "openrouter"],
+    "k3*": ["kimi-coding", "kimi", "openrouter"],
     "minimax-*": ["minimax-coding", "minimax", "openrouter"],
     "glm-*": ["glm-coding", "glm", "openrouter"],
     "z-ai-*": ["z-ai", "openrouter"],
@@ -42638,7 +42622,7 @@ function matchRoutingRule(modelName, rules) {
     return rules["*"];
   return null;
 }
-function buildRoutingChain(entries, originalModelName) {
+function buildRoutingChain(entries, originalModelName, cachePath) {
   const routes = [];
   for (const entry of entries) {
     const atIdx = entry.indexOf("@");
@@ -42652,6 +42636,13 @@ function buildRoutingChain(entries, originalModelName) {
       modelName = originalModelName;
     }
     const provider = PROVIDER_SHORTCUTS[providerRaw.toLowerCase()] ?? providerRaw.toLowerCase();
+    if (atIdx === -1) {
+      const routing = resolveSubscriptionRouting(modelName, provider, cachePath);
+      if (routing.kind === "not-served")
+        continue;
+      if (routing.kind === "serves")
+        modelName = routing.externalId;
+    }
     let modelSpec;
     if (provider === "openrouter") {
       const resolution = resolveModelNameSync(modelName, "openrouter");
@@ -42678,7 +42669,7 @@ function globMatch(pattern, value) {
 async function hasCredentialsForProvider(provider) {
   return credentials.isAvailable(provider);
 }
-async function routeExplicit(modelSpec, model, provider) {
+async function routeExplicit(modelSpec, model, provider, cachePath) {
   if (!await hasCredentialsForProvider(provider)) {
     return {
       kind: "no-route",
@@ -42686,7 +42677,7 @@ async function routeExplicit(modelSpec, model, provider) {
       hint: buildCredentialHint(model, [provider]) ?? undefined
     };
   }
-  const built = buildRoutingChain([modelSpec], model)[0];
+  const built = buildRoutingChain([modelSpec], model, cachePath)[0];
   if (!built) {
     return {
       kind: "no-route",
@@ -42695,7 +42686,7 @@ async function routeExplicit(modelSpec, model, provider) {
   }
   return { kind: "ok", primary: built, fallbacks: [] };
 }
-async function routeBare(model, nativeProvider, rules, defaultProvider) {
+async function routeBare(model, nativeProvider, rules, defaultProvider, cachePath) {
   const matched = matchRoutingRule(model, rules) ?? [];
   const entries = [...matched];
   if (defaultProvider && defaultProvider.length > 0) {
@@ -42716,7 +42707,7 @@ async function routeBare(model, nativeProvider, rules, defaultProvider) {
       hint: buildCredentialHint(model, [nativeProvider]) ?? undefined
     };
   }
-  const candidates = buildRoutingChain(entries, model);
+  const candidates = buildRoutingChain(entries, model, cachePath);
   const credentialed = [];
   const skipped = [];
   const checks4 = await Promise.all(candidates.map((candidate) => hasCredentialsForProvider(candidate.provider)));
@@ -42737,16 +42728,17 @@ async function routeBare(model, nativeProvider, rules, defaultProvider) {
   const [primary, ...fallbacks] = credentialed;
   return { kind: "ok", primary, fallbacks };
 }
-async function route(modelSpec, rulesOverride, defaultProviderOverride) {
+async function route(modelSpec, rulesOverride, defaultProviderOverride, cachePath) {
   const parsed = parseModelSpec(modelSpec);
   if (parsed.isExplicitProvider) {
-    return routeExplicit(modelSpec, parsed.model, parsed.provider);
+    return routeExplicit(modelSpec, parsed.model, parsed.provider, cachePath);
   }
   const rules = rulesOverride ?? loadRoutingRules();
   const defaultProvider = defaultProviderOverride !== undefined ? defaultProviderOverride : rulesOverride !== undefined ? undefined : loadConfig().defaultProvider;
-  return routeBare(parsed.model, parsed.provider, rules, defaultProvider);
+  return routeBare(parsed.model, parsed.provider, rules, defaultProvider, cachePath);
 }
 var init_routing_rules = __esm(() => {
+  init_model_catalog();
   init_authority();
   init_profile_config();
   init_auto_route();
@@ -58419,8 +58411,7 @@ 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_ANTHROPIC_API_BILLING: "CLAUDISH_ANTHROPIC_API_BILLING"
+    CLAUDISH_DEBUG: "CLAUDISH_DEBUG"
   };
   OPENROUTER_HEADERS = {
     "HTTP-Referer": "https://claudish.com",
@@ -58577,6 +58568,114 @@ var init_model_catalog2 = __esm(() => {
   NO_CATALOG_VENDOR_SLUGS = new Set(["litellm", "ollama", "lmstudio", "lm-studio"]);
 });
 
+// src/providers/model-discovery.ts
+function resolveBaseUrl2(catalogName) {
+  const def = getProviderByName(catalogName);
+  if (!def)
+    return null;
+  for (const envVar of def.baseUrlEnvVars ?? []) {
+    const v = process.env[envVar];
+    if (v)
+      return v.replace(/\/+$/, "");
+  }
+  return (def.baseUrl || "").replace(/\/+$/, "") || null;
+}
+function readContextWindow(row) {
+  for (const field of ["context_length", "context_window", "max_context_length"]) {
+    const v = row[field];
+    if (typeof v === "number" && Number.isFinite(v) && v > 0)
+      return v;
+  }
+  return;
+}
+function parseOpenAIModelsList(body) {
+  const data = body?.data;
+  if (!Array.isArray(data))
+    return [];
+  const models = [];
+  for (const raw2 of data) {
+    if (!raw2 || typeof raw2 !== "object")
+      continue;
+    const row = raw2;
+    const id = row.id;
+    if (typeof id !== "string" || id.trim().length === 0)
+      continue;
+    const displayName = typeof row.display_name === "string" ? row.display_name : undefined;
+    models.push({ id, displayName, contextWindow: readContextWindow(row) });
+  }
+  return models;
+}
+async function discoverProviderModels(providerName) {
+  const cached2 = _cache2.get(providerName);
+  if (cached2 && cached2.expiresAt > Date.now())
+    return cached2.models;
+  const def = getProviderByName(providerName);
+  const descriptor = def?.modelDiscovery;
+  if (!def || !descriptor)
+    return [];
+  const baseUrl = resolveBaseUrl2(providerName);
+  if (!baseUrl)
+    return [];
+  const endpoint = `${baseUrl}${descriptor.path}`;
+  let headers = {};
+  try {
+    const auth = await credentials.getRequestAuth(providerName, { model: "" });
+    headers = { ...auth.headers };
+  } catch (e) {
+    log(`[model-discovery:${providerName}] no credentials, skipping: ${e?.message}`);
+    return [];
+  }
+  let response;
+  try {
+    response = await fetch(endpoint, {
+      method: "GET",
+      headers,
+      signal: AbortSignal.timeout(FETCH_TIMEOUT_MS2)
+    });
+  } catch (e) {
+    log(`[model-discovery:${providerName}] fetch failed: ${e?.message}`);
+    return [];
+  }
+  if (!response.ok) {
+    log(`[model-discovery:${providerName}] HTTP ${response.status} from ${endpoint}`);
+    return [];
+  }
+  let body;
+  try {
+    body = await response.json();
+  } catch {
+    log(`[model-discovery:${providerName}] response was not JSON`);
+    return [];
+  }
+  const models = parseOpenAIModelsList(body);
+  if (models.length === 0) {
+    log(`[model-discovery:${providerName}] endpoint reachable but listed no models`);
+    return [];
+  }
+  log(`[model-discovery:${providerName}] discovered ${models.length} models: ` + models.map((m) => `${m.id}(${m.contextWindow ?? "?"})`).join(", "));
+  _cache2.set(providerName, { models, expiresAt: Date.now() + CACHE_TTL_MS3 });
+  return models;
+}
+async function discoverContextWindow(providerName, modelId) {
+  const models = await discoverProviderModels(providerName);
+  const match2 = models.find((m) => m.id.toLowerCase() === modelId.toLowerCase());
+  return match2?.contextWindow;
+}
+function rankDiscoveredModels(models) {
+  return [...models].sort((a, b) => {
+    const diff = (b.contextWindow ?? 0) - (a.contextWindow ?? 0);
+    return diff !== 0 ? diff : a.id.localeCompare(b.id);
+  });
+}
+var CACHE_TTL_MS3, FETCH_TIMEOUT_MS2 = 5000, _cache2;
+var init_model_discovery = __esm(() => {
+  init_authority();
+  init_logger();
+  init_provider_definitions();
+  CACHE_TTL_MS3 = 5 * 60 * 1000;
+  _cache2 = new Map;
+});
+
 // src/providers/ollama-discovery.ts
 function ollamaBaseUrl() {
   return process.env.OLLAMA_HOST || process.env.OLLAMA_BASE_URL || "http://localhost:11434";
@@ -59146,8 +59245,22 @@ async function selectModelFromProvider(provider, tierName, recommendedModels, _f
   const prefix = PROVIDER_MODEL_PREFIX[provider] || `${provider}@`;
   const displayName = getPickerDisplayName(provider);
   const def = getProviderByName(provider);
-  if (def?.fixedModel) {
-    return buildExplicitModelSpec(provider, def.fixedModel);
+  if (def?.modelDiscovery) {
+    const discovered = rankDiscoveredModels(await discoverProviderModels(provider));
+    if (discovered.length > 0) {
+      const discoveredModels = discovered.map((m) => ({
+        id: m.id,
+        name: m.displayName || m.id,
+        description: m.contextWindow ? `${m.displayName ?? m.id} \xB7 ${Math.round(m.contextWindow / 1024)}K context` : m.displayName ?? m.id,
+        provider: displayName,
+        supportsTools: true,
+        isFree: true,
+        source: displayName
+      }));
+      const picked = await pickModelFromList(provider, displayName, tierName, discoveredModels);
+      if (picked)
+        return picked;
+    }
   }
   if (provider === "ollama") {
     const ollamaModels = await fetchOllamaModels2({ enrichCapabilities: false });
@@ -59326,6 +59439,7 @@ var init_model_selector = __esm(() => {
   init_authority();
   init_model_loader();
   init_model_catalog2();
+  init_model_discovery();
   init_provider_definitions();
   pickerProviderToFirebaseSlug = {
     openrouter: "openrouter",
@@ -62377,8 +62491,6 @@ 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) {
@@ -63428,10 +63540,6 @@ ${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).")}
@@ -63607,7 +63715,6 @@ ${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
@@ -64733,7 +64840,7 @@ function writeProbeModelsCache(data, path2 = PROBE_MODELS_CACHE_PATH) {
   mkdirSync14(dirname7(path2), { recursive: true });
   writeFileSync16(path2, JSON.stringify(data), "utf-8");
 }
-function isCacheFresh(data, ttlMs = CACHE_TTL_MS3) {
+function isCacheFresh(data, ttlMs = CACHE_TTL_MS4) {
   if (!data?.generatedAt)
     return false;
   const generatedMs = Date.parse(data.generatedAt);
@@ -64741,7 +64848,7 @@ function isCacheFresh(data, ttlMs = CACHE_TTL_MS3) {
     return false;
   return Date.now() - generatedMs < ttlMs;
 }
-async function fetchProbeModels(url2 = PROBE_MODELS_URL, timeoutMs = FETCH_TIMEOUT_MS2) {
+async function fetchProbeModels(url2 = PROBE_MODELS_URL, timeoutMs = FETCH_TIMEOUT_MS3) {
   let response;
   try {
     response = await fetch(url2, { signal: AbortSignal.timeout(timeoutMs) });
@@ -64847,9 +64954,9 @@ function isValidResponse(raw2) {
     return false;
   return true;
 }
-var PROBE_MODELS_URL = "https://us-central1-claudish-6da10.cloudfunctions.net/probeModels", CACHE_TTL_MS3, FETCH_TIMEOUT_MS2 = 15000, PROBE_MODELS_CACHE_PATH, _inFlight = null;
+var PROBE_MODELS_URL = "https://us-central1-claudish-6da10.cloudfunctions.net/probeModels", CACHE_TTL_MS4, FETCH_TIMEOUT_MS3 = 15000, PROBE_MODELS_CACHE_PATH, _inFlight = null;
 var init_probe_catalog = __esm(() => {
-  CACHE_TTL_MS3 = 60 * 60 * 1000;
+  CACHE_TTL_MS4 = 60 * 60 * 1000;
   PROBE_MODELS_CACHE_PATH = join24(homedir23(), ".claudish", "probe-models.json");
 });
 
@@ -71177,8 +71284,7 @@ __export(exports_claude_runner, {
   isProxyAuthMode: () => isProxyAuthMode,
   computeMainThreadContextWindow: () => computeMainThreadContextWindow,
   checkClaudeInstalled: () => checkClaudeInstalled,
-  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay,
-  MIN_AUTO_COMPACT_WINDOW: () => MIN_AUTO_COMPACT_WINDOW
+  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay
 });
 import { spawn as spawn4 } from "child_process";
 import {
@@ -71209,18 +71315,6 @@ 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);
 }
@@ -71428,7 +71522,7 @@ async function computeMainThreadContextWindow(config3, cachePath) {
           continue;
         provider = plan.primary.provider;
       }
-      const win = lookupModelForProvider(parsed.model, provider, cachePath);
+      const win = await discoverContextWindow(provider, parsed.model) ?? lookupModelForProvider(parsed.model, provider, cachePath);
       if (typeof win === "number" && win > 0) {
         min = Math.min(min, win);
       }
@@ -71484,7 +71578,6 @@ 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;
@@ -71498,23 +71591,16 @@ async function runClaudeWithProxy(config3, proxyUrl, onCleanup) {
       env[ENV.ANTHROPIC_MODEL] = modelId;
       env[ENV.ANTHROPIC_SMALL_FAST_MODEL] = modelId;
     }
-    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 (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 (!process.env[ENV.CLAUDE_CODE_AUTO_COMPACT_WINDOW]) {
         const autoCompactWindow = await computeMainThreadContextWindow(config3);
-        if (autoCompactWindow >= MIN_AUTO_COMPACT_WINDOW) {
+        if (autoCompactWindow > 0) {
           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.");
         }
       }
     }
@@ -71530,9 +71616,6 @@ 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(`
@@ -71699,12 +71782,12 @@ async function checkClaudeInstalled() {
   const binary = await findClaudeBinary();
   return binary !== null;
 }
-var restoreTerminal = null, MIN_AUTO_COMPACT_WINDOW = 200000;
+var restoreTerminal = null;
 var init_claude_runner = __esm(() => {
   init_model_catalog();
   init_config();
   init_logger();
-  init_profile_config();
+  init_model_discovery();
   init_model_parser();
   init_routing_rules();
   init_telemetry();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudish",
-  "version": "7.19.1",
+  "version": "7.20.0",
   "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": "7.19.1",
-    "@claudish/magmux-darwin-x64": "7.19.1",
-    "@claudish/magmux-linux-arm64": "7.19.1",
-    "@claudish/magmux-linux-x64": "7.19.1"
+    "@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"
   },
   "author": "Jack Rudenko <i@madappgang.com>",
   "license": "MIT",