@iowarp/clio-coder 0.4.2 → 0.4.3

package/dist/{chunk-BYMNWQ7O.js → chunk-XPWWI35G.js}

@@ -3,29 +3,26 @@ import {
   createSentinelStripper,
   stripTokenizerSentinels
 } from "./chunk-CFGTUFWB.js";
-import {
-  credentialsPresent
-} from "./chunk-72GZI5EV.js";
 import {
   ceilChars,
   estimateAgentMessageTokens,
   toolSchemaChars
-} from "./chunk-CWVRRIEI.js";
+} from "./chunk-NZMNUPZZ.js";
 import {
-  getSharedBus
-} from "./chunk-PMZVFWSR.js";
+  credentialsPresent
+} from "./chunk-2JH2WHGE.js";
 import {
   readSettings
-} from "./chunk-34BHNEE3.js";
+} from "./chunk-GHS5EBTQ.js";
 import {
   warnLegacyNaming
 } from "./chunk-D7E7WCDF.js";
+import {
+  getSharedBus
+} from "./chunk-PMZVFWSR.js";
 import {
   BusChannels
 } from "./chunk-F4EKGO4N.js";
-import {
-  sleep
-} from "./chunk-EBY63B7G.js";
 import {
   CLIO_CONTEXT_WINDOW_WARN_BELOW,
   CLIO_MIN_CONTEXT_WINDOW,
@@ -42,31 +39,26 @@ import {
   registerBuiltinRuntimes,
   requestLmStudioJson,
   resolveLmStudioInstance
-} from "./chunk-IMXMHHMQ.js";
+} from "./chunk-VW6DOEDG.js";
+import {
+  sleep
+} from "./chunk-EBY63B7G.js";
 import {
   listKnownModelsForRuntime
-} from "./chunk-CDNVLKUX.js";
+} from "./chunk-LLDJM5XK.js";
 import {
   authNotRequiredStatus,
   openAuthStorage,
   registerClioOAuthProviders,
   resolveAuthTarget,
   targetRequiresAuth
-} from "./chunk-WNP7O5WZ.js";
+} from "./chunk-ID64D7PE.js";
 import {
   withStateFileLock
 } from "./chunk-IKCO5N3L.js";
 import {
   require_dist
 } from "./chunk-APJ265NV.js";
-import {
-  clioCacheDir,
-  clioConfigDir,
-  clioStateDir,
-  clioStatePath,
-  resolveClioDirs,
-  stateRootRemoved
-} from "./chunk-BYP5D4HI.js";
 import {
   safeResourceWrite
 } from "./chunk-WEPFGWHJ.js";
@@ -78,13 +70,21 @@ import {
   getCatalogModelForRuntime,
   mergeCapabilities,
   resolveCostProvenance
-} from "./chunk-O42A54GG.js";
+} from "./chunk-5OIVVPHF.js";
 import {
   activateExternalPluginApiBridge,
   calculateEngineCost,
   ensurePiAiRegistered,
   registerEngineApiProvider
-} from "./chunk-VXMFAE2W.js";
+} from "./chunk-YPC6ZR5L.js";
+import {
+  clioCacheDir,
+  clioConfigDir,
+  clioStateDir,
+  clioStatePath,
+  resolveClioDirs,
+  stateRootRemoved
+} from "./chunk-BYP5D4HI.js";
 import {
   __toESM,
   init_esm_shims
@@ -797,6 +797,12 @@ function applyModelCapabilityPatch(model, caps) {
   if (typeof caps.contextWindow === "number") model.contextWindow = caps.contextWindow;
   if (typeof caps.maxTokens === "number") model.maxTokens = caps.maxTokens;
   if (typeof caps.reasoning === "boolean") model.reasoning = caps.reasoning;
+  if (model.clioCoder || caps.thinkingControlRuntime !== void 0) {
+    const metadata2 = { ...model.clioCoder };
+    delete metadata2.thinkingControlRuntime;
+    if (caps.thinkingControlRuntime !== void 0) metadata2.thinkingControlRuntime = caps.thinkingControlRuntime;
+    model.clioCoder = metadata2;
+  }
   return model;
 }
 function applyReasoningResolution(caps, kbHit, detectedReasoning) {
@@ -1068,7 +1074,8 @@ function effectiveThinkingLevel(configured, available) {
   const fallback = available[0] ?? "off";
   if (!configured) return fallback;
   if (available.includes(configured)) return configured;
-  if (configured === "max" && available.includes("xhigh")) return "xhigh";
+  if ((configured === "max" || configured === "high" && !available.includes("high")) && available.includes("xhigh"))
+    return "xhigh";
   if ((configured === "high" || configured === "xhigh" || configured === "max") && available.includes("high")) {
     return "high";
   }
@@ -1263,7 +1270,12 @@ function resolveModelRuntimeCapabilities(input) {
     family,
     capabilities: input.capabilities,
     thinking,
-    request: resolveRequestCapability(thinking, parser, input.runtimeId, quirks),
+    request: resolveRequestCapability(
+      thinking,
+      parser,
+      input.runtimeId === "litellm" ? input.capabilities.thinkingControlRuntime ?? input.runtimeId : input.runtimeId,
+      quirks
+    ),
     response: {
       parser,
       stripTokenizerSentinels: true
@@ -1329,6 +1341,7 @@ function thinkingFormatFromModelApi(api) {
 }
 function capabilitiesFromModel(model) {
   const format = model.compat?.thinkingFormat ?? thinkingFormatFromModelApi(model.api);
+  const controlRuntime = (model.clioCoder ?? model.clio)?.thinkingControlRuntime;
   const caps = {
     chat: true,
     tools: true,
@@ -1339,7 +1352,8 @@ function capabilitiesFromModel(model) {
     rerank: false,
     fim: false,
     contextWindow: model.contextWindow,
-    maxTokens: model.maxTokens
+    maxTokens: model.maxTokens,
+    ...controlRuntime ? { thinkingControlRuntime: controlRuntime } : {}
   };
   if (format === "qwen-chat-template" || format === "openrouter" || format === "zai" || format === "anthropic-extended" || format === "deepseek-r1" || format === "openai-codex" || format === "harmony") {
     caps.thinkingFormat = format;
@@ -1411,6 +1425,21 @@ function normalizedModels(models) {
   }
   return out;
 }
+function normalizedLabels(models, labels) {
+  if (typeof labels !== "object" || labels === null || Array.isArray(labels)) return void 0;
+  const source = labels;
+  const out = {};
+  for (const id of models) {
+    const label = source[id];
+    if (typeof label !== "string") continue;
+    const trimmed = label.trim();
+    if (trimmed.length === 0 || trimmed.length > TARGET_MODEL_CACHE_MAX_MODEL_ID_CHARS || /[\r\n\0]/u.test(trimmed)) {
+      continue;
+    }
+    out[id] = trimmed;
+  }
+  return Object.keys(out).length > 0 ? out : void 0;
+}
 function parseSnapshot(value) {
   if (typeof value !== "object" || value === null || Array.isArray(value)) return null;
   const row = value;
@@ -1422,7 +1451,7 @@ function parseSnapshot(value) {
   if (typeof row.observedAt !== "string" || !Number.isFinite(Date.parse(row.observedAt))) return null;
   const models = normalizedModels(row.models);
   if (models.length !== row.models.length) return null;
-  return {
+  const snapshot = {
     version: TARGET_MODEL_CACHE_VERSION,
     targetId: row.targetId,
     runtimeId: row.runtimeId,
@@ -1430,6 +1459,9 @@ function parseSnapshot(value) {
     models,
     observedAt: row.observedAt
   };
+  const modelLabels = normalizedLabels(models, row.modelLabels);
+  if (modelLabels !== void 0) snapshot.modelLabels = modelLabels;
+  return snapshot;
 }
 function readTargetModelSnapshot(target, options = {}) {
   const path = targetModelSnapshotPath(target.id, options.cacheDir);
@@ -1458,6 +1490,8 @@ function recordTargetModelSnapshot(target, models, options = {}) {
     models: normalizedModels(models),
     observedAt: new Date(options.nowMs ?? Date.now()).toISOString()
   };
+  const modelLabels = normalizedLabels(snapshot.models, options.modelLabels);
+  if (modelLabels !== void 0) snapshot.modelLabels = modelLabels;
   try {
     safeResourceWrite(targetModelSnapshotPath(target.id, options.cacheDir), `${JSON.stringify(snapshot, null, 2)}
 `, {
@@ -1531,6 +1565,7 @@ function modelCandidatesForStatus(status) {
     const state = status.discoveredModelStates?.[trimmed];
     out.push({
       id: trimmed,
+      ...status.discoveredModelLabels?.[trimmed] ? { label: status.discoveredModelLabels[trimmed] } : {},
       source,
       ...state ? { loadState: state.state } : {},
       ...state?.detail ? { loadStateDetail: state.detail } : {}
@@ -2796,6 +2831,115 @@ import {
 } from "@earendil-works/pi-ai";
 import { openAICompletionsApi } from "@earendil-works/pi-ai/api/openai-completions.lazy";
 
+// src/core/gateway-routing.ts
+init_esm_shims();
+function liteLLMRouteFailureMessage(errorMessage, targetId, modelId) {
+  const providerMessage = errorMessage?.trim() || "LiteLLM request failed.";
+  const advice = `LiteLLM route '${modelId}' on target '${targetId}' failed. Clio did not retry or substitute another model; select a different route with /model, then resend.`;
+  return providerMessage.includes(advice) ? providerMessage : `${providerMessage}
+
+${advice}`;
+}
+function nonEmptyString(value) {
+  return typeof value === "string" && value.trim().length > 0 ? value.trim() : void 0;
+}
+function nonnegativeNumber(value) {
+  if (typeof value === "number" && Number.isFinite(value) && value >= 0) return value;
+  if (typeof value !== "string" || value.trim().length === 0) return void 0;
+  const parsed = Number(value);
+  return Number.isFinite(parsed) && parsed >= 0 ? parsed : void 0;
+}
+function nonnegativeInteger(value) {
+  const parsed = nonnegativeNumber(value);
+  return parsed === void 0 ? void 0 : Math.floor(parsed);
+}
+function apiBaseHost(value) {
+  const raw = nonEmptyString(value);
+  if (raw === void 0) return void 0;
+  try {
+    return new URL(raw).host || void 0;
+  } catch {
+    return void 0;
+  }
+}
+function storedApiBaseHost(value) {
+  const raw = nonEmptyString(value);
+  if (raw === void 0) return void 0;
+  try {
+    return new URL(raw.includes("://") ? raw : `http://${raw}`).host || void 0;
+  } catch {
+    return void 0;
+  }
+}
+function liteLLMGatewayRoutingFromHeaders(headers) {
+  const callId = nonEmptyString(headers.get("x-litellm-call-id"));
+  const modelGroup = nonEmptyString(headers.get("x-litellm-model-group"));
+  const modelName = nonEmptyString(headers.get("x-litellm-model-name"));
+  const host = apiBaseHost(headers.get("x-litellm-model-api-base"));
+  const deploymentId = nonEmptyString(headers.get("x-litellm-model-id"));
+  const attemptedFallbacks = nonnegativeInteger(headers.get("x-litellm-attempted-fallbacks"));
+  const attemptedRetries = nonnegativeInteger(headers.get("x-litellm-attempted-retries"));
+  const overheadMs = nonnegativeNumber(headers.get("x-litellm-overhead-duration-ms"));
+  const responseDurationMs = nonnegativeNumber(headers.get("x-litellm-response-duration-ms"));
+  const version = nonEmptyString(headers.get("x-litellm-version"));
+  const observation = {
+    gateway: "litellm",
+    ...callId !== void 0 ? { callId } : {},
+    ...modelGroup !== void 0 ? { modelGroup } : {},
+    ...modelName !== void 0 ? { modelName } : {},
+    ...host !== void 0 ? { apiBaseHost: host } : {},
+    ...deploymentId !== void 0 ? { deploymentId } : {},
+    ...attemptedFallbacks !== void 0 ? { attemptedFallbacks } : {},
+    ...attemptedRetries !== void 0 ? { attemptedRetries } : {},
+    ...overheadMs !== void 0 ? { overheadMs } : {},
+    ...responseDurationMs !== void 0 ? { responseDurationMs } : {},
+    ...version !== void 0 ? { version } : {}
+  };
+  return Object.keys(observation).length > 1 ? observation : null;
+}
+function gatewayRoutingObservationFromRecord(record) {
+  const value = record.gatewayRouting;
+  if (value === null || typeof value !== "object" || Array.isArray(value)) return null;
+  const raw = value;
+  if (raw.gateway !== "litellm") return null;
+  const callId = nonEmptyString(raw.callId);
+  const modelGroup = nonEmptyString(raw.modelGroup);
+  const modelName = nonEmptyString(raw.modelName);
+  const host = storedApiBaseHost(raw.apiBaseHost);
+  const deploymentId = nonEmptyString(raw.deploymentId);
+  const attemptedFallbacks = nonnegativeInteger(raw.attemptedFallbacks);
+  const attemptedRetries = nonnegativeInteger(raw.attemptedRetries);
+  const overheadMs = nonnegativeNumber(raw.overheadMs);
+  const responseDurationMs = nonnegativeNumber(raw.responseDurationMs);
+  const version = nonEmptyString(raw.version);
+  const observation = {
+    gateway: "litellm",
+    ...callId !== void 0 ? { callId } : {},
+    ...modelGroup !== void 0 ? { modelGroup } : {},
+    ...modelName !== void 0 ? { modelName } : {},
+    ...host !== void 0 ? { apiBaseHost: host } : {},
+    ...deploymentId !== void 0 ? { deploymentId } : {},
+    ...attemptedFallbacks !== void 0 ? { attemptedFallbacks } : {},
+    ...attemptedRetries !== void 0 ? { attemptedRetries } : {},
+    ...overheadMs !== void 0 ? { overheadMs } : {},
+    ...responseDurationMs !== void 0 ? { responseDurationMs } : {},
+    ...version !== void 0 ? { version } : {}
+  };
+  return Object.keys(observation).length > 1 ? observation : null;
+}
+function gatewayRoutingObservationLabel(observation) {
+  const route = [observation.modelGroup, observation.modelName].filter(Boolean).join("->");
+  const destination = [route, observation.apiBaseHost ? `@${observation.apiBaseHost}` : ""].filter(Boolean).join("");
+  const parts = [
+    "gateway=litellm",
+    ...destination ? [`route=${destination}`] : [],
+    ...observation.attemptedFallbacks !== void 0 ? [`fallbacks=${observation.attemptedFallbacks}`] : [],
+    ...observation.attemptedRetries !== void 0 ? [`retries=${observation.attemptedRetries}`] : [],
+    ...observation.overheadMs !== void 0 ? [`overhead_ms=${observation.overheadMs}`] : []
+  ];
+  return parts.join(" ");
+}
+
 // src/engine/harmony-response.ts
 init_esm_shims();
 var START = "<|start|>";
@@ -3443,7 +3587,10 @@ function observeResponseModelIdBytes(chunk, capture, flush = false) {
   }
   if (captureCanStopEarly(capture) || flush) capture.decoder = null;
 }
-function captureResponseModelId(response, capture) {
+function captureResponseModelId(response, capture, model) {
+  if (runtimeMetadata(model)?.runtimeId === "litellm") {
+    capture.gatewayRouting = liteLLMGatewayRoutingFromHeaders(response.headers);
+  }
   if (!response.body || !response.headers.get("content-type")?.toLowerCase().includes("text/event-stream")) {
     return response;
   }
@@ -3473,13 +3620,14 @@ function withResponseModelIdCapture(model, options, sourceFactory) {
     modelIdDone: false,
     backendTimings: null,
     backendTimingsSource: backendTimingsSourceForModel(model),
+    gatewayRouting: null,
     buffer: "",
     decoder: new TextDecoder()
   };
   const fetchImpl = options.fetch ?? ((input, init) => globalThis.fetch(input, init));
   const capturedOptions = {
     ...options,
-    fetch: async (input, init) => captureResponseModelId(await fetchImpl(input, init), capture)
+    fetch: async (input, init) => captureResponseModelId(await fetchImpl(input, init), capture, model)
   };
   const source = sourceFactory(capturedOptions);
   const annotated = createAssistantMessageEventStream3();
@@ -3489,9 +3637,11 @@ function withResponseModelIdCapture(model, options, sourceFactory) {
         const observation = capture.observed ? capture.reportedModelId === null ? { state: "not-reported" } : { state: "reported", reportedModelId: capture.reportedModelId } : { state: "not-observed" };
         if (event.type === "done") {
           event.message.responseModelIdObservation = observation;
+          if (capture.gatewayRouting !== null) event.message.gatewayRouting = capture.gatewayRouting;
           if (capture.backendTimings !== null) event.message.backendTimings = capture.backendTimings;
         } else if (event.type === "error") {
           event.error.responseModelIdObservation = observation;
+          if (capture.gatewayRouting !== null) event.error.gatewayRouting = capture.gatewayRouting;
           if (capture.backendTimings !== null) event.error.backendTimings = capture.backendTimings;
         }
         annotated.push(event);
@@ -3504,6 +3654,26 @@ function withResponseModelIdCapture(model, options, sourceFactory) {
   })();
   return annotated;
 }
+function withLiteLLMRouteFailureAdvice(model, source) {
+  const metadata2 = runtimeMetadata(model);
+  if (metadata2?.runtimeId !== "litellm") return source;
+  const advised = createAssistantMessageEventStream3();
+  (async () => {
+    try {
+      for await (const event of source) {
+        if (event.type === "error") {
+          event.error.errorMessage = liteLLMRouteFailureMessage(event.error.errorMessage, metadata2.targetId, model.id);
+        }
+        advised.push(event);
+      }
+      advised.end();
+    } catch (err) {
+      const message = err instanceof Error ? err.message : String(err);
+      throw new Error(liteLLMRouteFailureMessage(message, metadata2.targetId, model.id));
+    }
+  })();
+  return advised;
+}
 function samplingParamsFromProfile(profile) {
   return {
     ...profile.topP !== void 0 ? { top_p: profile.topP } : {},
@@ -3591,6 +3761,10 @@ function applyThinkingPayload(payload, applied, resolved, model) {
   if (resolved.request.reasoningEffort && (next.reasoning_effort === void 0 || resolved.response.parser === "harmony")) {
     next.reasoning_effort = resolved.request.reasoningEffort;
   }
+  if (resolved.runtimeId === "litellm" && resolved.request.reasoningEffort && next.reasoning_effort === resolved.request.reasoningEffort) {
+    const allowed = Array.isArray(next.allowed_openai_params) ? next.allowed_openai_params : [];
+    next.allowed_openai_params = [.../* @__PURE__ */ new Set([...allowed, "reasoning_effort"])];
+  }
   if (resolved.request.chatTemplateKwargs && !chatTemplateKwargsUnsupported(model)) {
     const existing = isPlainRecord(next.chat_template_kwargs) ? next.chat_template_kwargs : {};
     next.chat_template_kwargs = { ...existing, ...resolved.request.chatTemplateKwargs };
@@ -3612,13 +3786,13 @@ function applyLmStudioPayload(payload, model, resolved) {
   if (request?.ttlSeconds !== void 0) next.ttl = request.ttlSeconds;
   if (request?.draftModel !== void 0) next.draft_model = request.draftModel;
   if (resolved.thinking.mechanism === "none" || resolved.thinking.mechanism === "always-on") return next;
-  const resolvedEffort = resolved.request.reasoningEffort ?? lmStudioReasoningEffort(resolved.thinking.effectiveLevel, runtimeMetadata(model)?.lmstudioReasoningOptions);
+  const resolvedEffort = runtimeMetadata(model)?.lmstudioReasoningOptions ? lmStudioReasoningEffort(resolved.thinking.effectiveLevel, runtimeMetadata(model)?.lmstudioReasoningOptions) : resolved.request.reasoningEffort ?? lmStudioReasoningEffort(resolved.thinking.effectiveLevel);
   switch (request?.reasoning) {
     case "off":
-      next.reasoning_effort = "none";
+      next.reasoning_effort = lmStudioReasoningEffort("off", runtimeMetadata(model)?.lmstudioReasoningOptions);
       break;
     case "on":
-      next.reasoning_effort = "low";
+      next.reasoning_effort = lmStudioReasoningEffort("low", runtimeMetadata(model)?.lmstudioReasoningOptions);
       break;
     case "low":
     case "medium":
@@ -3722,6 +3896,36 @@ function withRemainingContextBudget(model, context, options) {
     maxTokens: remainingContextMaxTokens(model, context, options)
   };
 }
+function hasHeader(headers, name) {
+  if (!headers) return false;
+  const wanted = name.toLowerCase();
+  return Object.keys(headers).some((key) => key.toLowerCase() === wanted);
+}
+function withLiteLLMRequestOptions(model, options) {
+  const metadata2 = runtimeMetadata(model);
+  if (metadata2?.runtimeId !== "litellm") return options;
+  const request = metadata2.litellm?.request;
+  const modelHeaders = model.headers;
+  const optionHeaders = options.headers;
+  const headers = { ...options.headers ?? {} };
+  const setDefault = (name, value) => {
+    if (value === void 0 || hasHeader(modelHeaders, name) || hasHeader(optionHeaders, name)) return;
+    headers[name] = value;
+  };
+  setDefault("x-litellm-tags", ["clio-coder", ...request?.tags ?? []].join(","));
+  if (request?.sendSessionId !== false) setDefault("x-litellm-session-id", options.sessionId);
+  setDefault("x-litellm-timeout", request?.timeoutSeconds?.toString());
+  setDefault("x-litellm-stream-timeout", request?.streamTimeoutSeconds?.toString());
+  setDefault("x-litellm-num-retries", request?.numRetries?.toString());
+  return {
+    ...options,
+    headers,
+    // Never hide duplicate attempts inside the OpenAI SDK beneath Clio's
+    // operator-visible failure. The gateway may still be configured to retry,
+    // but physical-routing deployments should keep that policy at zero too.
+    maxRetries: options.maxRetries ?? 0
+  };
+}
 function requiredToolArguments(tool) {
   const schema = tool.parameters;
   if (schema === null || typeof schema !== "object" || Array.isArray(schema)) return [];
@@ -3990,21 +4194,28 @@ var openAICompletionsApiProvider = {
     const resolved = resolvedCapabilitiesForModel(model, defaultLevel);
     const effectiveContext = stripsThinking(resolved) ? stripThinkingFromContext(context) : context;
     const withSamplers = withSamplingOverrides(model, options, resolved);
+    const withGatewayOptions = withLiteLLMRequestOptions(
+      model,
+      withRemainingContextBudget(model, effectiveContext, withSamplers)
+    );
     return guardMalformedToolCalls(
       withReasoningTokenEstimate(
         stripSentinelsFromStream(
           stripNeverReasoningFromStream(
             filterGemmaChannelStream(
-              withResponseModelIdCapture(
+              withLiteLLMRouteFailureAdvice(
                 model,
-                withRemainingContextBudget(model, effectiveContext, withSamplers),
-                (capturedOptions) => withLocalResidency(
+                withResponseModelIdCapture(
                   model,
-                  {
-                    ...options?.apiKey !== void 0 ? { apiKey: options.apiKey } : {},
-                    ...options?.signal !== void 0 ? { signal: options.signal } : {}
-                  },
-                  (requestModel) => piOpenAICompletions.stream(requestModel, effectiveContext, capturedOptions)
+                  withGatewayOptions,
+                  (capturedOptions) => withLocalResidency(
+                    model,
+                    {
+                      ...options?.apiKey !== void 0 ? { apiKey: options.apiKey } : {},
+                      ...options?.signal !== void 0 ? { signal: options.signal } : {}
+                    },
+                    (requestModel) => piOpenAICompletions.stream(requestModel, effectiveContext, capturedOptions)
+                  )
                 )
               ),
               usesGemmaChannelMarkers(resolved.modelId)
@@ -4022,21 +4233,28 @@ var openAICompletionsApiProvider = {
     const resolved = resolvedCapabilitiesForModel(model, thinkingLevelFromSimple2(options));
     const effectiveContext = stripsThinking(resolved) ? stripThinkingFromContext(context) : context;
     const withSamplers = withSamplingOverrides(model, options, resolved);
+    const withGatewayOptions = withLiteLLMRequestOptions(
+      model,
+      withRemainingContextBudget(model, effectiveContext, withSamplers)
+    );
     return guardMalformedToolCalls(
       withReasoningTokenEstimate(
         stripSentinelsFromStream(
           stripNeverReasoningFromStream(
             filterGemmaChannelStream(
-              withResponseModelIdCapture(
+              withLiteLLMRouteFailureAdvice(
                 model,
-                withRemainingContextBudget(model, effectiveContext, withSamplers),
-                (capturedOptions) => withLocalResidency(
+                withResponseModelIdCapture(
                   model,
-                  {
-                    ...options?.apiKey !== void 0 ? { apiKey: options.apiKey } : {},
-                    ...options?.signal !== void 0 ? { signal: options.signal } : {}
-                  },
-                  (requestModel) => piOpenAICompletions.streamSimple(requestModel, effectiveContext, capturedOptions)
+                  withGatewayOptions,
+                  (capturedOptions) => withLocalResidency(
+                    model,
+                    {
+                      ...options?.apiKey !== void 0 ? { apiKey: options.apiKey } : {},
+                      ...options?.signal !== void 0 ? { signal: options.signal } : {}
+                    },
+                    (requestModel) => piOpenAICompletions.streamSimple(requestModel, effectiveContext, capturedOptions)
+                  )
                 )
               ),
               usesGemmaChannelMarkers(resolved.modelId)
@@ -4376,7 +4594,7 @@ function discoveredModelsSource(probe, preservePreviousProbe, previous, cachedMo
 function sameProbeIdentity(previous, next) {
   return previous.id === next.id && previous.runtime === next.runtime && (previous.url ?? "") === (next.url ?? "") && (previous.defaultModel ?? "") === (next.defaultModel ?? "");
 }
-function mergeProbeResult(desc, target, probe, previous, cachedModels = []) {
+function mergeProbeResult(desc, target, probe, previous, cachedModels = [], cachedModelLabels = {}) {
   const probeSucceeded = probe?.ok ?? false;
   const preservePrevious = !probeSucceeded && previous !== void 0 && sameProbeIdentity(previous.target, target);
   const probeCapabilities = probe?.discoveredCapabilities ?? (preservePrevious ? previous.probeCapabilities : null) ?? null;
@@ -4388,8 +4606,10 @@ function mergeProbeResult(desc, target, probe, previous, cachedModels = []) {
     probe?.models ?? (preservePrevious ? previous.discoveredModels : void 0) ?? (cachedModels.length > 0 ? cachedModels : void 0) ?? desc.knownModels ?? []
   );
   const discoveredModelStates = probe?.modelStates ?? (preservePrevious ? previous.discoveredModelStates : null) ?? null;
+  const discoveredModelLabels = probe?.modelLabels ?? (preservePrevious ? previous.discoveredModelLabels : void 0) ?? (cachedModels.length > 0 ? cachedModelLabels : void 0) ?? {};
   const merge = {
     discoveredModels,
+    discoveredModelLabels,
     discoveredModelsSource: discoveredModelsSource(probe, preservePrevious, previous, cachedModels, desc),
     discoveredModelStates,
     probeCapabilities,
@@ -4404,7 +4624,9 @@ function unservedDefaultModelReason(desc, target, merge) {
   const model = target.defaultModel;
   if (!model) return null;
   if (merge.discoveredModelsSource !== "probe" || merge.discoveredModels.length === 0) return null;
-  if (listKnownModelsForRuntime(desc.id).length > 0) return null;
+  if (listKnownModelsForRuntime(desc.id).length > 0 && desc.externalAgentLoop?.modelCatalog !== "live-authoritative") {
+    return null;
+  }
   if (merge.discoveredModels.includes(model)) return null;
   if (merge.discoveredModelStates && model in merge.discoveredModelStates) return null;
   return `default model '${model}' is not advertised by the target`;
@@ -4424,14 +4646,18 @@ function createProvidersBundle(context) {
     if (config) return config.get();
     return readSettings();
   }
-  async function buildProbeContextForTarget(target, desc) {
+  async function buildProbeContextForTarget(target, desc, signal) {
     const probeCtx = {
       credentialsPresent: credentialsPresent(),
-      httpTimeoutMs: DEFAULT_PROBE_TIMEOUT_MS
+      httpTimeoutMs: DEFAULT_PROBE_TIMEOUT_MS,
+      ...signal ? { signal } : {}
     };
     if (!targetRequiresAuth(target, desc)) return probeCtx;
     try {
-      const resolution = await authStore.resolveForTarget(resolveAuthTarget(target, desc), { includeFallback: false });
+      const resolution = await authStore.resolveForTarget(resolveAuthTarget(target, desc), {
+        includeFallback: false,
+        ...signal ? { signal } : {}
+      });
       if (resolution.apiKey) probeCtx.authToken = resolution.apiKey;
     } catch {
     }
@@ -4483,8 +4709,9 @@ function createProvidersBundle(context) {
       return out2;
     }
     const availability = availabilityFor(desc, target, authStatusFor);
-    const cachedModels = readTargetModelSnapshot(target)?.models ?? [];
-    const merge = mergeProbeResult(desc, target, probe, previous, cachedModels);
+    const cachedSnapshot = readTargetModelSnapshot(target);
+    const cachedModels = cachedSnapshot?.models ?? [];
+    const merge = mergeProbeResult(desc, target, probe, previous, cachedModels, cachedSnapshot?.modelLabels ?? {});
     const { capabilities, contextWindowProvenance } = capabilitiesFor(desc, target, merge, kb);
     const healthy = probe !== null ? probe.ok : null;
     const unservedDefault = probe?.ok ? unservedDefaultModelReason(desc, target, merge) : null;
@@ -4508,6 +4735,7 @@ function createProvidersBundle(context) {
       probeModelCapabilities: merge.probeModelCapabilities,
       probeModelId: merge.probeModelId,
       discoveredModels: merge.discoveredModels,
+      discoveredModelLabels: merge.discoveredModelLabels,
       discoveredModelsSource: merge.discoveredModelsSource,
       discoveredModelStates: merge.discoveredModelStates
     };
@@ -4516,6 +4744,7 @@ function createProvidersBundle(context) {
     return out;
   }
   async function probeTargetInternal(target, live, options) {
+    options?.signal?.throwIfAborted();
     const previous = statuses.get(target.id);
     const desc = registry.get(target.runtime);
     if (!desc) {
@@ -4529,13 +4758,15 @@ function createProvidersBundle(context) {
       statuses.set(target.id, status2);
       return status2;
     }
-    const probeCtx = await buildProbeContextForTarget(target, desc);
+    const probeCtx = await buildProbeContextForTarget(target, desc, options?.signal);
+    options?.signal?.throwIfAborted();
     let probeResult;
     try {
       probeResult = await desc.probe(target, probeCtx);
     } catch (err) {
       probeResult = { ok: false, error: err instanceof Error ? err.message : String(err) };
     }
+    options?.signal?.throwIfAborted();
     if (probeResult.ok && typeof desc.probeModels === "function" && !probeResult.models) {
       try {
         const ids = await desc.probeModels(target, probeCtx);
@@ -4543,6 +4774,7 @@ function createProvidersBundle(context) {
       } catch {
       }
     }
+    options?.signal?.throwIfAborted();
     if (probeResult.ok && options?.reasoning !== false && typeof desc.probeReasoning === "function") {
       const settings = readConfig();
       const orchestratorTarget = settings.chat.target === target.id ? settings.chat.model : null;
@@ -4550,6 +4782,7 @@ function createProvidersBundle(context) {
       if (candidateModelId) {
         try {
           const result = await desc.probeReasoning(target, candidateModelId, probeCtx);
+          options?.signal?.throwIfAborted();
           reasoningCache.set(reasoningCacheKey(target.id, candidateModelId), result.reasoning);
           const capabilityModelId = probeResult.capabilityModelId ?? null;
           if (capabilityModelId === null || capabilityModelId === candidateModelId) {
@@ -4570,10 +4803,15 @@ function createProvidersBundle(context) {
         }
       }
     }
+    options?.signal?.throwIfAborted();
     const status = buildStatus(target, desc, probeResult, previous);
     statuses.set(target.id, status);
     if (probeResult.ok && probeResult.models !== void 0) {
-      recordTargetModelSnapshot(target, probeResult.models);
+      recordTargetModelSnapshot(
+        target,
+        probeResult.models,
+        probeResult.modelLabels ? { modelLabels: probeResult.modelLabels } : {}
+      );
     }
     void recordEndpointSlotsFromStatus(status);
     context.bus.emit(BusChannels.ProviderHealth, { id: target.id, status });
@@ -4759,6 +4997,8 @@ export {
   setResidencyNoticeSink,
   setProtectedModelsProvider,
   withRunOverrides,
+  gatewayRoutingObservationFromRecord,
+  gatewayRoutingObservationLabel,
   applyModelCapabilityPatch,
   probeCapabilitiesForModel,
   resolveModelCapabilities,
@@ -4776,6 +5016,7 @@ export {
   isOrchestratorEligibleRuntime,
   isDispatchEligibleRuntime,
   modelResidencyForStatus,
+  hasLiveModelCatalog,
   modelCandidatesForStatus,
   modelIdsForStatus,
   canonicalizeWireModelId,
@@ -4790,4 +5031,4 @@ export {
   resolveCostProvenance2 as resolveCostProvenance,
   ProvidersDomainModule
 };
-//# sourceMappingURL=chunk-BYMNWQ7O.js.map
+//# sourceMappingURL=chunk-XPWWI35G.js.map