claudish 14.16.2 → 14.17.0

package/dist/index.js

@@ -133,6 +133,9 @@ var init_config = __esm(() => {
     CLAUDISH_PROVIDER_NAME: "CLAUDISH_PROVIDER_NAME",
     ANTHROPIC_MODEL: "ANTHROPIC_MODEL",
     ANTHROPIC_SMALL_FAST_MODEL: "ANTHROPIC_SMALL_FAST_MODEL",
+    CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST: "CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST",
+    CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL: "_CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL",
+    CLAUDISH_SET_HOST_ROUTING_VARS: "CLAUDISH_SET_HOST_ROUTING_VARS",
     CLAUDISH_MODEL_OPUS: "CLAUDISH_MODEL_OPUS",
     CLAUDISH_MODEL_SONNET: "CLAUDISH_MODEL_SONNET",
     CLAUDISH_MODEL_HAIKU: "CLAUDISH_MODEL_HAIKU",
@@ -5491,7 +5494,7 @@ var init_onepassword_config = __esm(() => {
 });
 
 // src/version.ts
-var VERSION = "14.16.2";
+var VERSION = "14.17.0";
 
 // src/redact.ts
 function isPlaceholder(value) {
@@ -34202,6 +34205,14 @@ function toModelInfo(raw) {
   if (tp && typeof tp.mode === "string" && typeof tp.limit === "number") {
     info.truncationPolicy = { mode: tp.mode, limit: tp.limit };
   }
+  if (Array.isArray(m.service_tiers)) {
+    info.serviceTiers = m.service_tiers.flatMap((tier) => {
+      const t = tier;
+      return t && typeof t.id === "string" && typeof t.name === "string" ? [{ id: t.id, name: t.name }] : [];
+    });
+  }
+  if (typeof m.default_service_tier === "string")
+    info.defaultServiceTier = m.default_service_tier;
   return info;
 }
 function toModelTable(body) {
@@ -34298,7 +34309,7 @@ async function firstAnswer(entry, slug, binding, opts, now) {
       log(`[CodexDiscovery] ${slug}: /models not answered within ${waitMs} ms`);
     }
   }
-  return slug ? fromModelsCache(slug, opts.codexHome) : undefined;
+  return slug && !opts.accountOnly ? fromModelsCache(slug, opts.codexHome) : undefined;
 }
 function fromModelsCache(slug, codexHome) {
   const info = toModelTable(readModelsCache(codexHome))?.get(slug.toLowerCase());
@@ -41084,6 +41095,34 @@ function parseModelParams(spec, into = {}) {
   return into;
 }
 
+// src/providers/codex-service-tier.ts
+function findCodexServiceTier(speed, tiers) {
+  if (!tiers)
+    return;
+  const matches = (tier) => tier.id === speed || tier.name.toLowerCase() === speed;
+  return tiers.find(matches);
+}
+function decideCodexSpeed(speed, modelInfo) {
+  if (speed.value === "standard") {
+    return { note: "speed standard: field omitted (connection default)" };
+  }
+  if (modelInfo === "unsigned") {
+    const message = `speed "${speed.value}" is not available on this connection (no subscription account signed the request)`;
+    return speed.explicit ? { note: message, refusal: message } : { note: message };
+  }
+  const tier = findCodexServiceTier(speed.value, modelInfo?.serviceTiers);
+  if (tier) {
+    return {
+      tier: tier.id,
+      note: `speed ${speed.value}: ${CODEX_SERVICE_TIER_POINTER} set to the account's ${speed.value} tier (from the account's dynamic models catalog)`
+    };
+  }
+  const listed = modelInfo?.serviceTiers?.length ?? 0;
+  const message = modelInfo ? `speed "${speed.value}" is not offered on this connection (the account lists ${listed} service tier${listed === 1 ? "" : "s"})` : `speed "${speed.value}" is not available on this connection (the account's dynamic models catalog did not answer; retry)`;
+  return speed.explicit ? { note: message, refusal: message } : { note: message };
+}
+var CODEX_SERVICE_TIER_POINTER = "/service_tier";
+
 // src/handlers/shared/call-context.ts
 function createSigningSecrets() {
   return { values: new Set, lease: acquireSigningLease() };
@@ -48151,6 +48190,34 @@ var init_authority = __esm(() => {
   installBillingProbes();
 });
 
+// src/providers/catalog-endpoints.ts
+function catalogUrl() {
+  return process.env.CLAUDISH_CATALOG_URL ?? DEFAULT_CATALOG_URL;
+}
+function siblingEndpoint(pathname, fallback) {
+  try {
+    const url = new URL(catalogUrl());
+    url.pathname = url.pathname.replace(/\/queryModels$/, pathname);
+    url.search = "";
+    return url.toString();
+  } catch {
+    return fallback;
+  }
+}
+function plansUrl() {
+  return process.env.CLAUDISH_PLANS_URL ?? siblingEndpoint("/queryPlans", `${DEFAULT_HOST}/queryPlans`);
+}
+function probeModelsUrl() {
+  return process.env.CLAUDISH_PROBE_MODELS_URL ?? siblingEndpoint("/probeModels", `${DEFAULT_HOST}/probeModels`);
+}
+function modelsBaseUrl() {
+  return siblingEndpoint("/queryModels", `${DEFAULT_HOST}/queryModels`);
+}
+var DEFAULT_HOST = "https://us-central1-claudish-6da10.cloudfunctions.net", DEFAULT_CATALOG_URL;
+var init_catalog_endpoints = __esm(() => {
+  DEFAULT_CATALOG_URL = `${DEFAULT_HOST}/queryModels?status=all&catalog=slim&includeRouteVariants=true&limit=1000`;
+});
+
 // src/providers/catalog-v3.ts
 function parseCatalogV3Envelope(value) {
   if (!value || typeof value !== "object")
@@ -48619,22 +48686,6 @@ var init_probe_discovery = __esm(() => {
 });
 
 // src/providers/catalog-client.ts
-function catalogUrl() {
-  return process.env.CLAUDISH_CATALOG_URL ?? DEFAULT_CATALOG_URL;
-}
-function derivePlansUrl(modelsUrl) {
-  try {
-    const url = new URL(modelsUrl);
-    url.pathname = url.pathname.replace(/\/queryModels$/, "/queryPlans");
-    url.search = "";
-    return url.toString();
-  } catch {
-    return "https://us-central1-claudish-6da10.cloudfunctions.net/queryPlans";
-  }
-}
-function plansUrl() {
-  return process.env.CLAUDISH_PLANS_URL ?? derivePlansUrl(catalogUrl());
-}
 function buildRoutesUrl(modelsUrl, generationId) {
   const url = new URL(modelsUrl);
   url.search = "";
@@ -48834,8 +48885,9 @@ async function refreshCatalogOnce(timeoutMs, options) {
   let pages = 0;
   const seenCursors = new Set;
   let expectedTotal;
+  const base = catalogUrl();
   for (;; ) {
-    const result = await fetchEnvelope(buildCatalogPageUrl(catalogUrl(), cursor, CATALOG_PAGE_LIMIT, generationId), timeoutMs);
+    const result = await fetchEnvelope(buildCatalogPageUrl(base, cursor, CATALOG_PAGE_LIMIT, generationId), timeoutMs);
     if (!result.ok) {
       if (result.reason === "incompatible") {
         return { kind: "incompatible", serverContractVersion: result.serverContractVersion };
@@ -48884,7 +48936,7 @@ async function refreshCatalogOnce(timeoutMs, options) {
     }
     return { kind: "fetch_failed", reason: planResult.reason };
   }
-  const registryResult = await fetchRouteRegistry(timeoutMs, generationId);
+  const registryResult = await fetchRouteRegistry(timeoutMs, generationId, base);
   if (!registryResult.ok) {
     if (registryResult.reason === "incompatible") {
       return { kind: "incompatible", serverContractVersion: registryResult.serverContractVersion };
@@ -48962,8 +49014,8 @@ async function fetchSubscriptionPlans(timeoutMs, generationId) {
   }
   return { ok: true, plans };
 }
-async function fetchRouteRegistry(timeoutMs, generationId) {
-  const result = await fetchEnvelope(buildRoutesUrl(catalogUrl(), generationId), timeoutMs);
+async function fetchRouteRegistry(timeoutMs, generationId, modelsUrl) {
+  const result = await fetchEnvelope(buildRoutesUrl(modelsUrl, generationId), timeoutMs);
   if (!result.ok)
     return result;
   if (result.envelope.generationId !== generationId) {
@@ -49005,7 +49057,7 @@ async function ensureRouteRegistryReady(timeoutMs = 5000) {
   if (cache?.routeRegistry)
     return;
   if (cache && !catalogWarmDisabled()) {
-    const result = await fetchRouteRegistry(timeoutMs, cache.catalogGenerationId);
+    const result = await fetchRouteRegistry(timeoutMs, cache.catalogGenerationId, catalogUrl());
     if (result.ok) {
       const routeRegistry = { generationId: cache.catalogGenerationId, registry: result.registry };
       if (wiresMatchRegistry(cache.entries, routeRegistry, cache.catalogGenerationId)) {
@@ -49028,10 +49080,11 @@ async function waitForCatalogWarm(timeoutMs) {
   if (timedOut)
     log(`[catalog] still refreshing after ${timeoutMs} ms; the caller goes on without it`);
 }
-var DEFAULT_CATALOG_URL = "https://us-central1-claudish-6da10.cloudfunctions.net/queryModels?status=all&catalog=slim&includeRouteVariants=true&limit=1000", MAX_CATALOG_PAGES = 40, CATALOG_PAGE_LIMIT = 1000, _memCache = null, _catalogEntriesForTest, _warmPromise = null;
+var MAX_CATALOG_PAGES = 40, CATALOG_PAGE_LIMIT = 1000, _memCache = null, _catalogEntriesForTest, _warmPromise = null;
 var init_catalog_client = __esm(() => {
   init_logger();
   init_all_models_cache();
+  init_catalog_endpoints();
   init_catalog_route_bindings();
   init_connection_wire();
   init_probe_discovery();
@@ -50254,7 +50307,7 @@ async function discoverEveryServedModel(providerName) {
         detail: oneLine3(e?.message ?? "")
       });
     }
-    if (!hasAuthHeader(headers)) {
+    if (!hasAuthHeader(headers) && def.authScheme !== "none") {
       return recordFailure({ kind: "no-credentials", provider: providerName, endpoint });
     }
   }
@@ -50534,6 +50587,15 @@ function setPointer(body, path, value) {
   }
   node[tokens[tokens.length - 1]] = value;
 }
+function readPointer(body, path) {
+  let node = body;
+  for (const token of pointerTokens(path)) {
+    if (node === null || typeof node !== "object" || Array.isArray(node))
+      return;
+    node = node[token];
+  }
+  return node;
+}
 function deletePointer(body, path) {
   const tokens = pointerTokens(path);
   const trail = [body];
@@ -50610,6 +50672,18 @@ function applyControl(body, control, requested, label, explicit, result) {
   result.headers.push(...control.headers);
   result.notes.push(`${label} ${requested}: ${control.path} = ${JSON.stringify(native)}${control.access !== "open" ? ` (access: ${control.access})` : ""}`);
 }
+function wireSpeedOffered(control, speed) {
+  return control.status === "supported" && control.values[speed] !== undefined;
+}
+function applyWireSpeed(body, wire, speed) {
+  const result = { headers: [], notes: [], borrowed: [] };
+  if (wire.speed.status === "supported" && readPointer(body, wire.speed.path) !== undefined) {
+    result.notes.push(`speed ${speed.value}: ${wire.speed.path} already set by --model-params`);
+    return result;
+  }
+  applyControl(body, wire.speed, speed.value, "speed", speed.explicit, result);
+  return result;
+}
 function writeAssignments(put, assignments) {
   for (const assignment of assignments)
     put(assignment.path, assignment.value);
@@ -50695,7 +50769,8 @@ function wireChoicesFromClaudeRequest(claudeRequest, pinnedEffort) {
     effort: effort === "none" ? undefined : effort,
     reasoningOff,
     hasTools: Array.isArray(claudeRequest?.tools) && claudeRequest.tools.length > 0,
-    maxTokens: typeof claudeRequest?.max_tokens === "number" ? claudeRequest.max_tokens : undefined
+    maxTokens: typeof claudeRequest?.max_tokens === "number" ? claudeRequest.max_tokens : undefined,
+    ...claudeRequest?.speed === "fast" && { speed: { value: "fast", explicit: false } }
   };
 }
 function outputLimitToSend(wire, choices, result) {
@@ -50762,7 +50837,7 @@ function recoverBorrowedWireFields(payload, borrowed, errorText) {
     deletePointer(next, path);
   return {
     payload: next,
-    note: `dropped ${drop.join(", ")} (borrowed from the owner's connection; this connection's value is unknown)`
+    note: `dropped ${drop.join(", ")} (not published by this connection's catalog wire)`
   };
 }
 function mergeWireHeaders(headers, extra) {
@@ -56140,6 +56215,17 @@ class ComposedHandler {
       subscription: CREDENTIAL_DECIDED_PROVIDERS.has(name) ? binding?.auth?.arm === "oauth" : isSubscriptionProvider(name)
     };
   }
+  wireWhere() {
+    const connection = this.options.connection;
+    return `${connection?.route.routeId}/${connection?.route.routeProfileId} \xB7 ${connection?.externalModelId}`;
+  }
+  noticeFastIgnored(source) {
+    if (fastIgnoredNoticed.has(this.targetModel))
+      return;
+    fastIgnoredNoticed.add(this.targetModel);
+    const advice = source === "catalog" ? "the fast tier is billed at a higher per-token price. Add #fast to the model spec to send it anyway." : "the Codex fast tier is sent only on a request signed by a Codex subscription account.";
+    logStderr(`${this.targetModel}: Claude Code /fast ignored, the request is metered and ${advice}`);
+  }
   respondCallAborted(startTime, fallbackMeta) {
     log(`[${this.provider.displayName}] call stopped: its deadline passed`);
     try {
@@ -56583,6 +56669,8 @@ class ComposedHandler {
     const toolNameMap = adapter.getToolNameMap();
     let wireHeaders = [];
     let wireBorrowed = [];
+    let codexSpeed;
+    let fastSpeed;
     const wire = this.options.connection?.wire;
     if (wire?.status === "known") {
       const choices = {
@@ -56605,6 +56693,13 @@ class ComposedHandler {
       }
       if (wireSpeed)
         choices.speed = { value: wireSpeed, explicit: true };
+      if (choices.speed && this.provider.name === "openai-codex" && wire.speed.status === "unknown") {
+        codexSpeed = choices.speed;
+        choices.speed = undefined;
+      } else if (choices.speed && !choices.speed.explicit) {
+        fastSpeed = choices.speed;
+        choices.speed = undefined;
+      }
       const shaped = shapeRequestFromWire(requestPayload, wire, choices, this.options.connection?.ownerWire);
       wireBorrowed = shaped.borrowed;
       const where = `${this.options.connection?.route.routeId}/${this.options.connection?.route.routeProfileId} \xB7 ${this.options.connection?.externalModelId}`;
@@ -56705,6 +56800,18 @@ class ComposedHandler {
         this.tokenTracker.setContextWindow(providerWindow);
       }
     }
+    if (fastSpeed && wire?.status === "known") {
+      if (!this.signedBilling(binding).subscription) {
+        log(`[Wire] ${this.wireWhere()}: speed ${fastSpeed.value} (Claude Code /fast) not sent: the request is metered`);
+        if (wireSpeedOffered(wire.speed, fastSpeed.value))
+          this.noticeFastIgnored("catalog");
+      } else {
+        const applied = applyWireSpeed(requestPayload, wire, fastSpeed);
+        for (const note of applied.notes)
+          log(`[Wire] ${this.wireWhere()}: ${note}`);
+        wireHeaders = [...wireHeaders, ...applied.headers];
+      }
+    }
     if (this.provider.name === "openai-codex") {
       const resume = (mayRequestModule) => resumeFromToolResults(claudeRequest, binding, peekCodeModeStores(), claudeRequest?.tool_choice, {
         setting: this.options.codeMode ?? DEFAULT_CODE_MODE_SETTING,
@@ -56729,6 +56836,30 @@ class ComposedHandler {
     if (this.provider.transformPayload) {
       requestPayload = this.provider.transformPayload(requestPayload, claudeRequest, binding);
     }
+    if (codexSpeed) {
+      addSigningSecrets(c, credentialValuesFromRequest(binding?.auth?.headers, this.signedEndpoint(binding)));
+      const accountKey = accountKeyFromBinding(binding);
+      const subscription = this.signedBilling(binding).subscription;
+      const modelInfo = !subscription ? "unsigned" : binding && accountKey ? await getCodexModelInfo(String(requestPayload?.model ?? ""), binding, accountKey, {
+        accountOnly: true
+      }) : undefined;
+      const outcome = decideCodexSpeed(codexSpeed, modelInfo);
+      const where = this.wireWhere();
+      log(`[Wire] ${where}: ${outcome.note}`);
+      if (!subscription && !codexSpeed.explicit && codexSpeed.value !== "standard") {
+        this.noticeFastIgnored("codex");
+      }
+      if (outcome.refusal) {
+        return c.json(wrapAnthropicError(400, redactForClient(`claudish: ${outcome.refusal} (${where})`), "invalid_request_error"), 400);
+      }
+      if (outcome.tier && requestPayload.service_tier !== undefined) {
+        log(`[Wire] ${where}: service_tier already set by --model-params; speed tier not written`);
+      } else if (outcome.tier) {
+        requestPayload.service_tier = outcome.tier;
+        if (!codexSpeed.explicit)
+          wireBorrowed = [...wireBorrowed, CODEX_SERVICE_TIER_POINTER];
+      }
+    }
     let codeModeShape;
     let codeModePrePayload;
     let codeModeContext;
@@ -58049,7 +58180,7 @@ function getRecoveryHint(status, errorText, providerName, transportTerminal429,
   }
   return `Unexpected HTTP ${status} from ${providerName}.`;
 }
-var STREAM_RETRY_DELAYS_MS;
+var STREAM_RETRY_DELAYS_MS, fastIgnoredNoticed;
 var init_composed_handler = __esm(() => {
   init_base_api_format();
   init_minimax_model_dialect();
@@ -58076,6 +58207,7 @@ var init_composed_handler = __esm(() => {
   init_raw();
   init_request_stage();
   init_logger();
+  init_codex_models_discovery();
   init_model_parser();
   init_provider_definitions();
   init_openai();
@@ -58114,6 +58246,7 @@ var init_composed_handler = __esm(() => {
   init_transient_retry();
   init_upstream_error_capture();
   STREAM_RETRY_DELAYS_MS = [3000, 15000, 30000];
+  fastIgnoredNoticed = new Set;
 });
 
 // src/providers/api-key-provenance.ts
@@ -71096,6 +71229,14 @@ function loadCustomEndpoints(config) {
   }
   return result;
 }
+function openaiModelsDiscovery(wireFormat, apiPath, modelPrefix) {
+  if (modelPrefix)
+    return;
+  const api = ["/chat/completions", "/responses"].find((suffix) => apiPath.endsWith(suffix));
+  if (wireFormat !== "openai" || !api)
+    return;
+  return { path: `${apiPath.slice(0, -api.length)}/models`, format: "openai-models-list" };
+}
 function buildProviderDefinition(name, ep, ovr) {
   const apiKeyEnvVar = ovr?.apiKeyEnvVar ?? customEndpointKeyEnvVar(name);
   const apiKeyAliases = ovr?.apiKeyAliases;
@@ -71120,9 +71261,11 @@ function buildProviderDefinition(name, ep, ovr) {
       isDirectApi: true,
       shortestPrefix: name,
       description: ovr?.description ?? `Custom endpoint: ${name}`,
-      authScheme: ep.authScheme ?? "bearer"
+      authScheme: ep.authScheme ?? "bearer",
+      modelDiscovery: openaiModelsDiscovery(ep.format, "/chat/completions", ep.modelPrefix)
     };
   }
+  const apiPath = ep.apiPath ?? "/v1/chat/completions";
   return {
     createHandler: runtimeHandler(name),
     tier: CUSTOM_ENDPOINT_TIER,
@@ -71131,7 +71274,7 @@ function buildProviderDefinition(name, ep, ovr) {
     transport: ep.transport,
     baseUrl: stripTrailingSlash(ep.baseUrl),
     baseUrlEnvVars,
-    apiPath: ep.apiPath ?? "/v1/chat/completions",
+    apiPath,
     apiKeyEnvVar,
     apiKeyAliases,
     apiKeyDescription: ovr?.apiKeyDescription ?? `${ep.displayName} (custom endpoint)`,
@@ -71142,7 +71285,8 @@ function buildProviderDefinition(name, ep, ovr) {
     shortestPrefix: name,
     description: ovr?.description ?? `Custom endpoint: ${ep.displayName}`,
     headers: ep.headers,
-    authScheme: ep.authScheme ?? "bearer"
+    authScheme: ep.authScheme ?? "bearer",
+    modelDiscovery: openaiModelsDiscovery(ep.transport, apiPath, ep.modelPrefix)
   };
 }
 function buildProviderProfile(ep, ovr) {
@@ -72000,7 +72144,7 @@ function isCacheFresh(data, ttlMs = CACHE_TTL_MS4) {
   const generatedMs = Date.parse(data.generatedAt);
   return !Number.isNaN(generatedMs) && Date.now() - generatedMs < ttlMs;
 }
-async function fetchProbeModels(url = PROBE_MODELS_URL, timeoutMs = FETCH_TIMEOUT_MS5) {
+async function fetchProbeModels(url = probeModelsUrl(), timeoutMs = FETCH_TIMEOUT_MS5) {
   let response;
   try {
     response = await fetch(url, {
@@ -72131,8 +72275,9 @@ function isValidResponse(raw) {
   const data = raw;
   return data.version === 3 && typeof data.generationId === "string" && data.generationId.length > 0 && typeof data.generatedAt === "string" && !!data.providers && typeof data.providers === "object" && !!data.unavailable && typeof data.unavailable === "object";
 }
-var PROBE_MODELS_URL = "https://us-central1-claudish-6da10.cloudfunctions.net/probeModels", CACHE_TTL_MS4, FETCH_TIMEOUT_MS5 = 15000, PROBE_MODELS_CACHE_PATH, _inFlight = null, DISCOVERY_ATTEMPTS = 3, DISCOVERY_TIMEOUT_MS = 8000;
+var CACHE_TTL_MS4, FETCH_TIMEOUT_MS5 = 15000, PROBE_MODELS_CACHE_PATH, _inFlight = null, DISCOVERY_ATTEMPTS = 3, DISCOVERY_TIMEOUT_MS = 8000;
 var init_probe_catalog = __esm(() => {
+  init_catalog_endpoints();
   init_catalog_route_bindings();
   CACHE_TTL_MS4 = 60 * 60 * 1000;
   PROBE_MODELS_CACHE_PATH = join33(homedir31(), ".claudish", "probe-models.json");
@@ -73517,6 +73662,46 @@ function claudeCodeKeychainServiceCandidates(env = process.env) {
 var BASE_KEYCHAIN_SERVICE = "Claude Code", KEYCHAIN_SERVICE_SUFFIX = "-credentials";
 var init_claude_code_facts = () => {};
 
+// src/fast-mode-gates.ts
+function applyProxyFastModeGates(env) {
+  scrubInheritedFastModeGates(env);
+  if (env.CLAUDE_CODE_DISABLE_FAST_MODE)
+    return;
+  const opened = [];
+  if (env.CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK === undefined) {
+    env.CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK = "1";
+    opened.push("skip");
+  }
+  const caps = env.CLAUDE_CODE_MODEL_CAPABILITIES;
+  env.CLAUDE_CODE_MODEL_CAPABILITIES = caps ? `${FAST_MODE_CAPABILITY};${caps}` : FAST_MODE_CAPABILITY;
+  opened.push("caps");
+  env[CLAUDISH_FAST_MODE_GATES_VAR] = opened.join(",");
+}
+function scrubInheritedFastModeGates(env) {
+  const marker = env[CLAUDISH_FAST_MODE_GATES_VAR];
+  if (marker === undefined)
+    return { removed: [] };
+  const opened = marker.split(",");
+  const removed = [];
+  if (opened.includes("skip") && env.CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK === "1") {
+    delete env.CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK;
+    removed.push("CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK");
+  }
+  const caps = env.CLAUDE_CODE_MODEL_CAPABILITIES;
+  if (opened.includes("caps") && caps !== undefined) {
+    if (caps === FAST_MODE_CAPABILITY) {
+      delete env.CLAUDE_CODE_MODEL_CAPABILITIES;
+      removed.push("CLAUDE_CODE_MODEL_CAPABILITIES");
+    } else if (caps.startsWith(`${FAST_MODE_CAPABILITY};`)) {
+      env.CLAUDE_CODE_MODEL_CAPABILITIES = caps.slice(FAST_MODE_CAPABILITY.length + 1);
+      removed.push("CLAUDE_CODE_MODEL_CAPABILITIES");
+    }
+  }
+  delete env[CLAUDISH_FAST_MODE_GATES_VAR];
+  return { removed };
+}
+var CLAUDISH_FAST_MODE_GATES_VAR = "CLAUDISH_FAST_MODE_GATES", FAST_MODE_CAPABILITY = "fast_mode";
+
 // src/hooks/session/binding.ts
 function sessionHooksNeedBinding(engine) {
   return BINDING_EVENTS.some((e) => engine.select(e).length > 0);
@@ -77309,6 +77494,34 @@ var init_no_surface = __esm(() => {
   init_binding();
 });
 
+// src/host-routing-env.ts
+function applyHostRouting(env) {
+  env[ENV.CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST] = "1";
+  env[ENV.CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL] = "";
+  env[ENV.CLAUDISH_SET_HOST_ROUTING_VARS] = "1";
+}
+function scrubInheritedHostRouting(env) {
+  if (env[ENV.CLAUDISH_SET_HOST_ROUTING_VARS] !== "1")
+    return { removed: [] };
+  const removed = [];
+  for (const name of HOST_ROUTING_VARS) {
+    if (name in env) {
+      delete env[name];
+      removed.push(name);
+    }
+  }
+  return { removed };
+}
+var HOST_ROUTING_VARS;
+var init_host_routing_env = __esm(() => {
+  init_config();
+  HOST_ROUTING_VARS = [
+    ENV.CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST,
+    ENV.CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL,
+    ENV.CLAUDISH_SET_HOST_ROUTING_VARS
+  ];
+});
+
 // src/launcher/magmux-binary.ts
 import { execSync } from "child_process";
 import { existsSync as existsSync28 } from "fs";
@@ -78112,7 +78325,8 @@ async function getRecommendedModels(opts = {}) {
     const generationId = readAllModelsCache()?.catalogGenerationId;
     if (!generationId)
       throw new Error("No complete v3 model and plan snapshot");
-    const response = await fetchCatalogData(FIREBASE_RECOMMENDED_URL, RECOMMENDED_FETCH_TIMEOUT_MS, generationId);
+    const base = modelsBaseUrl();
+    const response = await fetchCatalogData(`${base}${RECOMMENDED_VIEW}`, RECOMMENDED_FETCH_TIMEOUT_MS, generationId);
     if (response.mode !== "recommended")
       throw new Error("Unexpected recommendation mode");
     if (response.recommendations.generationId !== generationId) {
@@ -78123,7 +78337,7 @@ async function getRecommendedModels(opts = {}) {
     } catch (err) {
       log(`[effort] sidecar write failed for generation ${generationId}: ${err instanceof Error ? err.message : String(err)}`);
     }
-    const top100 = await fetchCatalogData(`${FIREBASE_BASE_URL}?catalog=top100`, RECOMMENDED_FETCH_TIMEOUT_MS, generationId);
+    const top100 = await fetchCatalogData(`${base}?catalog=top100`, RECOMMENDED_FETCH_TIMEOUT_MS, generationId);
     const data = recommendationProjection(response.recommendations, top100.models);
     if (data.models.length > 0) {
       _cachedRecommendedModels = data;
@@ -78167,27 +78381,27 @@ function isFreshEnough(doc) {
   return ageHours <= FIREBASE_CACHE_TTL_HOURS;
 }
 async function searchModels(query, limit = 50) {
-  const url = `${FIREBASE_BASE_URL}?search=${encodeURIComponent(query)}&limit=${limit}&status=active`;
+  const url = `${modelsBaseUrl()}?search=${encodeURIComponent(query)}&limit=${limit}&status=active`;
   const data = await fetchCatalogData(url, SEARCH_FETCH_TIMEOUT_MS);
   return data.models ?? [];
 }
 async function getModelByIdFromFirebase(modelId) {
-  const url = `${FIREBASE_BASE_URL}?modelId=${encodeURIComponent(modelId)}`;
+  const url = `${modelsBaseUrl()}?modelId=${encodeURIComponent(modelId)}`;
   const data = await fetchCatalogData(url, SEARCH_FETCH_TIMEOUT_MS);
   return data.mode === "exact" ? data.model : null;
 }
 async function getTop100Models() {
-  const url = `${FIREBASE_BASE_URL}?catalog=top100`;
+  const url = `${modelsBaseUrl()}?catalog=top100`;
   const data = await fetchCatalogData(url, SEARCH_FETCH_TIMEOUT_MS);
   return data;
 }
 async function getProviderList() {
-  const url = `${FIREBASE_BASE_URL}?catalog=providers`;
+  const url = `${modelsBaseUrl()}?catalog=providers`;
   const data = await fetchCatalogData(url, SEARCH_FETCH_TIMEOUT_MS);
   return data.providers ?? [];
 }
 async function getModelsByProvider(provider, pageSize = 200) {
-  const base = `${FIREBASE_BASE_URL}?provider=${encodeURIComponent(provider)}&status=active&limit=${pageSize}`;
+  const base = `${modelsBaseUrl()}?provider=${encodeURIComponent(provider)}&status=active&limit=${pageSize}`;
   return fetchPinnedModelDocs(base, "provider catalog");
 }
 async function fetchPinnedModelDocs(base, label) {
@@ -78225,7 +78439,7 @@ async function fetchPinnedModelDocs(base, label) {
   throw new Error(`${Label} exceeded the pagination limit`);
 }
 async function getAllModelDocs(pageSize = 200) {
-  const base = `${FIREBASE_BASE_URL}?status=active&limit=${pageSize}`;
+  const base = `${modelsBaseUrl()}?status=active&limit=${pageSize}`;
   return fetchPinnedModelDocs(base, "model catalog");
 }
 function loadModelInfo() {
@@ -78261,7 +78475,7 @@ function getAvailableModels() {
   _cachedModelIds = result;
   return result;
 }
-var RECOMMENDATION_SECTIONS, RECOMMENDATION_TIER_INDEX, _cachedModelInfo = null, _cachedModelIds = null, _cachedRecommendedModels = null, FIREBASE_BASE_URL = "https://us-central1-claudish-6da10.cloudfunctions.net/queryModels", FIREBASE_RECOMMENDED_URL, RECOMMENDED_PROJECTION_VERSION = 2, RECOMMENDED_MODELS_CACHE_PATH, RECOMMENDED_FETCH_TIMEOUT_MS = 5000, SEARCH_FETCH_TIMEOUT_MS = 1e4, RECOMMENDED_ROUTE_TIER_ORDER;
+var RECOMMENDATION_SECTIONS, RECOMMENDATION_TIER_INDEX, _cachedModelInfo = null, _cachedModelIds = null, _cachedRecommendedModels = null, RECOMMENDED_VIEW = "?catalog=recommended", RECOMMENDED_PROJECTION_VERSION = 2, RECOMMENDED_MODELS_CACHE_PATH, RECOMMENDED_FETCH_TIMEOUT_MS = 5000, SEARCH_FETCH_TIMEOUT_MS = 1e4, RECOMMENDED_ROUTE_TIER_ORDER;
 var init_model_loader = __esm(() => {
   init_fs_atomic();
   init_logger();
@@ -78269,6 +78483,7 @@ var init_model_loader = __esm(() => {
   init_auto_route();
   init_cache_ttl();
   init_catalog_client();
+  init_catalog_endpoints();
   init_catalog_route_bindings();
   init_effort_recommendations();
   init_route_candidates();
@@ -78290,7 +78505,6 @@ var init_model_loader = __esm(() => {
     { recommendationTier: "subscription", title: "Subscription", quickPickEligible: true }
   ];
   RECOMMENDATION_TIER_INDEX = new Map(RECOMMENDATION_SECTIONS.map((section, index) => [section.recommendationTier, index]));
-  FIREBASE_RECOMMENDED_URL = `${FIREBASE_BASE_URL}?catalog=recommended`;
   RECOMMENDED_MODELS_CACHE_PATH = join41(homedir33(), ".claudish", "recommended-models-cache-v2.json");
   RECOMMENDED_ROUTE_TIER_ORDER = {
     native: 0,
@@ -96304,6 +96518,12 @@ function scrubInheritedClaudishPlaceholders(env) {
   }
   return { removed };
 }
+function removeInheritedHostRouting(env) {
+  const { removed } = scrubInheritedHostRouting(env);
+  if (removed.length > 0) {
+    log(`[claude-runner] Removed inherited host-routing variables: ${removed.join(", ")}`);
+  }
+}
 function isClaudishPlaceholderCredential(name, value) {
   if (name === "ANTHROPIC_API_KEY")
     return value === CLAUDISH_PLACEHOLDER_API_KEY;
@@ -97072,6 +97292,8 @@ async function runClaudeWithProxy(config, proxyUrl, onCleanup) {
   if (config.monitor || advisorNativeSession) {
     delete env.ANTHROPIC_API_KEY;
     delete env.ANTHROPIC_AUTH_TOKEN;
+    scrubInheritedFastModeGates(env);
+    removeInheritedHostRouting(env);
     if (modelId) {
       env[ENV.ANTHROPIC_MODEL] = modelId;
       env[ENV.ANTHROPIC_SMALL_FAST_MODEL] = modelId;
@@ -97086,15 +97308,22 @@ async function runClaudeWithProxy(config, proxyUrl, onCleanup) {
       if (scrubbed.removed.length > 0) {
         log(`[claude-runner] Removed inherited claudish placeholder credentials: ${scrubbed.removed.join(", ")}`);
       }
+      removeInheritedHostRouting(env);
       if (shouldHideIncidentalAnthropicKey(config, env)) {
         delete env.ANTHROPIC_API_KEY;
         hidAnthropicApiKey = true;
       }
+      const closedGates = scrubInheritedFastModeGates(env);
+      if (closedGates.removed.length > 0) {
+        log(`[claude-runner] Removed inherited claudish fast-mode gates: ${closedGates.removed.join(", ")}`);
+      }
     } else {
       if (classifierPassthroughEnabled(config)) {
         console.error("[claudish] classifier passthrough enabled but no Anthropic credentials detected \u2014 " + "classifier requests will fail to authenticate against api.anthropic.com. " + "Map a role to a native Claude model (e.g. --model-sonnet claude-sonnet-5) or set ANTHROPIC_API_KEY.");
       }
       applyProxyPlaceholderAuth(env);
+      applyProxyFastModeGates(env);
+      applyHostRouting(env);
       const realWindow = await computeMainThreadContextWindow(config);
       const contextEnv = resolveContextWindowEnv(realWindow, process.env);
       Object.assign(env, contextEnv.vars);
@@ -97429,6 +97658,7 @@ var init_claude_runner = __esm(() => {
   init_interactive_driver();
   init_interactive_launch();
   init_no_surface();
+  init_host_routing_env();
   init_magmux_wrapper();
   init_logger();
   init_child_env();
@@ -106235,6 +106465,8 @@ function claudeCodePrintEnv(parent) {
   delete env.ANTHROPIC_API_KEY;
   delete env.ANTHROPIC_AUTH_TOKEN;
   delete env.ANTHROPIC_BASE_URL;
+  scrubInheritedFastModeGates(env);
+  scrubInheritedHostRouting(env);
   return env;
 }
 function isFrame(v) {
@@ -106425,6 +106657,7 @@ function startError(command, e) {
 var PRINT_FLAGS, PRINT_CHILD_DROPPED_MARKERS, inFlight4, KILL_GRACE_MS = 5000, STDERR_TAIL_CHARS = 2000;
 var init_run_prompt_claude_code = __esm(() => {
   init_config();
+  init_host_routing_env();
   init_child_env();
   init_pane_launch();
   init_redact();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudish",
-  "version": "14.16.2",
+  "version": "14.17.0",
   "description": "Run Claude Code with any model - OpenRouter, Ollama, LM Studio & local models",
   "type": "module",
   "main": "./dist/index.js",
@@ -71,10 +71,10 @@
     "ai"
   ],
   "optionalDependencies": {
-    "@claudish/magmux-darwin-arm64": "14.16.2",
-    "@claudish/magmux-darwin-x64": "14.16.2",
-    "@claudish/magmux-linux-arm64": "14.16.2",
-    "@claudish/magmux-linux-x64": "14.16.2"
+    "@claudish/magmux-darwin-arm64": "14.17.0",
+    "@claudish/magmux-darwin-x64": "14.17.0",
+    "@claudish/magmux-linux-arm64": "14.17.0",
+    "@claudish/magmux-linux-x64": "14.17.0"
   },
   "author": "Jack Rudenko <i@madappgang.com>",
   "license": "MIT",