@iowarp/clio-coder 0.3.7 → 0.3.9

package/dist/{chunk-VQNODYQ4.js → chunk-MIX5N5AC.js}

@@ -4,38 +4,50 @@ import {
   stripTokenizerSentinels
 } from "./chunk-CFGTUFWB.js";
 import {
-  authNotRequiredStatus,
-  openAuthStorage,
-  registerClioOAuthProviders,
-  resolveAuthTarget,
-  resolveRuntimeAuthTarget,
-  targetRequiresAuth
-} from "./chunk-JRIO5UD2.js";
+  credentialsPresent
+} from "./chunk-LU7P4LHA.js";
 import {
   ceilChars,
   estimateAgentMessageTokens,
   toolSchemaChars
-} from "./chunk-GEYXPTRF.js";
+} from "./chunk-AEYBF3TB.js";
 import {
   readSettings
-} from "./chunk-UHXRNZ2J.js";
-import {
-  THINKING_LEVELS
-} from "./chunk-4DGYLA73.js";
+} from "./chunk-PNY46YEY.js";
 import {
   sleep
 } from "./chunk-FQ4SKYE4.js";
+import {
+  getSharedBus
+} from "./chunk-ZGVHUX3M.js";
+import {
+  BusChannels
+} from "./chunk-RKRLDWD3.js";
+import {
+  CLIO_CONTEXT_WINDOW_WARN_BELOW,
+  CLIO_MIN_CONTEXT_WINDOW,
+  CLIO_MIN_MAX_OUTPUT_TOKENS,
+  extractLocalModelQuirks,
+  listLmStudioModels,
+  lmStudioReasoningEffort,
+  lmStudioRootUrl,
+  registerBuiltinRuntimes,
+  requestLmStudioJson,
+  resolveLmStudioInstance
+} from "./chunk-JEQQR47K.js";
+import {
+  listKnownModelsForRuntime
+} from "./chunk-774ILSRL.js";
+import {
+  authNotRequiredStatus,
+  openAuthStorage,
+  registerClioOAuthProviders,
+  resolveAuthTarget,
+  targetRequiresAuth
+} from "./chunk-EQ63NRB7.js";
 import {
   withStateFileLock
 } from "./chunk-IKCO5N3L.js";
-import {
-  activateExternalPluginApiBridge,
-  calculateEngineCost,
-  createEngineAi,
-  ensurePiAiRegistered,
-  getEngineSupportedThinkingLevels,
-  registerEngineApiProvider
-} from "./chunk-FCSXB6T2.js";
 import {
   require_dist
 } from "./chunk-APJ265NV.js";
@@ -45,329 +57,106 @@ import {
   resolveClioDirs
 } from "./chunk-BNAZZHFG.js";
 import {
-  getSharedBus
-} from "./chunk-ZGVHUX3M.js";
+  getRuntimeRegistry
+} from "./chunk-NUGM5KR6.js";
 import {
-  BusChannels
-} from "./chunk-OB5HIGJY.js";
+  capabilitiesFromCatalogModel,
+  catalogThinkingLevelsForRuntime,
+  getCatalogModelForRuntime,
+  mergeCapabilities,
+  resolveCostProvenance
+} from "./chunk-VO2LKSTM.js";
+import {
+  activateExternalPluginApiBridge,
+  calculateEngineCost,
+  ensurePiAiRegistered,
+  registerEngineApiProvider
+} from "./chunk-UOSL25KY.js";
 import {
   __toESM,
   init_esm_shims
 } from "./chunk-3R73A4XB.js";
 
-// src/domains/providers/capabilities.ts
+// src/domains/providers/endpoint-capacity.ts
 init_esm_shims();
-function mergeCapabilities(base, kb, probe, userOverride) {
-  const merged = { ...base };
-  applyLayer(merged, probe);
-  applyLayer(merged, kb);
-  applyDeploymentLimits(merged, probe);
-  applyLayer(merged, userOverride);
-  return merged;
-}
-function applyDeploymentLimits(target, probe) {
-  if (!probe) return;
-  if (probe.contextWindow !== void 0) target.contextWindow = probe.contextWindow;
-  if (probe.maxTokens !== void 0) target.maxTokens = probe.maxTokens;
-}
-function supportsAgentRoleTools(capabilities2) {
-  return capabilities2.tools === true;
-}
-var AGENT_ROLE_TOOLS_REQUIRED_REASON = "reports no tool support; Clio drives every agent role through typed tools";
-function applyLayer(target, layer) {
-  if (!layer) return;
-  for (const key of Object.keys(layer)) {
-    const value = layer[key];
-    if (value !== void 0) target[key] = value;
-  }
-}
 
-// src/domains/providers/catalog.ts
+// src/core/endpoint-key.ts
 init_esm_shims();
-var engineAi = createEngineAi();
-var CATALOG_PROVIDER_BY_RUNTIME_ID = /* @__PURE__ */ new Map([
-  ["anthropic", "anthropic"],
-  ["anthropic-max", "anthropic"],
-  ["claude-code", "anthropic"],
-  ["claude-sdk", "anthropic"],
-  ["bedrock", "amazon-bedrock"],
-  ["deepseek", "deepseek"],
-  ["google", "google"],
-  ["groq", "groq"],
-  ["mistral", "mistral"],
-  ["openai", "openai"],
-  ["openai-codex", "openai-codex"],
-  ["openrouter", "openrouter"]
-]);
-function catalogProviderForRuntime(runtimeId) {
-  return CATALOG_PROVIDER_BY_RUNTIME_ID.get(runtimeId);
-}
-function listCatalogModelsForRuntime(runtimeId) {
-  const provider = catalogProviderForRuntime(runtimeId);
-  if (!provider) return [];
-  try {
-    return engineAi.listModels(provider);
-  } catch {
-    return [];
-  }
-}
-function resolveEffectivePricing(target, runtimeId, wireModelId) {
-  if (target.pricing) {
-    const rates = {
-      input: target.pricing.input,
-      output: target.pricing.output,
-      cacheRead: target.pricing.cacheRead ?? 0,
-      cacheWrite: target.pricing.cacheWrite ?? 0
-    };
-    return {
-      rates,
-      provenance: Object.values(rates).every((rate) => rate === 0) ? "known_free" : "known"
-    };
-  }
-  const catalogModel = getCatalogModelForRuntime(runtimeId, wireModelId);
-  if (!catalogModel) return { rates: null, provenance: "unknown" };
-  return {
-    rates: {
-      input: catalogModel.cost.input,
-      output: catalogModel.cost.output,
-      cacheRead: catalogModel.cost.cacheRead,
-      cacheWrite: catalogModel.cost.cacheWrite
-    },
-    provenance: "estimated"
-  };
-}
-function resolveCostProvenance(target, runtimeId, wireModelId) {
-  return resolveEffectivePricing(target, runtimeId, wireModelId).provenance;
-}
-function getCatalogModelForRuntime(runtimeId, wireModelId) {
-  const provider = catalogProviderForRuntime(runtimeId);
-  if (!provider) return void 0;
+function canonicalEndpointUrl(raw) {
+  if (!raw?.trim()) return null;
   try {
-    return engineAi.getModel(provider, wireModelId);
+    const url = new URL(raw.trim().replace(/^ws:/u, "http:").replace(/^wss:/u, "https:"));
+    url.protocol = url.protocol.toLowerCase();
+    if (url.protocol !== "http:" && url.protocol !== "https:") return null;
+    url.hostname = url.hostname.toLowerCase();
+    url.username = "";
+    url.password = "";
+    url.hash = "";
+    url.search = "";
+    let path = url.pathname.replace(/\/{2,}/gu, "/").replace(/\/$/u, "");
+    if (path.endsWith("/v1")) path = path.slice(0, -3);
+    url.pathname = path || "/";
+    return url.toString().replace(/\/$/u, "");
   } catch {
-    return void 0;
-  }
-}
-function capabilitiesFromCatalogModel(defaultCapabilities25, model) {
-  if (!model) return defaultCapabilities25;
-  return {
-    ...defaultCapabilities25,
-    reasoning: model.reasoning,
-    vision: model.input.includes("image"),
-    contextWindow: model.contextWindow,
-    maxTokens: model.maxTokens
-  };
-}
-function catalogThinkingLevelsForRuntime(runtimeId, wireModelId) {
-  const model = getCatalogModelForRuntime(runtimeId, wireModelId);
-  return model ? getEngineSupportedThinkingLevels(model) : void 0;
-}
-function synthesizeCatalogBackedModel(input) {
-  const builtin = getCatalogModelForRuntime(input.runtimeId, input.wireModelId);
-  const caps = mergeCapabilities(
-    capabilitiesFromCatalogModel(input.defaultCapabilities, builtin),
-    input.kb?.entry.capabilities ?? null,
-    null,
-    input.target.capabilities ?? null
-  );
-  const pricing = input.target.pricing;
-  const targetHeaders = input.target.auth?.headers;
-  const model = {
-    ...builtin ?? {},
-    id: input.wireModelId,
-    name: `${input.wireModelId} (${input.target.id})`,
-    api: input.api,
-    provider: input.provider,
-    baseUrl: input.target.url ?? builtin?.baseUrl ?? input.defaultBaseUrl,
-    reasoning: caps.reasoning,
-    input: caps.vision ? builtin?.input.includes("image") ? builtin.input : ["text", "image"] : ["text"],
-    cost: {
-      input: pricing?.input ?? builtin?.cost.input ?? 0,
-      output: pricing?.output ?? builtin?.cost.output ?? 0,
-      cacheRead: pricing?.cacheRead ?? builtin?.cost.cacheRead ?? 0,
-      cacheWrite: pricing?.cacheWrite ?? builtin?.cost.cacheWrite ?? 0
-    },
-    contextWindow: caps.contextWindow,
-    maxTokens: caps.maxTokens
-  };
-  const headers = { ...input.defaultHeaders ?? {}, ...builtin?.headers ?? {}, ...targetHeaders ?? {} };
-  if (Object.keys(headers).length > 0) {
-    model.headers = headers;
+    return null;
   }
-  return model;
 }
 
-// src/domains/providers/registry.ts
-init_esm_shims();
-import { readdirSync, statSync } from "node:fs";
-import { join } from "node:path";
-import { pathToFileURL } from "node:url";
-var RUNTIME_KINDS = ["http", "sdk", "subprocess"];
-var RUNTIME_AUTHS = ["api-key", "oauth", "aws-sdk", "vertex-adc", "claude-cli", "none"];
-function createRuntimeRegistry() {
-  const byId = /* @__PURE__ */ new Map();
-  const canonical = /* @__PURE__ */ new Set();
-  const register = (desc) => {
-    const ids = [desc.id, ...desc.aliases ?? []];
-    const conflict = ids.find((id) => byId.has(id));
-    if (conflict) {
-      throw new Error(`runtime id '${conflict}' already registered`);
-    }
-    for (const id of ids) byId.set(id, desc);
-    canonical.add(desc.id);
-  };
-  const get = (id) => byId.get(id) ?? null;
-  const list = () => Array.from(canonical, (id) => byId.get(id)).filter((entry) => entry !== void 0);
-  const clear = () => {
-    byId.clear();
-    canonical.clear();
-  };
-  const loadFromDir = async (dir, beforeImport) => {
-    let entries;
-    try {
-      const stat = statSync(dir);
-      if (!stat.isDirectory()) return [];
-      entries = readdirSync(dir);
-    } catch {
-      return [];
-    }
-    const loaded = [];
-    for (const name of entries) {
-      if (!name.endsWith(".js")) continue;
-      const full = join(dir, name);
-      const desc = await importDescriptor(full, pathToFileURL(full).href, beforeImport);
-      if (desc === null) continue;
-      try {
-        register(desc);
-        loaded.push(desc.id);
-      } catch (err) {
-        process.stderr.write(`[providers] runtime plugin ${full} rejected: ${describeError(err)}
-`);
-      }
-    }
-    return loaded;
-  };
-  const loadFromPackage = async (packageName, beforeImport) => {
-    let mod;
-    try {
-      await beforeImport?.();
-      mod = await import(packageName);
-    } catch (err) {
-      process.stderr.write(`[providers] runtime package ${packageName} failed to import: ${describeError(err)}
-`);
-      return [];
-    }
-    const exported = mod.clioRuntimes;
-    if (!Array.isArray(exported)) {
-      process.stderr.write(`[providers] runtime package ${packageName} has no 'clioRuntimes' array export
-`);
-      return [];
-    }
-    const loaded = [];
-    for (const candidate of exported) {
-      const validation = validateRuntimeDescriptor(candidate);
-      if (!validation.ok) {
-        process.stderr.write(
-          `[providers] runtime package ${packageName} exported an invalid descriptor: ${validation.reason}
-`
-        );
-        continue;
-      }
-      try {
-        register(validation.descriptor);
-        loaded.push(validation.descriptor.id);
-      } catch (err) {
-        process.stderr.write(`[providers] runtime package ${packageName} id conflict: ${describeError(err)}
-`);
-      }
-    }
-    return loaded;
-  };
-  return { register, get, list, clear, loadFromDir, loadFromPackage };
+// src/domains/providers/endpoint-capacity.ts
+var UNBOUNDED_LOCAL_NATIVE_RUNTIMES = /* @__PURE__ */ new Set(["vllm", "sglang"]);
+function canonicalEndpointKey(value) {
+  return canonicalEndpointUrl(typeof value === "string" ? value : value.url);
 }
-var singleton = null;
-function getRuntimeRegistry() {
-  if (singleton === null) singleton = createRuntimeRegistry();
-  return singleton;
-}
-async function importDescriptor(file, href, beforeImport) {
-  let mod;
+function endpointLabel(endpointKey) {
   try {
-    await beforeImport?.();
-    mod = await import(href);
-  } catch (err) {
-    process.stderr.write(`[providers] runtime plugin ${file} failed to import: ${describeError(err)}
-`);
-    return null;
-  }
-  const candidate = mod.default;
-  const validation = validateRuntimeDescriptor(candidate);
-  if (!validation.ok) {
-    process.stderr.write(
-      `[providers] runtime plugin ${file} has invalid default-export RuntimeDescriptor: ${validation.reason}
-`
-    );
-    return null;
-  }
-  return validation.descriptor;
-}
-function validateRuntimeDescriptor(value) {
-  if (typeof value !== "object" || value === null || Array.isArray(value)) {
-    return { ok: false, reason: "descriptor must be an object" };
-  }
-  const v = value;
-  if (typeof v.id !== "string" || v.id.trim().length === 0) {
-    return { ok: false, reason: "id must be a non-empty string" };
-  }
-  if (v.aliases !== void 0 && (!Array.isArray(v.aliases) || v.aliases.some((alias) => typeof alias !== "string" || alias.trim().length === 0))) {
-    return { ok: false, reason: "aliases must be non-empty strings when present" };
-  }
-  if (typeof v.displayName !== "string" || v.displayName.trim().length === 0) {
-    return { ok: false, reason: "displayName must be a non-empty string" };
-  }
-  if (typeof v.kind !== "string" || !RUNTIME_KINDS.includes(v.kind)) {
-    return { ok: false, reason: `kind must be one of: ${RUNTIME_KINDS.join(", ")}` };
-  }
-  if (typeof v.apiFamily !== "string" || v.apiFamily.trim().length === 0) {
-    return { ok: false, reason: "apiFamily must be a non-empty string" };
-  }
-  if (typeof v.auth !== "string" || !RUNTIME_AUTHS.includes(v.auth)) {
-    return { ok: false, reason: `auth must be one of: ${RUNTIME_AUTHS.join(", ")}` };
-  }
-  if (typeof v.defaultCapabilities !== "object" || v.defaultCapabilities === null || Array.isArray(v.defaultCapabilities)) {
-    return { ok: false, reason: "defaultCapabilities must be an object" };
-  }
-  if (typeof v.synthesizeModel !== "function") {
-    return { ok: false, reason: "synthesizeModel must be a function" };
-  }
-  for (const field of ["probe", "probeModels", "complete", "infill", "embed", "rerank"]) {
-    if (v[field] !== void 0 && typeof v[field] !== "function") {
-      return { ok: false, reason: `${field} must be a function when present` };
+    const url = new URL(endpointKey);
+    return `${url.host}${url.pathname === "/" ? "" : url.pathname}`;
+  } catch {
+    return endpointKey;
+  }
+}
+function positiveInteger(value) {
+  return typeof value === "number" && Number.isInteger(value) && value > 0 ? value : void 0;
+}
+function endpointCapacityForStatus(status) {
+  const key = canonicalEndpointKey(status.target);
+  if (key === null) return null;
+  const override = positiveInteger(status.target.maxConcurrentRequests);
+  if (override !== void 0) return { key, label: endpointLabel(key), limit: override, source: "override" };
+  const discovered = positiveInteger(status.probeCapabilities?.parallelSlots);
+  if (discovered !== void 0) return { key, label: endpointLabel(key), limit: discovered, source: "discovered" };
+  if (status.runtime?.tier !== "local-native" || UNBOUNDED_LOCAL_NATIVE_RUNTIMES.has(status.runtime.id)) return null;
+  return { key, label: endpointLabel(key), limit: 1, source: "local-native-default" };
+}
+function sourceRank(source) {
+  return source === "override" ? 2 : source === "discovered" ? 1 : 0;
+}
+function endpointCapacitiesForStatuses(statuses) {
+  const capacities = {};
+  for (const status of statuses) {
+    const candidate = endpointCapacityForStatus(status);
+    if (candidate === null) continue;
+    const current = capacities[candidate.key];
+    if (current === void 0 || sourceRank(candidate.source) > sourceRank(current.source) || sourceRank(candidate.source) === sourceRank(current.source) && candidate.limit < current.limit) {
+      capacities[candidate.key] = candidate;
     }
   }
-  return { ok: true, descriptor: value };
-}
-function describeError(err) {
-  if (err instanceof Error) return err.message;
-  return String(err);
+  return capacities;
+}
+var foregroundStreams = /* @__PURE__ */ new Map();
+function registerForegroundStream(endpointKey) {
+  foregroundStreams.set(endpointKey, (foregroundStreams.get(endpointKey) ?? 0) + 1);
+  let held = true;
+  return () => {
+    if (!held) return;
+    held = false;
+    const next = (foregroundStreams.get(endpointKey) ?? 1) - 1;
+    if (next <= 0) foregroundStreams.delete(endpointKey);
+    else foregroundStreams.set(endpointKey, next);
+  };
 }
-
-// src/domains/providers/credentials.ts
-init_esm_shims();
-function credentialsPresent() {
-  const present = /* @__PURE__ */ new Set();
-  const registry = getRuntimeRegistry();
-  const auth = openAuthStorage();
-  for (const desc of registry.list()) {
-    const envVar = desc.credentialsEnvVar;
-    if (!envVar) continue;
-    const providerId = desc.id;
-    const status = auth.status(providerId, { explicitEnvVar: envVar, includeFallback: false });
-    if (status.available) {
-      present.add(envVar);
-    }
-  }
-  return present;
+function foregroundStreamUsage() {
+  return Object.fromEntries(foregroundStreams);
 }
 
 // src/engine/apis/residency.ts
@@ -375,10 +164,11 @@ init_esm_shims();
 
 // src/engine/apis/residency-lock.ts
 init_esm_shims();
-import { join as join2 } from "node:path";
+import { join } from "node:path";
 var LOCK_WAIT_MS = 12e4;
 function lockTargetFor(targetKey) {
-  return join2(clioStatePath(), "residency-locks", targetKey.replace(/[^a-zA-Z0-9._-]+/g, "_"));
+  const key = canonicalEndpointKey(targetKey) ?? targetKey;
+  return join(clioStatePath(), "residency-locks", key.replace(/[^a-zA-Z0-9._-]+/g, "_"));
 }
 async function withResidencyLock(targetKey, fn) {
   return withStateFileLock(lockTargetFor(targetKey), fn, {
@@ -407,6 +197,9 @@ function emitResidencyNotice(notice) {
   } catch {
   }
 }
+function emitResidencyMutation(mutation) {
+  getSharedBus().emit(BusChannels.ResidencyMutation, { ...mutation, at: Date.now() });
+}
 var protectedModelsProvider = null;
 function setProtectedModelsProvider(provider) {
   protectedModelsProvider = provider;
@@ -637,6 +430,13 @@ async function restoreEvictedModels(adapter, evicted, loadError) {
     try {
       await adapter.reloadEvicted(entry.modelId);
       restored.push(entry.modelId);
+      emitResidencyMutation({
+        targetKey: adapter.targetKey,
+        targetId: adapter.targetId,
+        runtimeId: adapter.runtimeId,
+        model: entry.modelId,
+        operation: "load"
+      });
     } catch {
       failed.push(entry.modelId);
     }
@@ -753,12 +553,28 @@ async function reconcileResidency(adapter) {
           await adapter.unload(entry.modelId);
           forgetClioLoaded(adapter.targetKey, entry.modelId);
           evicted.push(entry);
+          emitResidencyMutation({
+            targetKey: adapter.targetKey,
+            targetId: adapter.targetId,
+            runtimeId: adapter.runtimeId,
+            model: entry.modelId,
+            operation: "evict"
+          });
         } catch {
         }
       }
       if (!adapter.load) return;
       try {
         await adapter.load(adapter.keepModelId);
+        if (!plan.keepResident) {
+          emitResidencyMutation({
+            targetKey: adapter.targetKey,
+            targetId: adapter.targetId,
+            runtimeId: adapter.runtimeId,
+            model: adapter.keepModelId,
+            operation: "load"
+          });
+        }
       } catch (error) {
         await restoreEvictedModels(adapter, evicted, error);
         throw error;
@@ -927,130 +743,6 @@ function availableThinkingLevels(caps, options) {
 
 // src/domains/providers/model-capabilities.ts
 init_esm_shims();
-
-// src/domains/providers/types/local-model-quirks.ts
-init_esm_shims();
-var KV_CACHE_QUANTS = ["f32", "f16", "q8_0", "q4_0", "q4_1", "iq4_nl", "q5_0", "q5_1"];
-function isRecord(value) {
-  return value !== null && typeof value === "object" && !Array.isArray(value);
-}
-function asKvCacheQuant(value) {
-  if (value === false) return false;
-  if (typeof value !== "string") return void 0;
-  return KV_CACHE_QUANTS.includes(value) ? value : void 0;
-}
-function asPositive(value) {
-  return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : void 0;
-}
-function asInteger(value) {
-  const n = asPositive(value);
-  return n !== void 0 && Number.isInteger(n) ? n : void 0;
-}
-function asPenalty(value) {
-  return typeof value === "number" && Number.isFinite(value) ? value : void 0;
-}
-function extractKvCache(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  const k = asKvCacheQuant(raw.kQuant);
-  if (k !== void 0) out.kQuant = k;
-  const v = asKvCacheQuant(raw.vQuant);
-  if (v !== void 0) out.vQuant = v;
-  if (typeof raw.useFp16 === "boolean") out.useFp16 = raw.useFp16;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-function extractSamplingProfile(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  const temperature = asPositive(raw.temperature);
-  if (temperature !== void 0) out.temperature = temperature;
-  const topP = asPositive(raw.topP);
-  if (topP !== void 0) out.topP = topP;
-  const topK = asInteger(raw.topK);
-  if (topK !== void 0) out.topK = topK;
-  const minP = asPositive(raw.minP);
-  if (minP !== void 0) out.minP = minP;
-  const rp = asPenalty(raw.repeatPenalty) ?? asPenalty(raw.repetitionPenalty);
-  if (rp !== void 0) out.repeatPenalty = rp;
-  const pp = asPenalty(raw.presencePenalty);
-  if (pp !== void 0) out.presencePenalty = pp;
-  const fp = asPenalty(raw.frequencyPenalty);
-  if (fp !== void 0) out.frequencyPenalty = fp;
-  const maxTokens = asInteger(raw.maxTokens);
-  if (maxTokens !== void 0) out.maxTokens = maxTokens;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-function extractSampling(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  const thinking = extractSamplingProfile(raw.thinking);
-  if (thinking) out.thinking = thinking;
-  const instruct = extractSamplingProfile(raw.instruct);
-  if (instruct) out.instruct = instruct;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-var THINKING_MECHANISMS = [
-  "effort-levels",
-  "budget-tokens",
-  "on-off",
-  "always-on",
-  "none"
-];
-function asThinkingMechanism(value) {
-  if (typeof value !== "string") return void 0;
-  return THINKING_MECHANISMS.includes(value) ? value : void 0;
-}
-function extractBudgetByLevel(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  const minimal = asInteger(raw.minimal);
-  if (minimal !== void 0) out.minimal = minimal;
-  const low = asInteger(raw.low);
-  if (low !== void 0) out.low = low;
-  const medium = asInteger(raw.medium);
-  if (medium !== void 0) out.medium = medium;
-  const high = asInteger(raw.high);
-  if (high !== void 0) out.high = high;
-  const xhigh = asInteger(raw.xhigh);
-  if (xhigh !== void 0) out.xhigh = xhigh;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-function extractEffortByLevel(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  if (typeof raw.off === "string" && raw.off.length > 0) out.off = raw.off;
-  if (typeof raw.minimal === "string" && raw.minimal.length > 0) out.minimal = raw.minimal;
-  if (typeof raw.low === "string" && raw.low.length > 0) out.low = raw.low;
-  if (typeof raw.medium === "string" && raw.medium.length > 0) out.medium = raw.medium;
-  if (typeof raw.high === "string" && raw.high.length > 0) out.high = raw.high;
-  if (typeof raw.xhigh === "string" && raw.xhigh.length > 0) out.xhigh = raw.xhigh;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-function extractThinkingQuirks(raw) {
-  if (!isRecord(raw)) return void 0;
-  const mechanism = asThinkingMechanism(raw.mechanism);
-  if (!mechanism) return void 0;
-  const out = { mechanism };
-  const budgetByLevel = extractBudgetByLevel(raw.budgetByLevel);
-  if (budgetByLevel) out.budgetByLevel = budgetByLevel;
-  const effortByLevel = extractEffortByLevel(raw.effortByLevel);
-  if (effortByLevel) out.effortByLevel = effortByLevel;
-  if (typeof raw.guidance === "string" && raw.guidance.length > 0) out.guidance = raw.guidance;
-  return out;
-}
-function extractLocalModelQuirks(raw) {
-  if (!isRecord(raw)) return void 0;
-  const out = {};
-  const kvCache = extractKvCache(raw.kvCache);
-  if (kvCache) out.kvCache = kvCache;
-  const sampling = extractSampling(raw.sampling);
-  if (sampling) out.sampling = sampling;
-  const thinking = extractThinkingQuirks(raw.thinking);
-  if (thinking) out.thinking = thinking;
-  return Object.keys(out).length > 0 ? out : void 0;
-}
-
-// src/domains/providers/model-capabilities.ts
 function normalizedModelId(wireModelId) {
   const trimmed = wireModelId?.trim();
   return trimmed ? trimmed : null;
@@ -1478,2982 +1170,129 @@ function resolveModelRuntimeCapabilities(input) {
   const thinking = resolveThinkingCapability(input, quirks, parser);
   const result = {
     targetId: input.targetId ?? null,
-    runtimeId: input.runtimeId,
-    apiFamily: input.apiFamily ?? null,
-    modelId: input.modelId,
-    family,
-    capabilities: input.capabilities,
-    thinking,
-    request: resolveRequestCapability(thinking, parser, input.runtimeId),
-    response: {
-      parser,
-      stripTokenizerSentinels: true
-    }
-  };
-  if (quirks) result.quirks = quirks;
-  return result;
-}
-function resolveModelRuntimeCapabilitiesForStatus(status, wireModelId, knowledgeBase, options) {
-  const modelId = wireModelId?.trim() || status.target.defaultModel?.trim() || "";
-  const kbHit = modelId ? knowledgeBase?.lookup(modelId) ?? null : null;
-  const capabilities2 = resolveModelCapabilities(status, modelId, knowledgeBase, {
-    detectedReasoning: options?.detectedReasoning ?? null
-  });
-  const runtimeId = status.runtime?.id ?? status.target.runtime;
-  return resolveModelRuntimeCapabilities({
-    targetId: status.target.id,
-    runtimeId,
-    apiFamily: status.runtime?.apiFamily ?? null,
-    modelId,
-    capabilities: capabilities2,
-    kbHit,
-    ...thinkingHintsForCatalogModel(runtimeId, modelId),
-    ...options?.configuredThinkingLevel ? { configuredThinkingLevel: options.configuredThinkingLevel } : {}
-  });
-}
-function resolveModelRuntimeCapabilitiesForProviders(providers, targetId, wireModelId, configuredThinkingLevel) {
-  const id = targetId?.trim();
-  if (!id) return null;
-  const status = providers.list().find((entry) => entry.target.id === id);
-  if (!status) return null;
-  const modelId = wireModelId?.trim() || status.target.defaultModel?.trim() || "";
-  const detectedReasoning = modelId && typeof providers.getDetectedReasoning === "function" ? providers.getDetectedReasoning(id, modelId) : null;
-  return resolveModelRuntimeCapabilitiesForStatus(status, modelId, providers.knowledgeBase, {
-    detectedReasoning,
-    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
-  });
-}
-function thinkingHintsForModel(model) {
-  const out = {};
-  if (!model) return out;
-  const compat = model.compat;
-  if (compat?.forceAdaptiveThinking !== void 0) out.adaptiveThinking = compat.forceAdaptiveThinking;
-  if (model.thinkingLevelMap) out.thinkingLevelMap = model.thinkingLevelMap;
-  return out;
-}
-function thinkingHintsForCatalogModel(runtimeId, modelId) {
-  if (!runtimeId || modelId.length === 0) return {};
-  return thinkingHintsForModel(getCatalogModelForRuntime(runtimeId, modelId));
-}
-function thinkingFormatFromModelApi(api) {
-  switch (api) {
-    case "anthropic-messages":
-    case "bedrock-converse-stream":
-    case "claude-agent-sdk":
-    case "claude-code-subprocess":
-      return "anthropic-extended";
-    case "openai-codex-responses":
-      return "openai-codex";
-    default:
-      return void 0;
-  }
-}
-function capabilitiesFromModel(model) {
-  const format = model.compat?.thinkingFormat ?? thinkingFormatFromModelApi(model.api);
-  const caps = {
-    chat: true,
-    tools: true,
-    reasoning: model.reasoning === true,
-    vision: Array.isArray(model.input) && model.input.includes("image"),
-    audio: false,
-    embeddings: false,
-    rerank: false,
-    fim: false,
-    contextWindow: model.contextWindow,
-    maxTokens: model.maxTokens
-  };
-  if (format === "qwen-chat-template" || format === "openrouter" || format === "zai" || format === "anthropic-extended" || format === "deepseek-r1" || format === "openai-codex" || format === "harmony") {
-    caps.thinkingFormat = format;
-  }
-  return caps;
-}
-function resolveModelRuntimeCapabilitiesForModel(model, configuredThinkingLevel) {
-  const metadata2 = model.clio;
-  const caps = capabilitiesFromModel(model);
-  return resolveModelRuntimeCapabilities({
-    targetId: metadata2?.targetId ?? null,
-    runtimeId: metadata2?.runtimeId ?? model.provider,
-    apiFamily: model.api,
-    modelId: model.id,
-    capabilities: caps,
-    ...thinkingHintsForModel(model),
-    ...metadata2?.quirks ? { quirks: metadata2.quirks } : {},
-    kbHit: metadata2?.family ? {
-      matchKind: "family",
-      entry: {
-        family: metadata2.family,
-        matchPatterns: [metadata2.family],
-        capabilities: {}
-      }
-    } : null,
-    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
-  });
-}
-function resolveTargetRuntimeCapabilities(target, runtime, wireModelId, capabilities2, knowledgeBase, configuredThinkingLevel) {
-  const kbHit = knowledgeBase?.lookup(wireModelId) ?? null;
-  return resolveModelRuntimeCapabilities({
-    targetId: target.id,
-    runtimeId: runtime.id,
-    apiFamily: runtime.apiFamily,
-    modelId: wireModelId,
-    capabilities: capabilities2,
-    kbHit,
-    ...thinkingHintsForCatalogModel(runtime.id, wireModelId),
-    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
-  });
-}
-
-// src/domains/providers/runtimes/common/lmstudio-http.ts
-init_esm_shims();
-import { performance } from "node:perf_hooks";
-
-// src/domains/providers/runtimes/common/local-synth.ts
-init_esm_shims();
-function openAIThinkingFormat(caps) {
-  switch (caps.thinkingFormat) {
-    case "qwen-chat-template":
-    case "openrouter":
-    case "zai":
-      return caps.thinkingFormat;
-    case "deepseek-r1":
-      return "deepseek";
-    case "harmony":
-      return "harmony";
-    default:
-      return void 0;
-  }
-}
-function localOpenAICompat(caps, runtimeId) {
-  const compat = {
-    supportsStore: false,
-    supportsDeveloperRole: false,
-    supportsReasoningEffort: false,
-    supportsUsageInStreaming: true,
-    supportsFinishReason: false,
-    maxTokensField: "max_tokens",
-    supportsThinkingTokenBudget: runtimeId === "vllm",
-    supportsStrictMode: false
-  };
-  const thinkingFormat = openAIThinkingFormat(caps);
-  if (thinkingFormat) compat.thinkingFormat = thinkingFormat;
-  return compat;
-}
-function localAnthropicCompat() {
-  return {
-    supportsEagerToolInputStreaming: false,
-    supportsLongCacheRetention: false
-  };
-}
-function synthLocalModel(input) {
-  const { target, wireModelId, kb, defaultCapabilities: defaultCapabilities25, apiFamily, provider } = input;
-  const caps = mergeCapabilities(defaultCapabilities25, kb?.entry.capabilities ?? null, null, target.capabilities ?? null);
-  const rawUrl = target.url ?? "";
-  const baseUrl = rawUrl.length > 0 ? input.baseUrlForTarget(rawUrl) : "";
-  const pricing = target.pricing;
-  const headers = target.auth?.headers;
-  const quirks = extractLocalModelQuirks(kb?.entry.quirks);
-  const model = {
-    id: wireModelId,
-    name: `${wireModelId} (${target.id})`,
-    api: apiFamily,
-    provider,
-    baseUrl,
-    reasoning: caps.reasoning,
-    input: caps.vision ? ["text", "image"] : ["text"],
-    cost: {
-      input: pricing?.input ?? 0,
-      output: pricing?.output ?? 0,
-      cacheRead: pricing?.cacheRead ?? 0,
-      cacheWrite: pricing?.cacheWrite ?? 0
-    },
-    contextWindow: caps.contextWindow,
-    maxTokens: caps.maxTokens,
-    clio: {
-      targetId: target.id,
-      runtimeId: target.runtime,
-      ...target.lifecycle ? { lifecycle: target.lifecycle } : {},
-      ...target.gateway === true ? { gateway: true } : {},
-      ...kb?.entry.family ? { family: kb.entry.family } : {},
-      ...quirks ? { quirks } : {},
-      ...target.lmstudio ? { lmstudio: target.lmstudio } : {}
-    }
-  };
-  if (headers) model.headers = headers;
-  if (apiFamily === "openai-completions") {
-    model.compat = localOpenAICompat(caps, target.runtime);
-  }
-  if (apiFamily === "anthropic-messages") {
-    model.compat = localAnthropicCompat();
-  }
-  return model;
-}
-function stripTrailingSlash(url2) {
-  return url2.endsWith("/") ? url2.slice(0, -1) : url2;
-}
-function stripRedundantV1(url2) {
-  return url2.endsWith("/v1") ? url2.slice(0, -"/v1".length) : url2;
-}
-var withV1 = (url2) => `${stripRedundantV1(stripTrailingSlash(url2))}/v1`;
-var withAsIs = (url2) => stripTrailingSlash(url2);
-function targetBaseUrl(target) {
-  return target.url ? stripTrailingSlash(target.url) : null;
-}
-function targetRootUrl(target) {
-  return target.url ? stripRedundantV1(stripTrailingSlash(target.url)) : null;
-}
-
-// src/domains/providers/runtimes/common/lmstudio-http.ts
-var catalogsByHost = /* @__PURE__ */ new Map();
-function isRecord2(value) {
-  return value !== null && typeof value === "object" && !Array.isArray(value);
-}
-function positiveNumber(value) {
-  return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : void 0;
-}
-function nonEmptyString(value) {
-  return typeof value === "string" && value.trim().length > 0 ? value.trim() : void 0;
-}
-function lmStudioRootUrl(url2) {
-  const normalized = targetRootUrl({ id: "lmstudio", runtime: "lmstudio", url: url2 });
-  if (!normalized) return "";
-  if (normalized.startsWith("ws://")) return `http://${normalized.slice("ws://".length)}`;
-  if (normalized.startsWith("wss://")) return `https://${normalized.slice("wss://".length)}`;
-  return normalized;
-}
-function lmStudioRequestHeaders(base, apiKey) {
-  const headers = { ...base ?? {} };
-  const token = apiKey?.trim();
-  if (token) headers.authorization = `Bearer ${token}`;
-  return headers;
-}
-function lmStudioProbeHeaders(target, ctx) {
-  let token = ctx.authToken?.trim();
-  const envName = target.auth?.apiKeyEnvVar;
-  if (!token && envName && ctx.credentialsPresent.has(envName)) token = process.env[envName]?.trim();
-  return lmStudioRequestHeaders(target.auth?.headers, token);
-}
-function combinedSignal(timeoutMs, signal) {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(new Error(`timeout after ${timeoutMs}ms`)), timeoutMs);
-  const onAbort = () => controller.abort(signal?.reason);
-  if (signal?.aborted) controller.abort(signal.reason);
-  else signal?.addEventListener("abort", onAbort, { once: true });
-  return {
-    signal: controller.signal,
-    close() {
-      clearTimeout(timer);
-      signal?.removeEventListener("abort", onAbort);
-    }
-  };
-}
-async function requestLmStudioJson(url2, init, timeoutMs, signal) {
-  const started = performance.now();
-  const bounded = combinedSignal(timeoutMs, signal);
-  try {
-    const response = await fetch(url2, { ...init, signal: bounded.signal });
-    let data;
-    try {
-      data = await response.json();
-    } catch {
-      data = void 0;
-    }
-    const result = {
-      ok: response.ok,
-      status: response.status,
-      latencyMs: Math.round(performance.now() - started)
-    };
-    if (data !== void 0) result.data = data;
-    if (!response.ok) {
-      const message = isRecord2(data) ? nonEmptyString(data.error) : void 0;
-      result.error = message ?? `HTTP ${response.status}`;
-    }
-    return result;
-  } catch (error) {
-    return {
-      ok: false,
-      status: 0,
-      latencyMs: Math.round(performance.now() - started),
-      error: error instanceof Error ? error.message : String(error)
-    };
-  } finally {
-    bounded.close();
-  }
-}
-function loadedInstances(value) {
-  if (!Array.isArray(value)) return [];
-  const instances = [];
-  for (const raw of value) {
-    if (!isRecord2(raw)) continue;
-    const id = nonEmptyString(raw.id);
-    if (!id) continue;
-    instances.push({ id, config: isRecord2(raw.config) ? { ...raw.config } : {} });
-  }
-  return instances;
-}
-function reasoningOptions(capabilities2) {
-  if (!isRecord2(capabilities2) || !isRecord2(capabilities2.reasoning)) return void 0;
-  const allowed = capabilities2.reasoning.allowed_options;
-  if (!Array.isArray(allowed)) return void 0;
-  const options = allowed.filter((entry) => typeof entry === "string");
-  return options.length > 0 ? options : void 0;
-}
-function parseLmStudioV1Models(data) {
-  if (!isRecord2(data) || !Array.isArray(data.models)) return null;
-  const models = [];
-  for (const raw of data.models) {
-    if (!isRecord2(raw)) continue;
-    const key = nonEmptyString(raw.key);
-    if (!key) continue;
-    const capabilities2 = isRecord2(raw.capabilities) ? raw.capabilities : void 0;
-    const info = {
-      key,
-      loadedInstances: loadedInstances(raw.loaded_instances),
-      metadata: { ...raw }
-    };
-    const type = nonEmptyString(raw.type);
-    if (type) info.type = type;
-    const maxContextLength = positiveNumber(raw.max_context_length);
-    if (maxContextLength !== void 0) info.maxContextLength = maxContextLength;
-    if (capabilities2 && typeof capabilities2.vision === "boolean") info.vision = capabilities2.vision;
-    else if (type) info.vision = type === "vlm";
-    if (capabilities2 && typeof capabilities2.trained_for_tool_use === "boolean") {
-      info.tools = capabilities2.trained_for_tool_use;
-    }
-    const options = reasoningOptions(capabilities2);
-    if (options) info.reasoningOptions = options;
-    if (capabilities2 && "reasoning" in capabilities2) {
-      info.reasoning = capabilities2.reasoning === true || isRecord2(capabilities2.reasoning);
-    }
-    models.push(info);
-  }
-  return models;
-}
-function foldLmStudioModels(models) {
-  const byKey = /* @__PURE__ */ new Map();
-  for (const model of models) {
-    const previous = byKey.get(model.key);
-    if (!previous) {
-      byKey.set(model.key, { ...model, loadedInstances: [...model.loadedInstances] });
-      continue;
-    }
-    const knownIds = new Set(previous.loadedInstances.map((instance) => instance.id));
-    for (const instance of model.loadedInstances) {
-      if (!knownIds.has(instance.id)) previous.loadedInstances.push(instance);
-    }
-    if (previous.maxContextLength === void 0 && model.maxContextLength !== void 0) {
-      previous.maxContextLength = model.maxContextLength;
-    }
-    if (previous.vision === void 0 && model.vision !== void 0) previous.vision = model.vision;
-    if (previous.tools === void 0 && model.tools !== void 0) previous.tools = model.tools;
-    if (previous.reasoningOptions === void 0 && model.reasoningOptions !== void 0) {
-      previous.reasoningOptions = model.reasoningOptions;
-    }
-    if (previous.reasoning === void 0 && model.reasoning !== void 0) previous.reasoning = model.reasoning;
-  }
-  return [...byKey.values()];
-}
-function rememberHostCatalog(target, models) {
-  const rootUrl2 = lmStudioRootUrl(target.url ?? "");
-  if (!rootUrl2) return;
-  catalogsByHost.set(rootUrl2, { targetId: target.id, rootUrl: rootUrl2, models: foldLmStudioModels(models) });
-}
-function peerTargetsFor(target, instanceId) {
-  const rootUrl2 = lmStudioRootUrl(target.url ?? "");
-  const peers = [];
-  for (const catalog of catalogsByHost.values()) {
-    if (catalog.rootUrl === rootUrl2) continue;
-    if (catalog.models.some((model) => model.loadedInstances.some((instance) => instance.id === instanceId))) {
-      peers.push(catalog.targetId);
-    }
-  }
-  return [...new Set(peers)];
-}
-function resolveLmStudioInstance(target, models, requestedId, configuredDefault) {
-  const folded = foldLmStudioModels(models);
-  for (const model of folded) {
-    const explicit = model.loadedInstances.find((instance2) => instance2.id === requestedId);
-    if (explicit) {
-      return {
-        requestedId,
-        wireModelId: explicit.id,
-        model,
-        instance: explicit,
-        peerTargets: peerTargetsFor(target, explicit.id),
-        state: "instance"
-      };
-    }
-    if (model.key !== requestedId) continue;
-    if (model.loadedInstances.length === 0) {
-      return { requestedId, wireModelId: requestedId, model, peerTargets: [], state: "jit" };
-    }
-    const preferred = configuredDefault ? model.loadedInstances.find((instance2) => instance2.id === configuredDefault) : void 0;
-    const local = model.loadedInstances.find((instance2) => peerTargetsFor(target, instance2.id).length === 0);
-    const instance = preferred ?? local ?? model.loadedInstances[0];
-    if (!instance) return { requestedId, wireModelId: requestedId, model, peerTargets: [], state: "jit" };
-    return {
-      requestedId,
-      wireModelId: instance.id,
-      model,
-      instance,
-      peerTargets: peerTargetsFor(target, instance.id),
-      state: "instance"
-    };
-  }
-  return { requestedId, wireModelId: requestedId, peerTargets: [], state: "unknown" };
-}
-function lmStudioReasoningLevels(options) {
-  if (!options || options.length === 0) return [...THINKING_LEVELS];
-  if (options.includes("on") && !options.includes("low") && !options.includes("medium") && !options.includes("high")) {
-    return ["off", "low"];
-  }
-  const levels = ["off"];
-  if (options.includes("low")) levels.push("minimal", "low");
-  if (options.includes("medium")) levels.push("medium");
-  if (options.includes("high")) levels.push("high", "xhigh", "max");
-  return [...new Set(levels)];
-}
-function lmStudioReasoningEffort(level, options) {
-  if (level === "off") return "none";
-  if (options?.includes("on") && !options.includes("medium") && !options.includes("high")) return "low";
-  if (level === "medium" && options?.includes("medium") !== false) return "medium";
-  if ((level === "high" || level === "xhigh" || level === "max") && options?.includes("high") !== false) return "high";
-  return "low";
-}
-function v0Models(data) {
-  if (!isRecord2(data) || !Array.isArray(data.data)) return null;
-  const models = [];
-  for (const raw of data.data) {
-    if (!isRecord2(raw)) continue;
-    const key = nonEmptyString(raw.id);
-    if (!key) continue;
-    const state = nonEmptyString(raw.state);
-    const loadedContext = positiveNumber(raw.loaded_context_length);
-    const config = loadedContext === void 0 ? {} : { context_length: loadedContext };
-    const info = {
-      key,
-      loadedInstances: state === "loaded" || loadedContext !== void 0 ? [{ id: key, config }] : [],
-      metadata: { ...raw }
-    };
-    const type = nonEmptyString(raw.type);
-    if (type) {
-      info.type = type;
-      info.vision = type === "vlm";
-    }
-    const maxContextLength = positiveNumber(raw.max_context_length);
-    if (maxContextLength !== void 0) info.maxContextLength = maxContextLength;
-    if (Array.isArray(raw.capabilities)) info.tools = raw.capabilities.includes("tool_use");
-    models.push(info);
-  }
-  return models;
-}
-function openAIModels(data) {
-  if (!isRecord2(data) || !Array.isArray(data.data)) return null;
-  const models = [];
-  for (const raw of data.data) {
-    if (!isRecord2(raw)) continue;
-    const key = nonEmptyString(raw.id);
-    if (key) models.push({ key, loadedInstances: [], metadata: { ...raw } });
-  }
-  return models;
-}
-function authFailure(result) {
-  if (result.status !== 401 && result.status !== 403) return null;
-  return {
-    ok: false,
-    models: [],
-    latencyMs: result.latencyMs,
-    error: "LM Studio authentication required",
-    authRequired: true
-  };
-}
-async function listLmStudioModels(target, ctx) {
-  if (!target.url) return { ok: false, models: [], error: "target has no url" };
-  const root = lmStudioRootUrl(target.url);
-  const headers = lmStudioProbeHeaders(target, ctx);
-  const init = { headers };
-  const v1 = await requestLmStudioJson(`${root}/api/v1/models`, init, ctx.httpTimeoutMs, ctx.signal);
-  const v1Auth = authFailure(v1);
-  if (v1Auth) return v1Auth;
-  const parsedV1 = parseLmStudioV1Models(v1.data);
-  if (v1.ok && parsedV1) {
-    rememberHostCatalog(target, parsedV1);
-    return { ok: true, models: parsedV1, tier: "0.4+", latencyMs: v1.latencyMs };
-  }
-  const v0 = await requestLmStudioJson(`${root}/api/v0/models`, init, ctx.httpTimeoutMs, ctx.signal);
-  const v0Auth = authFailure(v0);
-  if (v0Auth) return v0Auth;
-  const parsedV0 = v0Models(v0.data);
-  if (v0.ok && parsedV0) {
-    rememberHostCatalog(target, parsedV0);
-    return { ok: true, models: parsedV0, tier: "0.3.x", latencyMs: v0.latencyMs };
-  }
-  const openAI = await requestLmStudioJson(`${root}/v1/models`, init, ctx.httpTimeoutMs, ctx.signal);
-  const openAIAuth = authFailure(openAI);
-  if (openAIAuth) return openAIAuth;
-  const parsedOpenAI = openAIModels(openAI.data);
-  if (openAI.ok && parsedOpenAI) {
-    rememberHostCatalog(target, parsedOpenAI);
-    return { ok: true, models: parsedOpenAI, tier: "openai-compat", latencyMs: openAI.latencyMs };
-  }
-  return {
-    ok: false,
-    models: [],
-    latencyMs: v1.latencyMs + v0.latencyMs + openAI.latencyMs,
-    error: v1.error ?? v0.error ?? openAI.error ?? "LM Studio returned no recognized model catalog"
-  };
-}
-async function greetLmStudio(target, ctx) {
-  if (!target.url) return { ok: false, error: "target has no url" };
-  const result = await requestLmStudioJson(
-    `${lmStudioRootUrl(target.url)}/lmstudio-greeting`,
-    { headers: lmStudioProbeHeaders(target, ctx) },
-    ctx.httpTimeoutMs,
-    ctx.signal
-  );
-  if (result.ok && isRecord2(result.data) && result.data.lmstudio === true) {
-    return { ok: true, latencyMs: result.latencyMs };
-  }
-  return { ok: false, latencyMs: result.latencyMs, error: result.error ?? "endpoint did not return {lmstudio:true}" };
-}
-function loadedContextLength(instance) {
-  return positiveNumber(instance?.config.context_length);
-}
-
-// src/domains/providers/runtimes/builtins.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/antigravity/antigravity-code.ts
-init_esm_shims();
-var ANTIGRAVITY_AUTH_NOTICE = "Uses your existing Antigravity (`agy`) login. Clio stores no Antigravity credentials. agy runs its own agent harness as a subprocess; Clio maps autonomy levels onto its CLI flags but cannot mediate individual agy tool calls.";
-var ANTIGRAVITY_MODELS = [
-  "Gemini 3.5 Flash (High)",
-  "Gemini 3.5 Flash (Medium)",
-  "Gemini 3.5 Flash (Low)",
-  "Gemini 3.1 Pro (High)",
-  "Gemini 3.1 Pro (Low)",
-  "Claude Sonnet 4.6 (Thinking)",
-  "Claude Opus 4.6 (Thinking)",
-  "GPT-OSS 120B (Medium)"
-];
-var antigravityCapabilities = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 1e6,
-  maxTokens: 8192
-};
-var antigravityCodeRuntime = {
-  id: "antigravity-code",
-  displayName: "Antigravity CLI",
-  kind: "subprocess",
-  tier: "subscription",
-  // Reuses the existing `google-generative-ai` api family: the worker branches
-  // to its own runner before any pi-ai inference, so this only needs to be a
-  // valid, google-backed family. The runner is selected by runtime id, and the
-  // plain-text parser is named by `outputParser`.
-  apiFamily: "google-generative-ai",
-  auth: "none",
-  authNotice: ANTIGRAVITY_AUTH_NOTICE,
-  knownModels: [...ANTIGRAVITY_MODELS],
-  binaryName: "agy",
-  headlessCommand: "agy --print",
-  outputParser: "antigravity-print-text",
-  defaultCapabilities: antigravityCapabilities,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: antigravityCapabilities,
-      runtimeId: "antigravity-code",
-      api: "google-generative-ai",
-      provider: "google",
-      defaultBaseUrl: "antigravity://local"
-    });
-  }
-};
-var antigravity_code_default = antigravityCodeRuntime;
-
-// src/domains/providers/runtimes/claude/claude-code.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/claude/common.ts
-init_esm_shims();
-var CLAUDE_CODE_AUTH_NOTICE = "Uses your existing Claude Code login from the installed `claude` command. Clio stores no Claude Code credentials.";
-var CLAUDE_CODE_MODELS = [
-  "sonnet",
-  "opus",
-  "haiku",
-  "claude-sonnet-4-5",
-  "claude-opus-4-5",
-  "claude-haiku-4-5",
-  "claude-3-7-sonnet-latest"
-];
-var claudeCodeCapabilities = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: true,
-  thinkingFormat: "anthropic-extended",
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 2e5,
-  maxTokens: 8192
-};
-function synthesizeClaudeDelegatedModel(input) {
-  return synthesizeCatalogBackedModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    runtimeId: input.runtimeId,
-    api: input.apiFamily,
-    provider: "anthropic",
-    defaultBaseUrl: "claude-code://local"
-  });
-}
-
-// src/domains/providers/runtimes/claude/claude-code.ts
-var claudeCodeRuntime = {
-  id: "claude-code",
-  displayName: "Claude Code CLI",
-  kind: "subprocess",
-  tier: "subscription",
-  apiFamily: "claude-code-subprocess",
-  auth: "claude-cli",
-  authNotice: CLAUDE_CODE_AUTH_NOTICE,
-  knownModels: [...CLAUDE_CODE_MODELS],
-  binaryName: "claude",
-  headlessCommand: "claude -p --output-format stream-json",
-  outputParser: "claude-code-stream-json",
-  defaultCapabilities: claudeCodeCapabilities,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeClaudeDelegatedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: claudeCodeCapabilities,
-      runtimeId: "claude-code",
-      apiFamily: "claude-code-subprocess"
-    });
-  }
-};
-var claude_code_default = claudeCodeRuntime;
-
-// src/domains/providers/runtimes/claude/claude-sdk.ts
-init_esm_shims();
-var claudeSdkRuntime = {
-  id: "claude-sdk",
-  displayName: "Claude Agent SDK",
-  kind: "sdk",
-  tier: "subscription",
-  apiFamily: "claude-agent-sdk",
-  auth: "claude-cli",
-  authNotice: CLAUDE_CODE_AUTH_NOTICE,
-  knownModels: [...CLAUDE_CODE_MODELS],
-  binaryName: "claude",
-  headlessCommand: "@anthropic-ai/claude-agent-sdk query()",
-  outputParser: "claude-agent-sdk-messages",
-  defaultCapabilities: claudeCodeCapabilities,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeClaudeDelegatedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: claudeCodeCapabilities,
-      runtimeId: "claude-sdk",
-      apiFamily: "claude-agent-sdk"
-    });
-  }
-};
-var claude_sdk_default = claudeSdkRuntime;
-
-// src/domains/providers/runtimes/cloud/alcf.ts
-init_esm_shims();
-
-// src/domains/providers/probe/http.ts
-init_esm_shims();
-import { performance as performance2 } from "node:perf_hooks";
-async function probeHttp(opts) {
-  const { response, latencyMs, error } = await runFetch(opts);
-  if (response === null) return { ok: false, error: error ?? "unknown transport error", latencyMs };
-  const method = opts.method ?? "GET";
-  if (response.ok) return { ok: true, latencyMs };
-  if (method === "HEAD" && response.status === 405) return { ok: true, latencyMs };
-  return {
-    ok: false,
-    latencyMs,
-    error: `HTTP ${response.status}: ${response.statusText}`
-  };
-}
-async function probeJson(opts) {
-  const { response, latencyMs, error } = await runFetch(opts);
-  if (response === null) return { ok: false, error: error ?? "unknown transport error", latencyMs };
-  const method = opts.method ?? "GET";
-  if (!response.ok && !(method === "HEAD" && response.status === 405)) {
-    return {
-      ok: false,
-      latencyMs,
-      error: `HTTP ${response.status}: ${response.statusText}`
-    };
-  }
-  let data;
-  try {
-    data = await response.json();
-  } catch (err) {
-    return { ok: false, latencyMs, error: `JSON parse: ${describeError2(err)}` };
-  }
-  return { ok: true, latencyMs, data };
-}
-async function runFetch(opts) {
-  const controller = new AbortController();
-  let timedOut = false;
-  const timer = setTimeout(() => {
-    timedOut = true;
-    controller.abort();
-  }, opts.timeoutMs);
-  const onExternalAbort = () => controller.abort();
-  if (opts.signal) {
-    if (opts.signal.aborted) controller.abort();
-    else opts.signal.addEventListener("abort", onExternalAbort, { once: true });
-  }
-  const init = {
-    method: opts.method ?? "GET",
-    signal: controller.signal
-  };
-  if (opts.headers) init.headers = opts.headers;
-  if (opts.body !== void 0) init.body = opts.body;
-  const started = performance2.now();
-  try {
-    const response = await fetch(opts.url, init);
-    return { response, latencyMs: Math.round(performance2.now() - started) };
-  } catch (err) {
-    const latencyMs = Math.round(performance2.now() - started);
-    if (timedOut) {
-      return { response: null, latencyMs, error: `timeout after ${opts.timeoutMs}ms` };
-    }
-    if (opts.signal?.aborted) {
-      return { response: null, latencyMs, error: "aborted by caller" };
-    }
-    return { response: null, latencyMs, error: describeError2(err) };
-  } finally {
-    clearTimeout(timer);
-    if (opts.signal) opts.signal.removeEventListener("abort", onExternalAbort);
-  }
-}
-function describeError2(err) {
-  if (!(err instanceof Error)) return String(err);
-  const cause = err.cause;
-  if (cause instanceof Error && cause.message.length > 0) {
-    const code = cause.code;
-    return typeof code === "string" && !cause.message.includes(code) ? `${cause.message} (${code})` : cause.message;
-  }
-  return err.message;
-}
-
-// src/domains/providers/runtimes/protocol/openai-compat.ts
-init_esm_shims();
-
-// src/core/context-floor.ts
-init_esm_shims();
-var CLIO_MIN_CONTEXT_WINDOW = 131072;
-var CLIO_MIN_MAX_OUTPUT_TOKENS = 32768;
-var CLIO_CONTEXT_WINDOW_WARN_BELOW = 128e3;
-
-// src/domains/providers/probe/reasoning.ts
-init_esm_shims();
-import { performance as performance3 } from "node:perf_hooks";
-var PROMPT = "What is 2+2? Think briefly, then answer.";
-function nonEmptyString2(value) {
-  return typeof value === "string" && value.trim().length > 0;
-}
-function detectReasoningField(data) {
-  const message = data.choices?.[0]?.message;
-  if (!message) return null;
-  if (nonEmptyString2(message.reasoning_content)) return "reasoning_content";
-  if (nonEmptyString2(message.reasoning)) return "reasoning";
-  if (nonEmptyString2(message.reasoning_text)) return "reasoning_text";
-  return null;
-}
-function trimTrailingSlash(url2) {
-  return url2.endsWith("/") ? url2.slice(0, -1) : url2;
-}
-async function probeOpenAICompatReasoning(opts) {
-  const controller = new AbortController();
-  let timedOut = false;
-  const timer = setTimeout(() => {
-    timedOut = true;
-    controller.abort();
-  }, opts.timeoutMs);
-  const onExternalAbort = () => controller.abort();
-  if (opts.signal) {
-    if (opts.signal.aborted) controller.abort();
-    else opts.signal.addEventListener("abort", onExternalAbort, { once: true });
-  }
-  const headers = { "content-type": "application/json" };
-  if (opts.apiKey && opts.apiKey.length > 0) headers.authorization = `Bearer ${opts.apiKey}`;
-  const body = JSON.stringify({
-    model: opts.modelId,
-    messages: [{ role: "user", content: PROMPT }],
-    max_tokens: 200,
-    temperature: 0,
-    stream: false,
-    reasoning_effort: "low"
-  });
-  const url2 = `${trimTrailingSlash(opts.baseUrl)}/v1/chat/completions`;
-  const started = performance3.now();
-  try {
-    const response = await fetch(url2, {
-      method: "POST",
-      headers,
-      body,
-      signal: controller.signal
-    });
-    const latencyMs = Math.round(performance3.now() - started);
-    if (!response.ok) {
-      return { reasoning: false, latencyMs, error: `HTTP ${response.status}: ${response.statusText}` };
-    }
-    const data = await response.json();
-    const field = detectReasoningField(data);
-    if (field) return { reasoning: true, field, latencyMs };
-    return { reasoning: false, latencyMs };
-  } catch (err) {
-    const latencyMs = Math.round(performance3.now() - started);
-    if (timedOut) return { reasoning: false, latencyMs, error: `timeout after ${opts.timeoutMs}ms` };
-    if (opts.signal?.aborted) return { reasoning: false, latencyMs, error: "aborted by caller" };
-    return { reasoning: false, latencyMs, error: err instanceof Error ? err.message : String(err) };
-  } finally {
-    clearTimeout(timer);
-    if (opts.signal) opts.signal.removeEventListener("abort", onExternalAbort);
-  }
-}
-
-// src/domains/providers/runtimes/common/probe-helpers.ts
-init_esm_shims();
-
-// src/domains/providers/types/context-window-slots.ts
-init_esm_shims();
-function formatContextWindowSlots(contextWindow, slots) {
-  const format = (n) => Math.round(n).toLocaleString("en-US");
-  return `${format(contextWindow)} (${format(slots.totalContextSize)} / ${slots.slots} slots)`;
-}
-
-// src/domains/providers/runtimes/common/probe-helpers.ts
-async function probeUrl(url2, ctx, method = "GET") {
-  const base = { url: url2, timeoutMs: ctx.httpTimeoutMs, method };
-  return ctx.signal ? probeHttp({ ...base, signal: ctx.signal }) : probeHttp(base);
-}
-async function probeOpenAIModels(base, ctx, modelsPath = "/v1/models") {
-  return (await probeOpenAIModelCatalog(base, ctx, modelsPath)).models;
-}
-var OPENAI_COMPAT_DETAIL_PATHS = ["/api/v0/models"];
-async function probeModelDetailRows(base, ctx, modelsPath) {
-  const rows = /* @__PURE__ */ new Map();
-  for (const detailPath of OPENAI_COMPAT_DETAIL_PATHS) {
-    if (detailPath === modelsPath) continue;
-    const opts = { url: `${base}${detailPath}`, timeoutMs: ctx.httpTimeoutMs };
-    const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-    if (!result.ok || !Array.isArray(result.data?.data)) continue;
-    for (const row of result.data.data) {
-      if (typeof row?.id !== "string" || row.id.length === 0) continue;
-      if (!rows.has(row.id)) rows.set(row.id, row);
-    }
-    if (rows.size > 0) break;
-  }
-  return rows;
-}
-async function probeOpenAIModelCatalog(base, ctx, modelsPath = "/v1/models") {
-  const opts = { url: `${base}${modelsPath}`, timeoutMs: ctx.httpTimeoutMs };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok || !result.data?.data) return { models: [], modelCapabilities: {}, modelStates: {} };
-  const detail = await probeModelDetailRows(base, ctx, modelsPath);
-  const models = [];
-  const modelCapabilities = {};
-  const modelStates = {};
-  for (const row of result.data.data) {
-    if (typeof row?.id !== "string" || row.id.length === 0) continue;
-    models.push(row.id);
-    const detailRow = detail.get(row.id);
-    const caps = {
-      ...detailRow ? capabilitiesFromOpenAIModelEntry(detailRow) : {},
-      ...capabilitiesFromOpenAIModelEntry(row)
-    };
-    if (Object.keys(caps).length > 0) modelCapabilities[row.id] = caps;
-    const loadedContext = loadedContextFromEntry(row) ?? (detailRow ? loadedContextFromEntry(detailRow) : void 0);
-    const state = modelStateFromOpenAIModelEntry(row) ?? (detailRow ? modelStateFromOpenAIModelEntry(detailRow) : void 0) ?? // A reported loaded context is itself the residency answer: nothing
-    // serves a window for a model it has not loaded.
-    (loadedContext !== void 0 ? { state: "loaded" } : void 0);
-    const contextSlots = contextSlotsFromEntry(row) ?? (detailRow ? contextSlotsFromEntry(detailRow) : void 0);
-    const withSlots = contextSlots ? { ...state ?? { state: "unknown" }, contextSlots } : state;
-    if (withSlots) {
-      modelStates[row.id] = loadedContext === void 0 ? withSlots : { ...withSlots, contextLength: loadedContext };
-    }
-  }
-  return { models, modelCapabilities, modelStates };
-}
-function isRecord3(value) {
-  return value !== null && typeof value === "object" && !Array.isArray(value);
-}
-function positiveNumber2(value) {
-  return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : void 0;
-}
-function firstPositiveNumber(record, keys) {
-  for (const key of keys) {
-    const value = positiveNumber2(record[key]);
-    if (value !== void 0) return value;
-  }
-  return void 0;
-}
-function capabilityListFlag(record, name) {
-  const list = record.capabilities;
-  if (!Array.isArray(list)) return void 0;
-  return list.some((entry) => entry === name) ? true : void 0;
-}
-function booleanFromAny(record, keys) {
-  for (const key of keys) {
-    const value = record[key];
-    if (typeof value === "boolean") return value;
-  }
-  return void 0;
-}
-function nestedRecord(record, key) {
-  const value = record[key];
-  return isRecord3(value) ? value : null;
-}
-function firstString(record, keys) {
-  if (!record) return void 0;
-  for (const key of keys) {
-    const value = record[key];
-    if (typeof value === "string" && value.trim().length > 0) return value.trim();
-  }
-  return void 0;
-}
-function firstBoolean(record, keys) {
-  if (!record) return void 0;
-  for (const key of keys) {
-    const value = record[key];
-    if (typeof value === "boolean") return value;
-  }
-  return void 0;
-}
-function normalizeModelState(raw) {
-  const value = raw?.trim().toLowerCase().replace(/[\s_]+/g, "-");
-  if (!value) return void 0;
-  if (value === "loaded" || value === "ready" || value === "running" || value === "active") return "loaded";
-  if (value === "loading" || value === "pending" || value === "queued" || value === "starting") return "loading";
-  if (value === "unloaded" || value === "not-loaded" || value === "idle" || value === "sleeping" || value === "stopped") {
-    return "unloaded";
-  }
-  if (value === "failed" || value === "error" || value === "errored") return "failed";
-  if (value === "unknown") return "unknown";
-  return void 0;
-}
-function modelStateFromOpenAIModelEntry(row) {
-  const status = nestedRecord(row, "status");
-  const failed = firstBoolean(status, ["failed"]) ?? firstBoolean(row, ["failed"]);
-  if (failed === true) {
-    const detail = firstString(status, ["error", "reason", "message"]) ?? firstString(row, ["error", "reason", "message"]);
-    return detail ? { state: "failed", detail } : { state: "failed" };
-  }
-  const raw = typeof row.status === "string" ? row.status : firstString(status, ["value", "state", "status"]) ?? firstString(row, ["state"]);
-  const normalized = normalizeModelState(raw);
-  if (normalized) {
-    const detail = firstString(status, ["detail", "message", "reason"]);
-    return detail ? { state: normalized, detail } : { state: normalized };
-  }
-  const loaded = firstBoolean(status, ["loaded"]) ?? firstBoolean(row, ["loaded"]);
-  if (loaded === true) return { state: "loaded" };
-  if (loaded === false) return { state: "unloaded" };
-  return void 0;
-}
-function loadedContextFromEntry(row) {
-  const reported = firstPositiveNumber(row, ["loaded_context_length", "loadedContextLength"]);
-  return reported === void 0 ? void 0 : Math.floor(reported);
-}
-function statusArgsFromEntry(row) {
-  const status = nestedRecord(row, "status");
-  return argsFromStatus(status);
-}
-function contextSlotsFromEntry(row) {
-  return llamaCppRequestContextWindow(parseLlamaCppServerFlags(statusArgsFromEntry(row)))?.slots;
-}
-function capabilitiesFromOpenAIModelEntry(row) {
-  const caps = {};
-  const meta = nestedRecord(row, "meta");
-  const flags = parseLlamaCppServerFlags(statusArgsFromEntry(row));
-  const contextWindow = llamaCppRequestContextWindow(flags)?.contextWindow ?? firstPositiveNumber(row, [
-    // What is actually loaded outranks what the model could support: a
-    // model served at 8k out of a possible 262k has an 8k window today,
-    // and the run has to be planned against the real one.
-    "loaded_context_length",
-    "loadedContextLength",
-    "context_window",
-    "contextWindow",
-    "context_length",
-    "contextLength",
-    "max_context_length",
-    "maxContextLength",
-    "n_ctx"
-  ]) ?? (meta ? firstPositiveNumber(meta, ["n_ctx", "n_ctx_train", "context_length", "contextWindow"]) : void 0);
-  if (contextWindow !== void 0) caps.contextWindow = Math.floor(contextWindow);
-  const maxTokens = positiveNumber2(flags.maxTokens) ?? firstPositiveNumber(row, [
-    "max_output_tokens",
-    "maxOutputTokens",
-    "max_completion_tokens",
-    "maxCompletionTokens",
-    "max_tokens",
-    "maxTokens",
-    "n_predict"
-  ]);
-  if (maxTokens !== void 0) caps.maxTokens = Math.floor(maxTokens);
-  const tools = flags.jinja ?? booleanFromAny(row, ["tools", "tool_use", "toolUse", "trained_for_tool_use"]) ?? capabilityListFlag(row, "tool_use");
-  if (tools !== void 0) caps.tools = tools;
-  const reasoning = flags.reasoning ?? (flags.reasoningBudget !== void 0 ? true : void 0) ?? booleanFromAny(row, ["reasoning", "thinking"]);
-  if (reasoning !== void 0) caps.reasoning = reasoning;
-  const architecture = nestedRecord(row, "architecture");
-  const architectureInput = architecture?.input_modalities;
-  const modalities = Array.isArray(row.modalities) ? row.modalities : Array.isArray(row.input) ? row.input : Array.isArray(architectureInput) ? architectureInput : null;
-  if (modalities) {
-    caps.vision = modalities.some((entry) => entry === "image" || entry === "vision");
-    if (modalities.some((entry) => entry === "audio")) caps.audio = true;
-  }
-  return caps;
-}
-function modelEntries(payload) {
-  if (!Array.isArray(payload?.data)) return [];
-  const out = [];
-  for (const row of payload.data) {
-    if (typeof row?.id !== "string" || row.id.length === 0) continue;
-    out.push({ id: row.id, status: row.status });
-  }
-  return out;
-}
-async function probeOpenAIModelEntries(base, ctx) {
-  const opts = { url: `${base}/v1/models`, timeoutMs: ctx.httpTimeoutMs };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok) return [];
-  return modelEntries(result.data);
-}
-function llamaCppRequestContextWindow(flags) {
-  const total = positiveNumber2(flags.contextSize);
-  if (total === void 0) return void 0;
-  const parallel = positiveNumber2(flags.parallel);
-  if (parallel === void 0 || parallel <= 1 || flags.kvUnified === true) return { contextWindow: Math.floor(total) };
-  const slots = Math.floor(parallel);
-  return {
-    contextWindow: Math.floor(total / slots),
-    slots: { totalContextSize: Math.floor(total), slots }
-  };
-}
-function argsFromStatus(status) {
-  if (!isRecord3(status)) return [];
-  const args = status.args;
-  if (Array.isArray(args)) return args.filter((entry) => typeof entry === "string");
-  if (typeof args === "string") return args.trim().split(/\s+/).filter(Boolean);
-  return [];
-}
-function looksLikeFlag(token) {
-  return token.startsWith("-") && Number.isNaN(Number(token));
-}
-function valueAfter(args, ...flags) {
-  for (const flag of flags) {
-    const index = args.indexOf(flag);
-    if (index < 0) continue;
-    const value = args[index + 1];
-    return value && !looksLikeFlag(value) ? value : void 0;
-  }
-  return void 0;
-}
-function numberFlag(args, ...flags) {
-  const value = valueAfter(args, ...flags);
-  if (value === void 0) return void 0;
-  const parsed = Number(value);
-  return Number.isFinite(parsed) ? parsed : void 0;
-}
-function booleanFlag(args, ...flags) {
-  const flag = flags.find((candidate) => args.includes(candidate));
-  if (flag === void 0) return void 0;
-  const value = valueAfter(args, flag);
-  if (value === void 0) return true;
-  const normalized = value.toLowerCase();
-  if (normalized === "true" || normalized === "on" || normalized === "1") return true;
-  if (normalized === "false" || normalized === "off" || normalized === "0") return false;
-  return void 0;
-}
-function parseLlamaCppServerFlags(args) {
-  const flags = {};
-  const ctxSize = numberFlag(args, "--ctx-size", "-c");
-  if (ctxSize !== void 0) flags.contextSize = ctxSize;
-  const maxTokens = numberFlag(args, "--n-predict");
-  if (maxTokens !== void 0) flags.maxTokens = maxTokens;
-  const flashAttention = booleanFlag(args, "--flash-attn");
-  if (flashAttention !== void 0) flags.flashAttention = flashAttention;
-  const jinja = booleanFlag(args, "--jinja");
-  if (jinja !== void 0) flags.jinja = jinja;
-  const reasoningRaw = valueAfter(args, "--reasoning");
-  if (reasoningRaw) flags.reasoning = reasoningRaw === "on" || reasoningRaw === "true" || reasoningRaw === "1";
-  const reasoningBudget = numberFlag(args, "--reasoning-budget");
-  if (reasoningBudget !== void 0) flags.reasoningBudget = reasoningBudget;
-  const temperature = numberFlag(args, "--temperature");
-  if (temperature !== void 0) flags.temperature = temperature;
-  const topP = numberFlag(args, "--top-p");
-  if (topP !== void 0) flags.topP = topP;
-  const topK = numberFlag(args, "--top-k");
-  if (topK !== void 0) flags.topK = topK;
-  const nGpuLayers = numberFlag(args, "--n-gpu-layers");
-  if (nGpuLayers !== void 0) flags.nGpuLayers = nGpuLayers;
-  const parallel = numberFlag(args, "--parallel", "-np");
-  if (parallel !== void 0) flags.parallel = parallel;
-  const kvUnifiedAt = Math.max(args.lastIndexOf("--kv-unified"), args.lastIndexOf("-kvu"));
-  const noKvUnifiedAt = args.lastIndexOf("--no-kv-unified");
-  if (noKvUnifiedAt > kvUnifiedAt) flags.kvUnified = false;
-  else if (kvUnifiedAt >= 0) flags.kvUnified = booleanFlag(args, "--kv-unified", "-kvu") ?? true;
-  const cacheTypeK = valueAfter(args, "--cache-type-k");
-  if (cacheTypeK) flags.cacheTypeK = cacheTypeK;
-  const cacheTypeV = valueAfter(args, "--cache-type-v");
-  if (cacheTypeV) flags.cacheTypeV = cacheTypeV;
-  const mmproj = valueAfter(args, "--mmproj");
-  if (mmproj) flags.mmproj = mmproj;
-  const chatTemplateKwargs = valueAfter(args, "--chat-template-kwargs");
-  if (chatTemplateKwargs) flags.chatTemplateKwargs = chatTemplateKwargs;
-  return flags;
-}
-function selectedModelEntry(entries, target) {
-  const expected = target.defaultModel?.trim();
-  if (expected) return entries.find((entry) => entry.id === expected) ?? null;
-  return entries[0] ?? null;
-}
-function statusNotes(id, status) {
-  if (!isRecord3(status)) return [];
-  const notes = [];
-  if (status.failed === true) notes.push(`llama.cpp router marks ${id} as failed`);
-  const state = typeof status.state === "string" ? status.state : void 0;
-  if (state === "loading") notes.push(`llama.cpp router reports ${id} is still loading`);
-  return notes;
-}
-async function probeLlamaCppModelStatus(base, target, ctx) {
-  const entries = await probeOpenAIModelEntries(base, ctx);
-  const selected = selectedModelEntry(entries, target);
-  if (!selected) return {};
-  const args = argsFromStatus(selected.status);
-  if (args.length === 0) return { notes: statusNotes(selected.id, selected.status) };
-  const flags = parseLlamaCppServerFlags(args);
-  const caps = {};
-  const window = llamaCppRequestContextWindow(flags);
-  if (window !== void 0) caps.contextWindow = window.contextWindow;
-  if (flags.maxTokens !== void 0 && flags.maxTokens > 0) caps.maxTokens = flags.maxTokens;
-  if (flags.reasoning === true || flags.reasoningBudget !== void 0) caps.reasoning = true;
-  if (flags.mmproj) caps.vision = true;
-  if (flags.jinja === true) caps.tools = true;
-  const enrichment = { modelId: selected.id, serverFlags: flags };
-  if (Object.keys(caps).length > 0) enrichment.discoveredCapabilities = caps;
-  const notes = statusNotes(selected.id, selected.status);
-  if (window?.slots) {
-    notes.push(
-      `${selected.id} context window ${formatContextWindowSlots(window.contextWindow, window.slots)}: --ctx-size is split across --parallel slots without --kv-unified`
-    );
-  }
-  if (notes.length > 0) enrichment.notes = notes;
-  return enrichment;
-}
-async function detectModelMismatch(base, target, ctx) {
-  const expected = target.defaultModel?.trim();
-  if (!expected) return null;
-  const ids = await probeOpenAIModels(base, ctx);
-  if (ids.length === 0) return null;
-  if (ids.includes(expected)) return null;
-  const loaded = ids[0] ?? "(unknown)";
-  return `wire model id ${expected} does not match server's loaded model ${loaded}; llama.cpp serves a single fixed model`;
-}
-async function probeLlamaCppProps(base, ctx) {
-  const opts = { url: `${base}/props`, timeoutMs: ctx.httpTimeoutMs };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok || !result.data) return {};
-  const data = result.data;
-  const enrichment = {};
-  const caps = {};
-  const nCtx = data.default_generation_settings?.n_ctx;
-  if (typeof nCtx === "number" && nCtx > 0) caps.contextWindow = nCtx;
-  const nPredict = data.default_generation_settings?.n_predict;
-  if (typeof nPredict === "number" && nPredict > 0) caps.maxTokens = nPredict;
-  const vision = data.modalities?.vision;
-  if (typeof vision === "boolean") caps.vision = vision;
-  if (Object.keys(caps).length > 0) enrichment.discoveredCapabilities = caps;
-  if (typeof data.build_info === "string" && data.build_info.length > 0) {
-    enrichment.serverVersion = data.build_info;
-  }
-  return enrichment;
-}
-
-// src/domains/providers/runtimes/protocol/openai-compat.ts
-function synthesizeOpenAICompatModel(input) {
-  return synthLocalModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    apiFamily: input.apiFamily ?? "openai-completions",
-    provider: input.provider,
-    baseUrlForTarget: input.baseUrlForTarget ?? withV1
-  });
-}
-function makeOpenAICompatRuntime(spec) {
-  const apiFamily = spec.apiFamily ?? "openai-completions";
-  const asIs = spec.baseUrlStyle === "asIs";
-  const baseUrlForTarget = asIs ? withAsIs : withV1;
-  const healthPath = spec.healthPath ?? "/v1/models";
-  const modelsPath = spec.modelsPath ?? "/v1/models";
-  const probeBase = asIs ? targetBaseUrl : targetRootUrl;
-  return {
-    id: spec.id,
-    displayName: spec.displayName,
-    kind: "http",
-    tier: spec.tier,
-    apiFamily,
-    auth: spec.auth,
-    defaultCapabilities: spec.defaultCapabilities,
-    async probe(target, ctx) {
-      const base = probeBase(target);
-      if (!base) return { ok: false, error: "target has no url" };
-      const health = await probeUrl(`${base}${healthPath}`, ctx);
-      if (!health.ok) return health;
-      const catalog = await probeOpenAIModelCatalog(base, ctx, modelsPath);
-      const result = { ...health };
-      if (catalog.models.length > 0) result.models = catalog.models;
-      if (Object.keys(catalog.modelStates).length > 0) result.modelStates = catalog.modelStates;
-      if (Object.keys(catalog.modelCapabilities).length > 0) {
-        result.modelCapabilities = catalog.modelCapabilities;
-        const selected = target.defaultModel?.trim();
-        const selectedCaps = selected ? catalog.modelCapabilities[selected] : void 0;
-        if (selected && selectedCaps) {
-          result.discoveredCapabilities = selectedCaps;
-          result.capabilityModelId = selected;
-        }
-      }
-      return result;
-    },
-    async probeModels(target, ctx) {
-      const base = probeBase(target);
-      if (!base) return [];
-      return probeOpenAIModels(base, ctx, modelsPath);
-    },
-    async probeReasoning(target, modelId, ctx) {
-      const base = probeBase(target);
-      if (!base) return { reasoning: false, latencyMs: 0, error: "target has no url" };
-      const apiKeyEnv = target.auth?.apiKeyEnvVar;
-      const apiKey = apiKeyEnv && ctx.credentialsPresent.has(apiKeyEnv) ? process.env[apiKeyEnv] : void 0;
-      const probeOpts = {
-        baseUrl: base,
-        modelId,
-        timeoutMs: Math.max(ctx.httpTimeoutMs, 8e3)
-      };
-      if (apiKey) probeOpts.apiKey = apiKey;
-      if (ctx.signal) probeOpts.signal = ctx.signal;
-      return probeOpenAICompatReasoning(probeOpts);
-    },
-    synthesizeModel(target, wireModelId, kb) {
-      return synthesizeOpenAICompatModel({
-        target,
-        wireModelId,
-        kb,
-        defaultCapabilities: spec.defaultCapabilities,
-        apiFamily,
-        provider: spec.provider,
-        baseUrlForTarget
-      });
-    }
-  };
-}
-var defaultCapabilities = {
-  chat: true,
-  tools: false,
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var openai_compat_default = makeOpenAICompatRuntime({
-  id: "openai-compat",
-  displayName: "Generic OpenAI-compatible",
-  provider: "openai-compat",
-  auth: "api-key",
-  tier: "protocol",
-  defaultCapabilities
-});
-
-// src/domains/providers/runtimes/cloud/alcf.ts
-var CATALOG_URL = "https://inference-api.alcf.anl.gov/resource_server/list-endpoints";
-var GATEWAY_ORIGIN = "https://inference-api.alcf.anl.gov/resource_server";
-var defaultCapabilities2 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 32768,
-  maxTokens: 4096
-};
-var KNOWN_MODELS = [
-  "openai/gpt-oss-120b",
-  "openai/gpt-oss-20b",
-  "gpt-oss-120b",
-  "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
-  "meta-llama/Llama-4-Scout-17B-16E-Instruct"
-];
-function clusterFromUrl(url2) {
-  if (!url2) return null;
-  const match = /\/resource_server\/([^/]+)\//.exec(url2);
-  return match ? match[1] ?? null : null;
-}
-function frameworkForCluster(cluster) {
-  return cluster === "metis" ? "api" : "vllm";
-}
-function catalogModels(payload, cluster, framework) {
-  const models = payload.clusters?.[cluster]?.frameworks?.[framework]?.models;
-  if (!Array.isArray(models)) return [];
-  return models.map((model) => String(model).trim()).filter((model) => model.length > 0);
-}
-function runningModels(payload) {
-  const out = [];
-  for (const job of payload.running ?? []) {
-    for (const raw of String(job.Models ?? "").split(",")) {
-      const id = raw.trim();
-      if (id.length > 0) out.push(id);
-    }
-  }
-  return out;
-}
-function dedupe(ids) {
-  const seen = /* @__PURE__ */ new Set();
-  const out = [];
-  for (const id of ids) {
-    if (seen.has(id)) continue;
-    seen.add(id);
-    out.push(id);
-  }
-  return out;
-}
-function authHeaders(ctx) {
-  return { Authorization: `Bearer ${ctx.authToken}`, Accept: "application/json" };
-}
-async function discover(target, ctx) {
-  if (!target.url) return { ok: false, error: "ALCF target has no url" };
-  const cluster = clusterFromUrl(target.url);
-  if (!cluster) {
-    return { ok: false, error: `cannot determine ALCF cluster from url ${target.url}` };
-  }
-  if (!ctx.authToken) {
-    return { ok: false, error: "ALCF requires Globus auth; run `clio-coder auth login alcf`." };
-  }
-  const framework = frameworkForCluster(cluster);
-  const headers = authHeaders(ctx);
-  const catalog = await probeJson({
-    url: CATALOG_URL,
-    headers,
-    timeoutMs: ctx.httpTimeoutMs,
-    ...ctx.signal ? { signal: ctx.signal } : {}
-  });
-  if (!catalog.ok || !catalog.data) {
-    const reason = catalog.error ?? "unknown error";
-    const hint = reason.includes("401") ? " (token expired or rejected; run `clio-coder auth login alcf` again)" : "";
-    return {
-      ok: false,
-      error: `ALCF endpoint catalog unreachable: ${reason}${hint}`,
-      ...catalog.latencyMs !== void 0 ? { latencyMs: catalog.latencyMs } : {}
-    };
-  }
-  const fromCatalog = catalogModels(catalog.data, cluster, framework);
-  let live = [];
-  try {
-    const jobs = await probeJson({
-      url: `${GATEWAY_ORIGIN}/${cluster}/jobs`,
-      headers,
-      timeoutMs: Math.min(ctx.httpTimeoutMs, 12e3),
-      ...ctx.signal ? { signal: ctx.signal } : {}
-    });
-    if (jobs.ok && jobs.data) live = runningModels(jobs.data);
-  } catch {
-  }
-  const models = dedupe([...fromCatalog, ...live]);
-  const result = { ok: true, models };
-  if (catalog.latencyMs !== void 0) result.latencyMs = catalog.latencyMs;
-  if (models.length === 0) {
-    result.notes = [`ALCF ${cluster}/${framework} reported no models; the gateway may have no running jobs.`];
-  }
-  return result;
-}
-var alcfRuntime = {
-  id: "alcf",
-  displayName: "ALCF Inference (Globus)",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-completions",
-  auth: "oauth",
-  knownModels: KNOWN_MODELS,
-  defaultCapabilities: defaultCapabilities2,
-  probe(target, ctx) {
-    return discover(target, ctx);
-  },
-  async probeModels(target, ctx) {
-    const result = await discover(target, ctx);
-    return result.models ?? [];
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    const model = synthesizeOpenAICompatModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities2,
-      apiFamily: "openai-completions",
-      provider: "alcf",
-      baseUrlForTarget: withAsIs
-    });
-    const meta = model.clio;
-    if (meta) meta.chatTemplateKwargsUnsupported = true;
-    return model;
-  }
-};
-var alcf_default = alcfRuntime;
-
-// src/domains/providers/runtimes/cloud/anthropic.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/protocol/anthropic-messages.ts
-init_esm_shims();
-function synthesizeAnthropicMessagesModel(input) {
-  return synthesizeCatalogBackedModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    runtimeId: "anthropic",
-    api: "anthropic-messages",
-    provider: input.provider,
-    defaultBaseUrl: input.defaultBaseUrl
-  });
-}
-
-// src/domains/providers/runtimes/cloud/anthropic.ts
-var defaultCapabilities3 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: true,
-  thinkingFormat: "anthropic-extended",
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 2e5,
-  maxTokens: 8192
-};
-var anthropicRuntime = {
-  id: "anthropic",
-  displayName: "Anthropic",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "anthropic-messages",
-  auth: "api-key",
-  credentialsEnvVar: "ANTHROPIC_API_KEY",
-  defaultCapabilities: defaultCapabilities3,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeAnthropicMessagesModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities3,
-      provider: "anthropic",
-      defaultBaseUrl: "https://api.anthropic.com"
-    });
-  }
-};
-var anthropic_default = anthropicRuntime;
-
-// src/domains/providers/runtimes/cloud/anthropic-max.ts
-init_esm_shims();
-var defaultCapabilities4 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: true,
-  thinkingFormat: "anthropic-extended",
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 2e5,
-  maxTokens: 8192
-};
-var anthropicMaxRuntime = {
-  id: "anthropic-max",
-  displayName: "Anthropic (Claude Pro/Max)",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "anthropic-messages",
-  auth: "oauth",
-  oauthProviderId: "anthropic",
-  authNotice: "Connects with your Claude Pro/Max subscription via OAuth (the same path Claude Code uses). Using subscription credentials outside Anthropic's first-party apps may not align with their terms of service; enable at your own discretion.",
-  defaultCapabilities: defaultCapabilities4,
-  async probeModels(_target, _ctx) {
-    return listCatalogModelsForRuntime("anthropic-max").map((model) => model.id);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities4,
-      runtimeId: "anthropic-max",
-      api: "anthropic-messages",
-      provider: "anthropic",
-      defaultBaseUrl: "https://api.anthropic.com"
-    });
-  }
-};
-var anthropic_max_default = anthropicMaxRuntime;
-
-// src/domains/providers/runtimes/cloud/bedrock.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/protocol/bedrock.ts
-init_esm_shims();
-function synthesizeBedrockModel(input) {
-  return synthesizeCatalogBackedModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    runtimeId: "bedrock",
-    api: "bedrock-converse-stream",
-    provider: "amazon-bedrock",
-    defaultBaseUrl: ""
-  });
-}
-
-// src/domains/providers/runtimes/cloud/bedrock.ts
-var defaultCapabilities5 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: true,
-  thinkingFormat: "anthropic-extended",
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 2e5,
-  maxTokens: 8192
-};
-var bedrockRuntime = {
-  id: "bedrock",
-  displayName: "Amazon Bedrock",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "bedrock-converse-stream",
-  auth: "aws-sdk",
-  defaultCapabilities: defaultCapabilities5,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeBedrockModel({ target, wireModelId, kb, defaultCapabilities: defaultCapabilities5 });
-  }
-};
-var bedrock_default = bedrockRuntime;
-
-// src/domains/providers/runtimes/cloud/deepseek.ts
-init_esm_shims();
-var defaultCapabilities6 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  thinkingFormat: "deepseek-r1",
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 128e3,
-  maxTokens: 65536
-};
-var deepseekRuntime = {
-  id: "deepseek",
-  displayName: "DeepSeek",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  credentialsEnvVar: "DEEPSEEK_API_KEY",
-  defaultCapabilities: defaultCapabilities6,
-  async probeModels(_target, _ctx) {
-    return listCatalogModelsForRuntime("deepseek").map((model) => model.id);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities6,
-      runtimeId: "deepseek",
-      api: "openai-completions",
-      provider: "deepseek",
-      defaultBaseUrl: "https://api.deepseek.com"
-    });
-  }
-};
-var deepseek_default = deepseekRuntime;
-
-// src/domains/providers/runtimes/cloud/google.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/protocol/google.ts
-init_esm_shims();
-function synthesizeGoogleModel(input) {
-  return synthesizeCatalogBackedModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    runtimeId: "google",
-    api: "google-generative-ai",
-    provider: "google",
-    defaultBaseUrl: input.defaultBaseUrl
-  });
-}
-
-// src/domains/providers/runtimes/cloud/google.ts
-var defaultCapabilities7 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 2e6,
-  maxTokens: 8192
-};
-var googleRuntime = {
-  id: "google",
-  displayName: "Google Generative AI",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "google-generative-ai",
-  auth: "api-key",
-  credentialsEnvVar: "GOOGLE_API_KEY",
-  defaultCapabilities: defaultCapabilities7,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeGoogleModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities7,
-      defaultBaseUrl: "https://generativelanguage.googleapis.com/v1beta"
-    });
-  }
-};
-var google_default = googleRuntime;
-
-// src/domains/providers/runtimes/cloud/groq.ts
-init_esm_shims();
-var defaultCapabilities8 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 128e3,
-  maxTokens: 8192
-};
-var groqRuntime = {
-  id: "groq",
-  displayName: "Groq",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  credentialsEnvVar: "GROQ_API_KEY",
-  defaultCapabilities: defaultCapabilities8,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities8,
-      runtimeId: "groq",
-      api: "openai-completions",
-      provider: "groq",
-      defaultBaseUrl: "https://api.groq.com/openai/v1"
-    });
-  }
-};
-var groq_default = groqRuntime;
-
-// src/domains/providers/runtimes/cloud/mistral.ts
-init_esm_shims();
-var defaultCapabilities9 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "mistral",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 128e3,
-  maxTokens: 8192
-};
-var mistralRuntime = {
-  id: "mistral",
-  displayName: "Mistral AI",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "mistral-conversations",
-  auth: "api-key",
-  credentialsEnvVar: "MISTRAL_API_KEY",
-  defaultCapabilities: defaultCapabilities9,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities9,
-      runtimeId: "mistral",
-      api: "mistral-conversations",
-      provider: "mistral",
-      defaultBaseUrl: "https://api.mistral.ai"
-    });
-  }
-};
-var mistral_default = mistralRuntime;
-
-// src/domains/providers/runtimes/cloud/openai.ts
-init_esm_shims();
-var defaultCapabilities10 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 272e3,
-  maxTokens: 16384
-};
-var openaiRuntime = {
-  id: "openai",
-  displayName: "OpenAI",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-responses",
-  auth: "api-key",
-  credentialsEnvVar: "OPENAI_API_KEY",
-  defaultCapabilities: defaultCapabilities10,
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities10,
-      runtimeId: "openai",
-      api: "openai-responses",
-      provider: "openai",
-      defaultBaseUrl: "https://api.openai.com/v1"
-    });
-  }
-};
-var openai_default = openaiRuntime;
-
-// src/domains/providers/runtimes/cloud/openai-codex.ts
-init_esm_shims();
-var defaultCapabilities11 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: true,
-  thinkingFormat: "openai-codex",
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 272e3,
-  maxTokens: 16384
-};
-var openaiCodexRuntime = {
-  id: "openai-codex",
-  displayName: "OpenAI Codex",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-codex-responses",
-  auth: "oauth",
-  defaultCapabilities: defaultCapabilities11,
-  async probeModels(_target, _ctx) {
-    return listCatalogModelsForRuntime("openai-codex").map((model) => model.id);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities11,
-      runtimeId: "openai-codex",
-      api: "openai-codex-responses",
-      provider: "openai-codex",
-      defaultBaseUrl: "https://chatgpt.com/backend-api"
-    });
-  }
-};
-var openai_codex_default = openaiCodexRuntime;
-
-// src/domains/providers/runtimes/cloud/openrouter.ts
-init_esm_shims();
-var OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1";
-var OPENROUTER_HEADERS = {
-  "HTTP-Referer": "https://github.com/iowarp/clio-coder",
-  "X-OpenRouter-Title": "Clio Coder"
-};
-var defaultCapabilities12 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  thinkingFormat: "openrouter",
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: 128e3,
-  maxTokens: 8192
-};
-function trimTrailingSlash2(value) {
-  return value.endsWith("/") && value.length > 1 ? value.slice(0, -1) : value;
-}
-function targetBaseUrl2(target) {
-  return trimTrailingSlash2(target.url ?? OPENROUTER_BASE_URL);
-}
-function modelsUrl(target) {
-  return `${targetBaseUrl2(target)}/models`;
-}
-function probeHeaders(target, ctx) {
-  const headers = { ...OPENROUTER_HEADERS, ...target.auth?.headers ?? {} };
-  const envName = target.auth?.apiKeyEnvVar ?? "OPENROUTER_API_KEY";
-  if (ctx.credentialsPresent.has(envName)) {
-    const key = process.env[envName]?.trim();
-    if (key) headers.authorization = `Bearer ${key}`;
-  }
-  return headers;
-}
-async function fetchModels(target, ctx) {
-  const opts = {
-    url: modelsUrl(target),
-    timeoutMs: ctx.httpTimeoutMs,
-    headers: probeHeaders(target, ctx)
-  };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok) return result;
-  const models = (result.data?.data ?? []).map((row) => typeof row?.id === "string" ? row.id : null).filter((id) => id !== null);
-  const out = { ok: true, models };
-  if (result.latencyMs !== void 0) out.latencyMs = result.latencyMs;
-  const configured = target.defaultModel?.trim();
-  if (configured && models.length > 0 && !models.includes(configured)) {
-    out.ok = false;
-    out.error = `configured model '${configured}' was not returned by OpenRouter`;
-  }
-  return out;
-}
-var openrouterRuntime = {
-  id: "openrouter",
-  displayName: "OpenRouter",
-  kind: "http",
-  tier: "cloud",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  credentialsEnvVar: "OPENROUTER_API_KEY",
-  defaultCapabilities: defaultCapabilities12,
-  probe(target, ctx) {
-    return fetchModels(target, ctx);
-  },
-  async probeModels(target, ctx) {
-    const result = await fetchModels(target, ctx);
-    return result.ok && result.models ? [...result.models] : [];
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthesizeCatalogBackedModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities12,
-      runtimeId: "openrouter",
-      api: "openai-completions",
-      provider: "openrouter",
-      defaultBaseUrl: OPENROUTER_BASE_URL,
-      defaultHeaders: OPENROUTER_HEADERS
-    });
-  }
-};
-var openrouter_default = openrouterRuntime;
-
-// src/domains/providers/runtimes/local-native/lemonade-anthropic.ts
-init_esm_shims();
-
-// src/domains/providers/runtimes/protocol/anthropic-compat.ts
-init_esm_shims();
-function synthesizeAnthropicCompatModel(input) {
-  return synthLocalModel({
-    target: input.target,
-    wireModelId: input.wireModelId,
-    kb: input.kb,
-    defaultCapabilities: input.defaultCapabilities,
-    apiFamily: "anthropic-messages",
-    provider: input.provider,
-    baseUrlForTarget: input.baseUrlForTarget ?? withAsIs
-  });
-}
-function makeAnthropicCompatRuntime(spec) {
-  const messagesPath = spec.messagesPath ?? "/v1/messages";
-  const modelsPath = spec.modelsPath ?? "/v1/models";
-  return {
-    id: spec.id,
-    displayName: spec.displayName,
-    kind: "http",
-    tier: spec.tier,
-    apiFamily: "anthropic-messages",
-    auth: spec.auth,
-    defaultCapabilities: spec.defaultCapabilities,
-    ...spec.hidden === true ? { hidden: true } : {},
-    async probe(target, ctx) {
-      const base = targetBaseUrl(target);
-      if (!base) return { ok: false, error: "target has no url" };
-      return probeUrl(`${base}${messagesPath}`, ctx, "HEAD");
-    },
-    async probeModels(target, ctx) {
-      const base = targetBaseUrl(target);
-      if (!base) return [];
-      return probeOpenAIModels(base, ctx, modelsPath);
-    },
-    synthesizeModel(target, wireModelId, kb) {
-      return synthesizeAnthropicCompatModel({
-        target,
-        wireModelId,
-        kb,
-        defaultCapabilities: spec.defaultCapabilities,
-        provider: spec.provider
-      });
-    }
-  };
-}
-var defaultCapabilities13 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var anthropic_compat_default = makeAnthropicCompatRuntime({
-  id: "anthropic-compat",
-  displayName: "Generic Anthropic-compatible",
-  provider: "anthropic-compat",
-  auth: "api-key",
-  tier: "protocol",
-  defaultCapabilities: defaultCapabilities13
-});
-
-// src/domains/providers/runtimes/local-native/lemonade-anthropic.ts
-var defaultCapabilities14 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var lemonade_anthropic_default = makeAnthropicCompatRuntime({
-  id: "lemonade-anthropic",
-  displayName: "Lemonade (Anthropic-compat)",
-  provider: "lemonade",
-  auth: "api-key",
-  tier: "local-native",
-  defaultCapabilities: defaultCapabilities14,
-  hidden: true
-});
-
-// src/domains/providers/runtimes/local-native/lemonade-openai.ts
-init_esm_shims();
-var defaultCapabilities15 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var lemonade_openai_default = makeOpenAICompatRuntime({
-  id: "lemonade",
-  displayName: "Lemonade (OpenAI-compat)",
-  provider: "lemonade",
-  auth: "api-key",
-  tier: "local-native",
-  defaultCapabilities: defaultCapabilities15
-});
-
-// src/domains/providers/runtimes/local-native/llamacpp.ts
-init_esm_shims();
-var defaultCapabilities16 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  structuredOutputs: "json-schema",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-function targetUrl(target) {
-  return targetRootUrl(target);
-}
-var llamacppRuntime = {
-  id: "llamacpp",
-  displayName: "llama.cpp",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities16,
-  async probe(target, ctx) {
-    const base = targetUrl(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
-    const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
-    if (!health.ok) return health;
-    const props = await probeLlamaCppProps(base, ctx);
-    const status = await probeLlamaCppModelStatus(base, target, ctx);
-    const catalog = await probeOpenAIModelCatalog(base, ctx);
-    const result = { ok: true };
-    if (catalog.models.length > 0) result.models = catalog.models;
-    if (Object.keys(catalog.modelCapabilities).length > 0) result.modelCapabilities = catalog.modelCapabilities;
-    if (Object.keys(catalog.modelStates).length > 0) result.modelStates = catalog.modelStates;
-    if (typeof health.latencyMs === "number") result.latencyMs = health.latencyMs;
-    const discoveredCapabilities = {
-      ...props.discoveredCapabilities ?? {},
-      ...status.discoveredCapabilities ?? {}
-    };
-    if (Object.keys(discoveredCapabilities).length > 0) {
-      result.discoveredCapabilities = discoveredCapabilities;
-      if (status.modelId) result.capabilityModelId = status.modelId;
-    }
-    if (props.serverVersion) result.serverVersion = props.serverVersion;
-    const note = await detectModelMismatch(base, target, ctx);
-    const notes = [...status.notes ?? [], ...note ? [note] : []];
-    if (notes.length > 0) result.notes = notes;
-    return result;
-  },
-  async probeModels(target, ctx) {
-    const base = targetUrl(target);
-    if (!base) return [];
-    return probeOpenAIModels(base, ctx);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities16,
-      apiFamily: "openai-completions",
-      provider: "llamacpp",
-      baseUrlForTarget: withV1
-    });
-  }
-};
-var llamacpp_default = llamacppRuntime;
-
-// src/domains/providers/runtimes/local-native/llamacpp-anthropic.ts
-init_esm_shims();
-var defaultCapabilities17 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "anthropic",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: true,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-function url(target) {
-  return target.url ? stripTrailingSlash(target.url) : null;
-}
-var llamacppAnthropicRuntime = {
-  id: "llamacpp-anthropic",
-  displayName: "llama.cpp (Anthropic-compat)",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "anthropic-messages",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities17,
-  hidden: true,
-  async probe(target, ctx) {
-    const base = url(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
-    const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
-    if (!health.ok) return health;
-    const headOpts = {
-      url: `${base}/v1/messages`,
-      method: "HEAD",
-      timeoutMs: ctx.httpTimeoutMs
-    };
-    const head = await (ctx.signal ? probeHttp({ ...headOpts, signal: ctx.signal }) : probeHttp(headOpts));
-    if (!head.ok) return head;
-    const props = await probeLlamaCppProps(base, ctx);
-    const enriched = { ...head };
-    if (props.discoveredCapabilities) enriched.discoveredCapabilities = props.discoveredCapabilities;
-    if (props.serverVersion) enriched.serverVersion = props.serverVersion;
-    const note = await detectModelMismatch(base, target, ctx);
-    if (note) enriched.notes = [note];
-    return enriched;
-  },
-  async probeModels(target, ctx) {
-    const base = url(target);
-    if (!base) return [];
-    const opts = { url: `${base}/v1/models`, timeoutMs: ctx.httpTimeoutMs };
-    const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-    if (!result.ok || !result.data?.data) return [];
-    return result.data.data.map((row) => typeof row?.id === "string" ? row.id : null).filter((id) => id !== null);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities17,
-      apiFamily: "anthropic-messages",
-      provider: "llamacpp",
-      baseUrlForTarget: withAsIs
-    });
-  }
-};
-var llamacpp_anthropic_default = llamacppAnthropicRuntime;
-
-// src/domains/providers/runtimes/local-native/llamacpp-completion.ts
-init_esm_shims();
-var defaultCapabilities18 = {
-  chat: false,
-  tools: false,
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: true,
-  structuredOutputs: "gbnf",
-  contextWindow: 8192,
-  maxTokens: 4096
-};
-function targetUrl2(target) {
-  return targetRootUrl(target);
-}
-function parseChunk(raw) {
-  const chunk = {
-    content: typeof raw.content === "string" ? raw.content : "",
-    stop: raw.stop === true
-  };
-  if (raw.stop_type === "eos" || raw.stop_type === "limit" || raw.stop_type === "word" || raw.stop_type === "none") {
-    chunk.stop_type = raw.stop_type;
-  }
-  if (typeof raw.tokens_predicted === "number") chunk.tokens_predicted = raw.tokens_predicted;
-  if (typeof raw.tokens_evaluated === "number") chunk.tokens_evaluated = raw.tokens_evaluated;
-  return chunk;
-}
-async function* streamSse(body) {
-  const reader = body.getReader();
-  const decoder = new TextDecoder("utf-8");
-  let buffered = "";
-  let droppedFrames = 0;
-  try {
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      buffered += decoder.decode(value, { stream: true });
-      let nl = buffered.indexOf("\n");
-      while (nl !== -1) {
-        const line = buffered.slice(0, nl).trimEnd();
-        buffered = buffered.slice(nl + 1);
-        nl = buffered.indexOf("\n");
-        if (line.length === 0) continue;
-        const payload = line.startsWith("data:") ? line.slice(5).trim() : line;
-        if (payload.length === 0 || payload === "[DONE]") continue;
-        try {
-          const parsed = JSON.parse(payload);
-          const chunk = parseChunk(parsed);
-          yield chunk;
-          if (chunk.stop) return;
-        } catch {
-          droppedFrames += 1;
-        }
-      }
-    }
-    const tail = buffered.trim();
-    if (tail.length > 0) {
-      const payload = tail.startsWith("data:") ? tail.slice(5).trim() : tail;
-      if (payload !== "[DONE]") {
-        try {
-          const parsed = JSON.parse(payload);
-          yield parseChunk(parsed);
-        } catch {
-          droppedFrames += 1;
-        }
-      }
-    }
-  } finally {
-    if (droppedFrames > 0) {
-      process.stderr.write(`[clio:llamacpp] dropped ${droppedFrames} malformed stream frame(s)
-`);
-    }
-    reader.releaseLock();
-  }
-}
-async function postStream(url2, body, signal) {
-  const init = {
-    method: "POST",
-    headers: { "content-type": "application/json", accept: "text/event-stream" },
-    body: JSON.stringify(body)
-  };
-  if (signal) init.signal = signal;
-  const response = await fetch(url2, init);
-  if (!response.ok || !response.body) {
-    throw new Error(`HTTP ${response.status}: ${response.statusText}`);
-  }
-  return response.body;
-}
-function buildCompleteBody(opts) {
-  const body = { prompt: opts.prompt, stream: true };
-  if (opts.n_predict !== void 0) body.n_predict = opts.n_predict;
-  if (opts.stop && opts.stop.length > 0) body.stop = opts.stop;
-  if (opts.grammar) body.grammar = opts.grammar;
-  if (opts.json_schema) body.json_schema = opts.json_schema;
-  if (opts.cache_prompt !== void 0) body.cache_prompt = opts.cache_prompt;
-  return body;
-}
-function buildInfillBody(opts) {
-  const body = buildCompleteBody(opts);
-  body.input_prefix = opts.input_prefix;
-  body.input_suffix = opts.input_suffix;
-  if (opts.input_extra) body.input_extra = opts.input_extra;
-  return body;
-}
-var llamacppCompletionRuntime = {
-  id: "llamacpp-completion",
-  displayName: "llama.cpp (completion / infill)",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities18,
-  hidden: true,
-  async probe(target, ctx) {
-    const base = targetUrl2(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
-    const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
-    if (!health.ok) return health;
-    const props = await probeLlamaCppProps(base, ctx);
-    const status = await probeLlamaCppModelStatus(base, target, ctx);
-    const enriched = { ...health };
-    const discoveredCapabilities = {
-      ...props.discoveredCapabilities ?? {},
-      ...status.discoveredCapabilities ?? {}
-    };
-    if (Object.keys(discoveredCapabilities).length > 0) {
-      enriched.discoveredCapabilities = discoveredCapabilities;
-      if (status.modelId) enriched.capabilityModelId = status.modelId;
-    }
-    if (props.serverVersion) enriched.serverVersion = props.serverVersion;
-    const note = await detectModelMismatch(base, target, ctx);
-    const notes = [...status.notes ?? [], ...note ? [note] : []];
-    if (notes.length > 0) enriched.notes = notes;
-    return enriched;
-  },
-  async probeModels(target, ctx) {
-    const base = targetUrl2(target);
-    if (!base) return [];
-    return probeOpenAIModels(base, ctx);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities18,
-      apiFamily: "openai-completions",
-      provider: "llamacpp",
-      baseUrlForTarget: withV1
-    });
-  },
-  async *complete(target, opts) {
-    const base = targetUrl2(target);
-    if (!base) throw new Error("target has no url");
-    const body = await postStream(`${base}/completion`, buildCompleteBody(opts), opts.signal);
-    for await (const chunk of streamSse(body)) yield chunk;
-  },
-  async *infill(target, opts) {
-    const base = targetUrl2(target);
-    if (!base) throw new Error("target has no url");
-    const body = await postStream(`${base}/infill`, buildInfillBody(opts), opts.signal);
-    for await (const chunk of streamSse(body)) yield chunk;
-  }
-};
-var llamacpp_completion_default = llamacppCompletionRuntime;
-
-// src/domains/providers/runtimes/local-native/llamacpp-embed.ts
-init_esm_shims();
-var defaultCapabilities19 = {
-  chat: false,
-  tools: false,
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: true,
-  rerank: false,
-  fim: false,
-  contextWindow: 8192,
-  maxTokens: 0
-};
-function targetUrl3(target) {
-  return targetRootUrl(target);
-}
-function meanPool(matrix) {
-  if (matrix.length === 0) return [];
-  const dim = matrix[0]?.length ?? 0;
-  const sum = new Array(dim).fill(0);
-  for (const row of matrix) {
-    for (let i = 0; i < dim; i++) sum[i] = (sum[i] ?? 0) + (row[i] ?? 0);
-  }
-  return sum.map((v) => v / matrix.length);
-}
-function flattenNativeEmbedding(entry) {
-  const value = entry.embedding;
-  if (!value || value.length === 0) return [];
-  if (Array.isArray(value[0])) return meanPool(value);
-  return value;
-}
-async function postJson(url2, body, signal) {
-  const init = {
-    method: "POST",
-    headers: { "content-type": "application/json" },
-    body: JSON.stringify(body)
-  };
-  if (signal) init.signal = signal;
-  const response = await fetch(url2, init);
-  if (!response.ok) return { status: response.status, data: null };
-  try {
-    const data = await response.json();
-    return { status: response.status, data };
-  } catch {
-    return { status: response.status, data: null };
-  }
-}
-var llamacppEmbedRuntime = {
-  id: "llamacpp-embed",
-  displayName: "llama.cpp (embeddings)",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities19,
-  hidden: true,
-  async probe(target, ctx) {
-    const base = targetUrl3(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
-    const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
-    if (!health.ok) return health;
-    const probeResponse = await fetch(`${base}/embedding`, {
-      method: "POST",
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({ content: "probe" }),
-      ...ctx.signal ? { signal: ctx.signal } : {}
-    }).catch((err) => {
-      return new Response(null, { status: 599, statusText: String(err) });
-    });
-    if (!probeResponse.ok) {
-      return { ok: false, error: `/embedding not available: HTTP ${probeResponse.status}` };
-    }
-    const props = await probeLlamaCppProps(base, ctx);
-    const result = { ok: true };
-    if (health.latencyMs !== void 0) result.latencyMs = health.latencyMs;
-    if (props.discoveredCapabilities) result.discoveredCapabilities = props.discoveredCapabilities;
-    if (props.serverVersion) result.serverVersion = props.serverVersion;
-    return result;
-  },
-  async probeModels(target, ctx) {
-    const base = targetUrl3(target);
-    if (!base) return [];
-    return probeOpenAIModels(base, ctx);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities19,
-      apiFamily: "openai-completions",
-      provider: "llamacpp",
-      baseUrlForTarget: withV1
-    });
-  },
-  async embed(target, input, ctx) {
-    const base = targetUrl3(target);
-    if (!base) throw new Error("target has no url");
-    const modelId = target.defaultModel ?? "default";
-    const inputs = Array.isArray(input) ? input : [input];
-    const oai = await postJson(
-      `${base}/v1/embeddings`,
-      { input: inputs, model: modelId, encoding_format: "float" },
-      ctx.signal
-    );
-    if (oai.data && Array.isArray(oai.data.data)) {
-      const rows = oai.data.data;
-      const sorted = [...rows].sort((a, b) => (a.index ?? 0) - (b.index ?? 0));
-      const vectors2 = sorted.map((row) => row.embedding ?? []);
-      const tokens = oai.data.usage?.total_tokens ?? oai.data.usage?.prompt_tokens ?? void 0;
-      const dim = vectors2[0]?.length ?? 0;
-      const result = {
-        vectors: vectors2,
-        model: oai.data.model ?? modelId,
-        dimensions: dim
-      };
-      if (tokens !== void 0) result.tokensUsed = tokens;
-      return result;
-    }
-    const probeOpts = { url: `${base}/embedding`, timeoutMs: ctx.httpTimeoutMs };
-    const native = await (ctx.signal ? probeJson({
-      ...probeOpts,
-      method: "POST",
-      body: JSON.stringify({ content: inputs }),
-      headers: { "content-type": "application/json" },
-      signal: ctx.signal
-    }) : probeJson({
-      ...probeOpts,
-      method: "POST",
-      body: JSON.stringify({ content: inputs }),
-      headers: { "content-type": "application/json" }
-    }));
-    if (!native.ok || !native.data) {
-      throw new Error(`llama.cpp embedding failed: ${native.error ?? "unknown"}`);
-    }
-    const items = native.data;
-    items.sort((a, b) => (a.index ?? 0) - (b.index ?? 0));
-    const vectors = items.map(flattenNativeEmbedding);
-    return {
-      vectors,
-      model: modelId,
-      dimensions: vectors[0]?.length ?? 0
-    };
-  }
-};
-var llamacpp_embed_default = llamacppEmbedRuntime;
-
-// src/domains/providers/runtimes/local-native/llamacpp-rerank.ts
-init_esm_shims();
-var defaultCapabilities20 = {
-  chat: false,
-  tools: false,
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: true,
-  fim: false,
-  contextWindow: 8192,
-  maxTokens: 0
-};
-function targetUrl4(target) {
-  return targetRootUrl(target);
-}
-var llamacppRerankRuntime = {
-  id: "llamacpp-rerank",
-  displayName: "llama.cpp (rerank)",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities20,
-  hidden: true,
-  async probe(target, ctx) {
-    const base = targetUrl4(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
-    const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
-    if (!health.ok) return health;
-    const modelId = target.defaultModel ?? "default";
-    const probeResponse = await fetch(`${base}/reranking`, {
-      method: "POST",
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({ query: "probe", documents: ["a"], model: modelId }),
-      ...ctx.signal ? { signal: ctx.signal } : {}
-    }).catch((err) => new Response(null, { status: 599, statusText: String(err) }));
-    if (!(probeResponse.status === 200 || probeResponse.status === 202)) {
-      return { ok: false, error: `/reranking not available: HTTP ${probeResponse.status}` };
-    }
-    const props = await probeLlamaCppProps(base, ctx);
-    const result = { ok: true };
-    if (health.latencyMs !== void 0) result.latencyMs = health.latencyMs;
-    if (props.discoveredCapabilities) result.discoveredCapabilities = props.discoveredCapabilities;
-    if (props.serverVersion) result.serverVersion = props.serverVersion;
-    return result;
-  },
-  async probeModels(target, ctx) {
-    const base = targetUrl4(target);
-    if (!base) return [];
-    return probeOpenAIModels(base, ctx);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities20,
-      apiFamily: "openai-completions",
-      provider: "llamacpp",
-      baseUrlForTarget: withV1
-    });
-  },
-  async rerank(target, query, documents, ctx) {
-    const base = targetUrl4(target);
-    if (!base) throw new Error("target has no url");
-    const modelId = target.defaultModel ?? "default";
-    const req = {
-      url: `${base}/reranking`,
-      method: "POST",
-      timeoutMs: ctx.httpTimeoutMs,
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({
-        query,
-        documents,
-        top_n: documents.length,
-        model: modelId
-      })
-    };
-    const result = await (ctx.signal ? probeJson({ ...req, signal: ctx.signal }) : probeJson(req));
-    if (!result.ok || !result.data) {
-      throw new Error(`llama.cpp rerank failed: ${result.error ?? "unknown"}`);
-    }
-    const rows = result.data.results ?? [];
-    const items = rows.map((row) => {
-      const idx = typeof row.index === "number" ? row.index : 0;
-      const score = typeof row.relevance_score === "number" ? row.relevance_score : 0;
-      const doc = typeof row.document === "string" ? row.document : typeof row.document?.text === "string" ? row.document.text : void 0;
-      const item = { index: idx, score };
-      if (doc !== void 0) item.document = doc;
-      return item;
-    });
-    return { items, model: result.data.model ?? modelId };
-  }
-};
-var llamacpp_rerank_default = llamacppRerankRuntime;
-
-// src/domains/providers/runtimes/local-native/lmstudio.ts
-init_esm_shims();
-var defaultCapabilities21 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  structuredOutputs: "json-schema",
-  reasoning: true,
-  vision: true,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var reasoningOptionsByTargetModel = /* @__PURE__ */ new Map();
-function reasoningOptionsKey(target, modelId) {
-  const url2 = (target.url ?? "").replace(/^ws:/u, "http:").replace(/^wss:/u, "https:").replace(/\/$/u, "");
-  return `${target.id}|${url2}|${modelId}`;
-}
-function rememberReasoningOptions(target, models) {
-  const prefix = reasoningOptionsKey(target, "");
-  for (const key of reasoningOptionsByTargetModel.keys()) {
-    if (key.startsWith(prefix)) reasoningOptionsByTargetModel.delete(key);
-  }
-  for (const model of models) {
-    if (!model.reasoningOptions) continue;
-    reasoningOptionsByTargetModel.set(reasoningOptionsKey(target, model.key), model.reasoningOptions);
-    for (const instance of model.loadedInstances) {
-      reasoningOptionsByTargetModel.set(reasoningOptionsKey(target, instance.id), model.reasoningOptions);
+    runtimeId: input.runtimeId,
+    apiFamily: input.apiFamily ?? null,
+    modelId: input.modelId,
+    family,
+    capabilities: input.capabilities,
+    thinking,
+    request: resolveRequestCapability(thinking, parser, input.runtimeId),
+    response: {
+      parser,
+      stripTokenizerSentinels: true
     }
-  }
+  };
+  if (quirks) result.quirks = quirks;
+  return result;
 }
-function configuredModel(target) {
-  const id = target.defaultModel?.trim();
-  return id && id.length > 0 ? id : void 0;
+function resolveModelRuntimeCapabilitiesForStatus(status, wireModelId, knowledgeBase, options) {
+  const modelId = wireModelId?.trim() || status.target.defaultModel?.trim() || "";
+  const kbHit = modelId ? knowledgeBase?.lookup(modelId) ?? null : null;
+  const capabilities = resolveModelCapabilities(status, modelId, knowledgeBase, {
+    detectedReasoning: options?.detectedReasoning ?? null
+  });
+  const runtimeId = status.runtime?.id ?? status.target.runtime;
+  return resolveModelRuntimeCapabilities({
+    targetId: status.target.id,
+    runtimeId,
+    apiFamily: status.runtime?.apiFamily ?? null,
+    modelId,
+    capabilities,
+    kbHit,
+    ...thinkingHintsForCatalogModel(runtimeId, modelId),
+    ...options?.configuredThinkingLevel ? { configuredThinkingLevel: options.configuredThinkingLevel } : {}
+  });
 }
-function resolveModel(models, id) {
-  if (id) {
-    for (const model2 of models) {
-      if (model2.key === id) return { model: model2, ...model2.loadedInstances[0] ? { instance: model2.loadedInstances[0] } : {} };
-      const instance = model2.loadedInstances.find((entry) => entry.id === id);
-      if (instance) return { model: model2, instance };
-    }
-    return null;
-  }
-  const model = models.find((entry) => entry.loadedInstances.length > 0) ?? models[0];
-  return model ? { model, ...model.loadedInstances[0] ? { instance: model.loadedInstances[0] } : {} } : null;
+function resolveModelRuntimeCapabilitiesForProviders(providers, targetId, wireModelId, configuredThinkingLevel) {
+  const id = targetId?.trim();
+  if (!id) return null;
+  const status = providers.list().find((entry) => entry.target.id === id);
+  if (!status) return null;
+  const modelId = wireModelId?.trim() || status.target.defaultModel?.trim() || "";
+  const detectedReasoning = modelId && typeof providers.getDetectedReasoning === "function" ? providers.getDetectedReasoning(id, modelId) : null;
+  return resolveModelRuntimeCapabilitiesForStatus(status, modelId, providers.knowledgeBase, {
+    detectedReasoning,
+    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
+  });
 }
-function capabilities(model, instance) {
+function thinkingHintsForModel(model) {
   const out = {};
-  if (model.vision !== void 0) out.vision = model.vision;
-  if (model.tools !== void 0) out.tools = model.tools;
-  if (model.reasoning !== void 0) out.reasoning = model.reasoning;
-  else if (model.reasoningOptions !== void 0)
-    out.reasoning = model.reasoningOptions.some((option) => option !== "off");
-  const contextWindow = loadedContextLength(instance) ?? model.maxContextLength;
-  if (contextWindow !== void 0) out.contextWindow = contextWindow;
+  if (!model) return out;
+  const compat = model.compat;
+  if (compat?.forceAdaptiveThinking !== void 0) out.adaptiveThinking = compat.forceAdaptiveThinking;
+  if (model.thinkingLevelMap) out.thinkingLevelMap = model.thinkingLevelMap;
   return out;
 }
-function statusFor(model, instance, reasoningLevels) {
-  const status = {
-    state: instance ? "loaded" : "unloaded",
-    key: model.key,
-    reasoningLevels
-  };
-  if (instance) {
-    status.instanceId = instance.id;
-    status.loadConfig = instance.config;
-    const contextLength = loadedContextLength(instance);
-    if (contextLength !== void 0) status.contextLength = contextLength;
-  }
-  return status;
+function thinkingHintsForCatalogModel(runtimeId, modelId) {
+  if (!runtimeId || modelId.length === 0) return {};
+  return thinkingHintsForModel(getCatalogModelForRuntime(runtimeId, modelId));
 }
-function probeFromCatalog(catalog, target) {
-  if (!catalog.ok) {
-    return {
-      ok: false,
-      ...catalog.latencyMs !== void 0 ? { latencyMs: catalog.latencyMs } : {},
-      error: catalog.error ?? "LM Studio model listing failed"
-    };
-  }
-  const models = foldLmStudioModels(catalog.models);
-  rememberReasoningOptions(target, models);
-  const ids = [];
-  const modelCapabilities = {};
-  const modelStates = {};
-  for (const model of models) {
-    const levels = lmStudioReasoningLevels(model.reasoningOptions);
-    const keyInstance = model.loadedInstances[0];
-    modelCapabilities[model.key] = capabilities(model, keyInstance);
-    modelStates[model.key] = statusFor(model, keyInstance, levels);
-    for (const instance of model.loadedInstances) {
-      if (!ids.includes(instance.id)) ids.push(instance.id);
-      modelCapabilities[instance.id] = capabilities(model, instance);
-      const instanceStatus = statusFor(model, instance, levels);
-      modelStates[instance.id] = instanceStatus;
-      const resolution = resolveLmStudioInstance(target, models, instance.id, configuredModel(target));
-      instanceStatus.detail = resolution.peerTargets.length > 0 ? `loaded on ${target.id}; also loaded on ${resolution.peerTargets.join(", ")}` : `loaded on ${target.id}`;
-    }
-    if (model.loadedInstances.length === 0) {
-      ids.push(model.key);
-      const keyStatus = modelStates[model.key];
-      if (keyStatus) keyStatus.detail = "not loaded (LM Studio will load it on first use)";
-    }
+function thinkingFormatFromModelApi(api) {
+  switch (api) {
+    case "anthropic-messages":
+    case "bedrock-converse-stream":
+    case "claude-agent-sdk":
+    case "claude-code-subprocess":
+      return "anthropic-extended";
+    case "openai-codex-responses":
+      return "openai-codex";
+    default:
+      return void 0;
   }
-  const selected = resolveModel(models, configuredModel(target));
-  const result = {
-    ok: true,
-    models: ids,
-    modelCapabilities,
-    modelStates,
-    serverVersion: catalog.tier === "0.4+" ? "LM Studio API 0.4+" : catalog.tier === "0.3.x" ? "LM Studio API 0.3.x" : "LM Studio OpenAI-compatible API",
-    surfaces: {
-      openaiChat: "/v1/chat/completions",
-      ...catalog.tier === "0.4+" ? { nativeV1: "/api/v1/models" } : {},
-      ...catalog.tier === "0.3.x" ? { nativeV0: "/api/v0/models" } : {}
-    }
-  };
-  if (catalog.latencyMs !== void 0) result.latencyMs = catalog.latencyMs;
-  if (selected) {
-    result.discoveredCapabilities = capabilities(selected.model, selected.instance);
-    result.capabilityModelId = configuredModel(target) ?? selected.model.key;
-  }
-  const loaded = models.flatMap(
-    (model) => model.loadedInstances.map((instance) => {
-      const context = loadedContextLength(instance);
-      return `${model.key} as ${instance.id}${context ? ` at ${context} tokens` : ""}`;
-    })
-  );
-  result.notes = [
-    `LM Studio surface tier ${catalog.tier ?? "unknown"}`,
-    `Surfaces: ${Object.values(result.surfaces ?? {}).join(", ")}`,
-    ...loaded.length > 0 ? [`Loaded instances: ${loaded.join(", ")}`] : []
-  ];
-  return result;
 }
-var lmstudioRuntime = {
-  id: "lmstudio",
-  aliases: ["lmstudio-native"],
-  displayName: "LM Studio",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "openai-completions",
-  auth: "api-key",
-  defaultCapabilities: defaultCapabilities21,
-  async probe(target, ctx) {
-    const [greeting, catalog] = await Promise.all([greetLmStudio(target, ctx), listLmStudioModels(target, ctx)]);
-    if (!greeting.ok && catalog.tier !== "0.3.x") {
-      return {
-        ok: false,
-        ...greeting.latencyMs !== void 0 ? { latencyMs: greeting.latencyMs } : {},
-        error: `URL did not identify itself as LM Studio: ${greeting.error ?? "greeting missing"}`
-      };
-    }
-    return probeFromCatalog(catalog, target);
-  },
-  async probeModels(target, ctx) {
-    const catalog = await listLmStudioModels(target, ctx);
-    return probeFromCatalog(catalog, target).models ?? [];
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    const canonicalTarget = target.runtime === "lmstudio" ? target : { ...target, runtime: "lmstudio" };
-    const model = synthLocalModel({
-      target: canonicalTarget,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities21,
-      apiFamily: "openai-completions",
-      provider: "lmstudio",
-      baseUrlForTarget: (url2) => withV1(url2.replace(/^ws:/u, "http:").replace(/^wss:/u, "https:"))
-    });
-    const metadata2 = model.clio;
-    if (metadata2) {
-      metadata2.chatTemplateKwargsUnsupported = true;
-      if (target.defaultModel) metadata2.lmstudioDefaultModel = target.defaultModel;
-      const options = reasoningOptionsByTargetModel.get(reasoningOptionsKey(canonicalTarget, wireModelId));
-      if (options) metadata2.lmstudioReasoningOptions = options;
-    }
-    return model;
+function capabilitiesFromModel(model) {
+  const format = model.compat?.thinkingFormat ?? thinkingFormatFromModelApi(model.api);
+  const caps = {
+    chat: true,
+    tools: true,
+    reasoning: model.reasoning === true,
+    vision: Array.isArray(model.input) && model.input.includes("image"),
+    audio: false,
+    embeddings: false,
+    rerank: false,
+    fim: false,
+    contextWindow: model.contextWindow,
+    maxTokens: model.maxTokens
+  };
+  if (format === "qwen-chat-template" || format === "openrouter" || format === "zai" || format === "anthropic-extended" || format === "deepseek-r1" || format === "openai-codex" || format === "harmony") {
+    caps.thinkingFormat = format;
   }
-};
-var lmstudio_default = lmstudioRuntime;
-
-// src/domains/providers/runtimes/local-native/ollama-native.ts
-init_esm_shims();
-var defaultCapabilities22 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-function positiveNumber3(value) {
-  return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : void 0;
+  return caps;
 }
-async function probeResidentModelStates(base, ctx) {
-  const opts = { url: `${base}/api/ps`, timeoutMs: ctx.httpTimeoutMs };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok || !result.data?.models) return void 0;
-  const states = {};
-  for (const row of result.data.models) {
-    const id = typeof row?.model === "string" ? row.model : typeof row?.name === "string" ? row.name : null;
-    if (!id) continue;
-    const status = { state: "loaded" };
-    const sizeVram = positiveNumber3(row?.size_vram);
-    if (sizeVram !== void 0) status.sizeVramBytes = sizeVram;
-    const size = positiveNumber3(row?.size);
-    if (size !== void 0) status.sizeBytes = size;
-    states[id] = status;
-  }
-  return Object.keys(states).length > 0 ? states : void 0;
-}
-var ollamaNativeRuntime = {
-  id: "ollama-native",
-  displayName: "Ollama (native)",
-  kind: "http",
-  tier: "local-native",
-  apiFamily: "ollama-native",
-  auth: "none",
-  defaultCapabilities: defaultCapabilities22,
-  async probe(target, ctx) {
-    const base = targetBaseUrl(target);
-    if (!base) return { ok: false, error: "target has no url" };
-    const opts = { url: `${base}/api/tags`, timeoutMs: ctx.httpTimeoutMs };
-    const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-    if (!result.ok) {
-      const failed = { ok: false };
-      if (result.error) failed.error = result.error;
-      if (result.latencyMs !== void 0) failed.latencyMs = result.latencyMs;
-      return failed;
-    }
-    const out = { ok: true };
-    if (result.latencyMs !== void 0) out.latencyMs = result.latencyMs;
-    const modelStates = await probeResidentModelStates(base, ctx);
-    if (modelStates) out.modelStates = modelStates;
-    return out;
-  },
-  async probeModels(target, ctx) {
-    const base = targetBaseUrl(target);
-    if (!base) return [];
-    const opts = { url: `${base}/api/tags`, timeoutMs: ctx.httpTimeoutMs };
-    const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-    if (!result.ok || !result.data?.models) return [];
-    return result.data.models.map((row) => typeof row?.name === "string" ? row.name : null).filter((name) => name !== null);
-  },
-  synthesizeModel(target, wireModelId, kb) {
-    return synthLocalModel({
-      target,
-      wireModelId,
-      kb,
-      defaultCapabilities: defaultCapabilities22,
-      apiFamily: "ollama-native",
-      provider: "ollama",
-      baseUrlForTarget: withAsIs
-    });
-  }
-};
-var ollama_native_default = ollamaNativeRuntime;
-
-// src/domains/providers/runtimes/local-native/sglang.ts
-init_esm_shims();
-var defaultCapabilities23 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var sglang_default = makeOpenAICompatRuntime({
-  id: "sglang",
-  displayName: "SGLang",
-  provider: "sglang",
-  auth: "api-key",
-  tier: "local-native",
-  defaultCapabilities: defaultCapabilities23
-});
-
-// src/domains/providers/runtimes/local-native/vllm.ts
-init_esm_shims();
-var defaultCapabilities24 = {
-  chat: true,
-  tools: true,
-  toolCallFormat: "openai",
-  reasoning: false,
-  vision: false,
-  audio: false,
-  embeddings: false,
-  rerank: false,
-  fim: false,
-  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
-  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
-};
-var vllm_default = makeOpenAICompatRuntime({
-  id: "vllm",
-  displayName: "vLLM",
-  provider: "vllm",
-  auth: "api-key",
-  tier: "local-native",
-  defaultCapabilities: defaultCapabilities24,
-  healthPath: "/health"
-});
-
-// src/domains/providers/runtimes/builtins.ts
-var BUILTIN_RUNTIMES = [
-  alcf_default,
-  anthropic_default,
-  anthropic_max_default,
-  bedrock_default,
-  deepseek_default,
-  google_default,
-  groq_default,
-  mistral_default,
-  openai_default,
-  openai_codex_default,
-  openrouter_default,
-  lemonade_anthropic_default,
-  lemonade_openai_default,
-  llamacpp_default,
-  llamacpp_anthropic_default,
-  llamacpp_completion_default,
-  llamacpp_embed_default,
-  llamacpp_rerank_default,
-  lmstudio_default,
-  ollama_native_default,
-  anthropic_compat_default,
-  openai_compat_default,
-  sglang_default,
-  vllm_default,
-  claude_code_default,
-  claude_sdk_default,
-  antigravity_code_default
-];
-function registerBuiltinRuntimes(registry) {
-  for (const desc of BUILTIN_RUNTIMES) {
-    if (registry.get(desc.id) !== null) continue;
-    registry.register(desc);
-  }
+function resolveModelRuntimeCapabilitiesForModel(model, configuredThinkingLevel) {
+  const metadata2 = model.clio;
+  const caps = capabilitiesFromModel(model);
+  return resolveModelRuntimeCapabilities({
+    targetId: metadata2?.targetId ?? null,
+    runtimeId: metadata2?.runtimeId ?? model.provider,
+    apiFamily: model.api,
+    modelId: model.id,
+    capabilities: caps,
+    ...thinkingHintsForModel(model),
+    ...metadata2?.quirks ? { quirks: metadata2.quirks } : {},
+    kbHit: metadata2?.family ? {
+      matchKind: "family",
+      entry: {
+        family: metadata2.family,
+        matchPatterns: [metadata2.family],
+        capabilities: {}
+      }
+    } : null,
+    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
+  });
+}
+function resolveTargetRuntimeCapabilities(target, runtime, wireModelId, capabilities, knowledgeBase, configuredThinkingLevel) {
+  const kbHit = knowledgeBase?.lookup(wireModelId) ?? null;
+  return resolveModelRuntimeCapabilities({
+    targetId: target.id,
+    runtimeId: runtime.id,
+    apiFamily: runtime.apiFamily,
+    modelId: wireModelId,
+    capabilities,
+    kbHit,
+    ...thinkingHintsForCatalogModel(runtime.id, wireModelId),
+    ...configuredThinkingLevel ? { configuredThinkingLevel } : {}
+  });
 }
 
 // src/domains/providers/eligibility.ts
@@ -4468,140 +1307,6 @@ function isDispatchEligibleRuntime(runtime) {
   return isTargetEligibleRuntime(runtime);
 }
 
-// src/domains/providers/support.ts
-init_esm_shims();
-var SUMMARY_BY_RUNTIME_ID = {
-  alcf: "ALCF inference gateway (Sophia/Metis) via Globus",
-  anthropic: "Anthropic API",
-  "anthropic-max": "Claude Pro/Max subscription via Anthropic OAuth",
-  bedrock: "Amazon Bedrock",
-  "claude-code": "Claude Code subscription via installed claude CLI",
-  "claude-sdk": "Claude Code subscription via Claude Agent SDK",
-  deepseek: "DeepSeek API",
-  google: "Google Gemini API",
-  groq: "Groq API",
-  mistral: "Mistral API",
-  openai: "OpenAI Platform API",
-  "openai-codex": "ChatGPT Plus/Pro via Codex OAuth",
-  openrouter: "OpenRouter API",
-  "ollama-native": "Ollama native API",
-  lmstudio: "LM Studio chat over OpenAI-compatible REST with native REST model management",
-  llamacpp: "llama.cpp server (auto-detect surface)",
-  "anthropic-compat": "Generic Anthropic-compatible REST",
-  "openai-compat": "Generic OpenAI-compatible REST"
-};
-function groupPriority(group) {
-  switch (group) {
-    case "featured":
-      return 0;
-    case "subscription":
-      return 1;
-    case "cloud-api":
-      return 2;
-    case "local-http":
-      return 3;
-  }
-}
-function supportGroupLabel(group) {
-  switch (group) {
-    case "featured":
-      return "Featured";
-    case "subscription":
-      return "Subscriptions";
-    case "cloud-api":
-      return "Cloud APIs";
-    case "local-http":
-      return "Local HTTP";
-  }
-}
-function classifyGroup(runtime) {
-  if (runtime.id === "openai-codex") return "featured";
-  if (runtime.id === "alcf") return "cloud-api";
-  if (runtime.auth === "oauth" || runtime.auth === "claude-cli") return "subscription";
-  if (catalogProviderForRuntime(runtime.id) || runtime.auth === "api-key" && !runtime.probe) {
-    return "cloud-api";
-  }
-  return "local-http";
-}
-function knownModelsFor(runtimeId, runtime) {
-  const catalogModels2 = listCatalogModelsForRuntime(runtimeId);
-  if (catalogModels2.length === 0) return runtime?.knownModels ? [...runtime.knownModels] : [];
-  return catalogModels2.map((model) => model.id);
-}
-function listKnownModelsForRuntime(runtimeId) {
-  return knownModelsFor(runtimeId, getRuntimeIfRegistered(runtimeId));
-}
-function getRuntimeIfRegistered(runtimeId) {
-  try {
-    return getRuntimeRegistry().get(runtimeId);
-  } catch {
-    return null;
-  }
-}
-function runtimeModelListSource(runtime) {
-  if (runtime.knownModels && runtime.knownModels.length > 0) return "runtime";
-  if (listCatalogModelsForRuntime(runtime.id).length > 0) return "catalog";
-  return "none";
-}
-function describeRuntimeModels(entry, sample) {
-  if (entry.modelSource === "catalog") return `${entry.modelHints.length} in catalog`;
-  if (entry.modelHints.length === 0) return "-";
-  return entry.modelHints.slice(0, sample).join(", ");
-}
-function buildProviderSupportEntry(runtime) {
-  const modelHints = knownModelsFor(runtime.id, runtime);
-  const modelSource = runtimeModelListSource(runtime);
-  const defaultModel = modelSource === "catalog" ? void 0 : modelHints[0];
-  return {
-    runtimeId: runtime.id,
-    label: runtime.displayName,
-    group: classifyGroup(runtime),
-    summary: SUMMARY_BY_RUNTIME_ID[runtime.id] ?? runtime.displayName,
-    ...defaultModel ? { defaultModel } : {},
-    modelHints,
-    modelSource,
-    featured: runtime.id === "openai-codex",
-    connectable: runtime.auth === "oauth" || runtime.auth === "api-key",
-    supportsCustomUrl: runtime.kind === "http" && (classifyGroup(runtime) === "local-http" || runtime.id === "openai-compat" || runtime.id === "anthropic-compat" || runtime.id === "alcf")
-  };
-}
-function compareProviderSupportEntries(a, b) {
-  return groupPriority(a.group) - groupPriority(b.group) || (a.featured === b.featured ? 0 : a.featured ? -1 : 1) || a.label.localeCompare(b.label) || a.runtimeId.localeCompare(b.runtimeId);
-}
-function listProviderSupportEntries(runtimes, options = {}) {
-  const filtered = options.includeHidden ? runtimes : runtimes.filter((runtime) => runtime.hidden !== true);
-  return filtered.map((runtime) => buildProviderSupportEntry(runtime)).sort(compareProviderSupportEntries);
-}
-function configuredTargetsForRuntime(settings, runtimeId) {
-  const canonical = getRuntimeIfRegistered(runtimeId)?.id ?? runtimeId;
-  return settings.targets.filter(
-    (target) => (getRuntimeIfRegistered(target.runtime)?.id ?? target.runtime) === canonical
-  );
-}
-function resolveProviderReference(input, settings, getRuntime) {
-  const trimmed = input.trim();
-  if (trimmed.length === 0) return null;
-  const target = settings.targets.find((entry) => entry.id === trimmed) ?? null;
-  if (target) {
-    const runtime2 = getRuntime(target.runtime);
-    if (!runtime2) return null;
-    return {
-      input: trimmed,
-      target,
-      runtime: runtime2,
-      authTarget: resolveAuthTarget(target, runtime2)
-    };
-  }
-  const runtime = getRuntime(trimmed);
-  if (!runtime) return null;
-  return {
-    input: trimmed,
-    target: null,
-    runtime,
-    authTarget: resolveRuntimeAuthTarget(runtime)
-  };
-}
-
 // src/domains/providers/model-discovery.ts
 init_esm_shims();
 function uniqueModels(ids) {
@@ -4703,24 +1408,24 @@ function diagnostic(severity, code, message) {
 function hasError(diagnostics) {
   return diagnostics.some((entry) => entry.severity === "error");
 }
-function statusFor2(providers, target, runtime, _wireModelId) {
+function statusFor(providers, target, runtime, _wireModelId) {
   const existing = providers.list().find((entry) => entry.target.id === target.id);
   if (existing) return existing;
-  const capabilities2 = { ...runtime.defaultCapabilities, ...target.capabilities ?? {} };
+  const capabilities = { ...runtime.defaultCapabilities, ...target.capabilities ?? {} };
   return {
     target,
     runtime,
     available: true,
     reason: "synthetic-status",
     health: { status: "unknown", lastCheckAt: null, lastError: null, latencyMs: null },
-    capabilities: capabilities2,
+    capabilities,
     probeCapabilities: null,
     probeModelId: null,
     discoveredModels: runtime.knownModels ?? []
   };
 }
-function requiredCapabilitySupported(capabilities2, name) {
-  const value = capabilities2[name];
+function requiredCapabilitySupported(capabilities, name) {
+  const value = capabilities[name];
   return value !== void 0 && value !== false && value !== 0 && value !== "";
 }
 function streamingDecision(runtime) {
@@ -4730,18 +1435,18 @@ function runtimeSupportsUse(runtime, use) {
   if (use === "dispatch") return isDispatchEligibleRuntime(runtime);
   return isOrchestratorEligibleRuntime(runtime);
 }
-function capabilityDecisions(runtime, capabilities2) {
+function capabilityDecisions(runtime, capabilities) {
   return {
-    chat: capabilities2.chat,
-    tools: capabilities2.tools,
-    reasoning: capabilities2.reasoning,
-    vision: capabilities2.vision,
+    chat: capabilities.chat,
+    tools: capabilities.tools,
+    reasoning: capabilities.reasoning,
+    vision: capabilities.vision,
     streaming: streamingDecision(runtime),
-    contextWindow: capabilities2.contextWindow,
-    maxTokens: capabilities2.maxTokens
+    contextWindow: capabilities.contextWindow,
+    maxTokens: capabilities.maxTokens
   };
 }
-function appendCapabilityDiagnostics(diagnostics, input, capabilities2, decisions, targetId) {
+function appendCapabilityDiagnostics(diagnostics, input, capabilities, decisions, targetId) {
   if (!decisions.chat) {
     diagnostics.push(diagnostic("error", "chat-unsupported", `target '${targetId}' does not advertise chat support`));
   }
@@ -4757,7 +1462,7 @@ function appendCapabilityDiagnostics(diagnostics, input, capabilities2, decision
     );
   }
   for (const capability of input.requiredCapabilities ?? []) {
-    if (!requiredCapabilitySupported(capabilities2, capability)) {
+    if (!requiredCapabilitySupported(capabilities, capability)) {
       diagnostics.push(
         diagnostic(
           "error",
@@ -4874,14 +1579,14 @@ function resolveRuntimeTarget(providers, input) {
       diagnostics: [diagnostic("error", "model-not-configured", `target '${targetId}' has no model configured`)]
     };
   }
-  const status = statusFor2(providers, target, runtime, wireModelId);
+  const status = statusFor(providers, target, runtime, wireModelId);
   const unknownModel = unknownModelDiagnostic(target, status, wireModelId);
   if (unknownModel) diagnostics.push(unknownModel);
   const requestedThinkingLevel = input.requestedThinkingLevel ?? "off";
   const capabilityResolution = modelCapabilitiesFor(providers, status, wireModelId);
-  const capabilities2 = { ...capabilityResolution.capabilities };
+  const capabilities = { ...capabilityResolution.capabilities };
   const probedContextWindow = probeCapabilitiesForModel(status, wireModelId)?.contextWindow ?? null;
-  const loadedContextWindow = loadedContextWindowForModel(status, wireModelId);
+  const loadedContextWindow = loadedContextWindowForModel(status, wireModelId) ?? input.knownLoadedContextWindow ?? null;
   const contextWindowDetails = resolveContextWindowDetails(
     target,
     runtime,
@@ -4892,7 +1597,7 @@ function resolveRuntimeTarget(providers, input) {
     void 0,
     contextSlotsForModel(status, wireModelId)
   );
-  capabilities2.contextWindow = contextWindowDetails.effectiveContextWindow;
+  capabilities.contextWindow = contextWindowDetails.effectiveContextWindow;
   if (contextWindowDetails.warning) {
     diagnostics.push(diagnostic("warning", "context-window-low", contextWindowDetails.warning));
   }
@@ -4903,12 +1608,12 @@ function resolveRuntimeTarget(providers, input) {
     target,
     runtime,
     wireModelId,
-    capabilities2,
+    capabilities,
     providers.knowledgeBase,
     requestedThinkingLevel
   );
-  const decisions = capabilityDecisions(runtime, capabilities2);
-  appendCapabilityDiagnostics(diagnostics, input, capabilities2, decisions, targetId);
+  const decisions = capabilityDecisions(runtime, capabilities);
+  appendCapabilityDiagnostics(diagnostics, input, capabilities, decisions, targetId);
   appendThinkingDiagnostics(diagnostics, modelRuntime, requestedThinkingLevel);
   if (hasError(diagnostics)) return { ok: false, diagnostics };
   const resolved = {
@@ -4924,7 +1629,7 @@ function resolveRuntimeTarget(providers, input) {
     costProvenance: resolveCostProvenance(target, runtime.id, wireModelId),
     requestedThinkingLevel,
     effectiveThinkingLevel: modelRuntime.thinking.effectiveLevel,
-    capabilities: capabilities2,
+    capabilities,
     capabilityDecisions: decisions,
     modelRuntime,
     modelReasoningAuthoritative: capabilityResolution.reasoningAuthoritative,
@@ -4960,7 +1665,7 @@ function refineRuntimeTargetWithModelHints(target, model, knowledgeBase) {
   const modelHintContextWindow = nonNegativeFiniteNumber(hintRecord?.contextWindow);
   const windowHintDiffers = modelHintContextWindow !== void 0 && modelHintContextWindow > 0 && modelHintContextWindow !== target.contextWindowDetails.effectiveContextWindow;
   if (Object.keys(patch).length === 0 && !windowHintDiffers) return target;
-  const capabilities2 = { ...target.capabilities, ...patch };
+  const capabilities = { ...target.capabilities, ...patch };
   const contextWindowDetails = resolveContextWindowDetails(
     target.target,
     target.runtime,
@@ -4974,22 +1679,22 @@ function refineRuntimeTargetWithModelHints(target, model, knowledgeBase) {
     modelHintContextWindow,
     target.contextWindowDetails.contextWindowSlots
   );
-  capabilities2.contextWindow = contextWindowDetails.effectiveContextWindow;
+  capabilities.contextWindow = contextWindowDetails.effectiveContextWindow;
   const modelRuntime = resolveModelRuntimeCapabilities({
     targetId: target.targetId,
     runtimeId: target.runtimeId,
     apiFamily: target.apiFamily,
     modelId: target.wireModelId,
-    capabilities: capabilities2,
+    capabilities,
     ...target.modelRuntime.quirks ? { quirks: target.modelRuntime.quirks } : {},
     configuredThinkingLevel: target.requestedThinkingLevel
   });
-  const decisions = capabilityDecisions(target.runtime, capabilities2);
+  const decisions = capabilityDecisions(target.runtime, capabilities);
   const diagnostics = withoutStaleRuntimeDiagnostics(target.diagnostics, decisions);
   appendThinkingDiagnostics(diagnostics, modelRuntime, target.requestedThinkingLevel);
   return {
     ...target,
-    capabilities: capabilities2,
+    capabilities,
     capabilityDecisions: decisions,
     modelRuntime,
     effectiveThinkingLevel: modelRuntime.thinking.effectiveLevel,
@@ -5109,6 +1814,12 @@ function resolveContextWindowDetails(target, runtime, wireModelId, knowledgeBase
   };
 }
 
+// src/domains/providers/types/cost-provenance.ts
+init_esm_shims();
+function normalizeCostProvenance(value) {
+  return value ?? "unknown";
+}
+
 // src/domains/providers/index.ts
 init_esm_shims();
 
@@ -5126,6 +1837,18 @@ import {
   Ollama
 } from "ollama";
 
+// src/core/residency-target-key.ts
+init_esm_shims();
+var RESIDENCY_RUNTIME_IDS = /* @__PURE__ */ new Set([
+  "llamacpp",
+  "lmstudio",
+  "ollama-native"
+]);
+function residencyTargetKey(runtimeId, baseUrl) {
+  if (!RESIDENCY_RUNTIME_IDS.has(runtimeId) || typeof baseUrl !== "string") return null;
+  return canonicalEndpointUrl(baseUrl) ?? baseUrl;
+}
+
 // src/engine/gemma-channel-filter.ts
 init_esm_shims();
 import { createAssistantMessageEventStream } from "@earendil-works/pi-ai";
@@ -5395,7 +2118,7 @@ async function reconcileOllamaResidency(model, headers) {
   if (!baseUrl) return;
   const metadata2 = model.clio;
   const adapter = {
-    targetKey: `ollama-native|${baseUrl}`,
+    targetKey: residencyTargetKey("ollama-native", baseUrl),
     targetId: ollamaTargetId(model),
     runtimeId: "ollama-native",
     keepModelId: model.id,
@@ -5766,7 +2489,7 @@ var ollamaNativeApiProvider = {
 
 // src/engine/apis/openai-completions.ts
 init_esm_shims();
-import { TextDecoder as TextDecoder2 } from "node:util";
+import { TextDecoder } from "node:util";
 import {
   createAssistantMessageEventStream as createAssistantMessageEventStream3
 } from "@earendil-works/pi-ai";
@@ -5893,30 +2616,30 @@ function harmonyPrefixTailLength(value) {
 
 // src/engine/apis/llamacpp-residency.ts
 init_esm_shims();
-import { performance as performance4 } from "node:perf_hooks";
+import { performance } from "node:perf_hooks";
 var LOAD_TIMEOUT_MS = 12e4;
 var POLL_INTERVAL_MS = 500;
 function isResidentState(state) {
   return state === "loaded" || state === "loading" || state === "sleeping";
 }
-function isRecord4(value) {
+function isRecord(value) {
   return value !== null && typeof value === "object" && !Array.isArray(value);
 }
 function routerModelState(entry) {
   const status = entry.status;
-  const state = isRecord4(status) ? status.value : status;
+  const state = isRecord(status) ? status.value : status;
   if (state === "loaded" || state === "loading" || state === "sleeping" || state === "unloaded" || state === "failed") {
     return state;
   }
   return "unknown";
 }
 function parseRouterModels(payload) {
-  if (!isRecord4(payload)) return [];
+  if (!isRecord(payload)) return [];
   const data = payload.data;
   if (!Array.isArray(data)) return [];
   const models = [];
   for (const entry of data) {
-    if (!isRecord4(entry) || typeof entry.id !== "string") continue;
+    if (!isRecord(entry) || typeof entry.id !== "string") continue;
     const tags = Array.isArray(entry.tags) ? entry.tags.filter((tag) => typeof tag === "string") : [];
     models.push({ id: entry.id, state: routerModelState(entry), tags });
   }
@@ -5926,14 +2649,14 @@ function residentModel(model) {
   return isResidentState(model.state);
 }
 function parseLlamaCppRouterProps(payload) {
-  if (!isRecord4(payload)) return {};
+  if (!isRecord(payload)) return {};
   const maxInstances = payload.max_instances;
   return typeof maxInstances === "number" && Number.isFinite(maxInstances) && maxInstances > 0 ? { maxInstances: Math.floor(maxInstances) } : {};
 }
 function rootUrl(baseUrl) {
   return baseUrl.replace(/\/+$/, "").replace(/\/v1$/, "");
 }
-function modelsUrl2(baseUrl) {
+function modelsUrl(baseUrl) {
   return `${baseUrl.replace(/\/+$/, "")}/models`;
 }
 function propsUrl(baseUrl) {
@@ -5957,25 +2680,65 @@ async function fetchRouterProps(input, fetchImpl) {
   }
 }
 async function fetchRouterModels(input, fetchImpl) {
-  const response = await fetchImpl(modelsUrl2(input.baseUrl), {
+  const response = await fetchImpl(modelsUrl(input.baseUrl), {
     signal: AbortSignal.timeout(input.timeoutMs ?? 1500)
   });
   if (!response.ok) throw new Error(`HTTP ${response.status}`);
   return parseRouterModels(await response.json());
 }
-async function postRouterModel(fetchImpl, url2, modelId) {
-  const response = await fetchImpl(url2, {
+var RouterModelPostError = class extends Error {
+  constructor(url, status, responseBody) {
+    super(`llama.cpp router rejected ${url}: HTTP ${status}; response: ${responseBody}`);
+    this.url = url;
+    this.status = status;
+    this.responseBody = responseBody;
+    this.name = "RouterModelPostError";
+  }
+  url;
+  status;
+  responseBody;
+};
+async function routerErrorBody(response) {
+  try {
+    const body = (await response.text()).trim();
+    return body.length > 0 ? body : "<empty body>";
+  } catch (error) {
+    return `<unreadable body: ${error instanceof Error ? error.message : String(error)}>`;
+  }
+}
+async function postRouterModel(fetchImpl, url, modelId) {
+  const response = await fetchImpl(url, {
     method: "POST",
     headers: { "content-type": "application/json" },
     body: JSON.stringify({ model: modelId }),
     signal: AbortSignal.timeout(LOAD_TIMEOUT_MS)
   });
   if (response.ok) return;
-  throw new Error(`llama.cpp router rejected ${url2}: HTTP ${response.status}`);
+  throw new RouterModelPostError(url, response.status, await routerErrorBody(response));
+}
+function alreadyRunningResponse(body) {
+  return body.toLowerCase().includes("already running");
+}
+async function postRouterLoad(input, fetchImpl, modelId) {
+  try {
+    await postRouterModel(fetchImpl, loadUrl(input.baseUrl), modelId);
+  } catch (error) {
+    if (!(error instanceof RouterModelPostError)) throw error;
+    let model;
+    try {
+      model = (await fetchRouterModels(input, fetchImpl)).find((entry) => entry.id === modelId);
+    } catch {
+      if (alreadyRunningResponse(error.responseBody)) return;
+      throw error;
+    }
+    if (alreadyRunningResponse(error.responseBody)) return;
+    if (model !== void 0 && model.state !== "unloaded" && model.state !== "failed") return;
+    throw error;
+  }
 }
 async function waitForLoaded(input, fetchImpl, modelId) {
-  const started = performance4.now();
-  while (performance4.now() - started < LOAD_TIMEOUT_MS) {
+  const started = performance.now();
+  while (performance.now() - started < LOAD_TIMEOUT_MS) {
     const models = await fetchRouterModels(input, fetchImpl);
     const model = models.find((entry) => entry.id === modelId);
     if (model?.state === "loaded" || model?.state === "sleeping") return;
@@ -5986,7 +2749,7 @@ async function waitForLoaded(input, fetchImpl, modelId) {
 }
 async function ensureModelLoaded(input, fetchImpl, modelId, knownState) {
   if (knownState === "loaded" || knownState === "sleeping") return;
-  if (knownState !== "loading") await postRouterModel(fetchImpl, loadUrl(input.baseUrl), modelId);
+  if (knownState !== "loading") await postRouterLoad(input, fetchImpl, modelId);
   await waitForLoaded(input, fetchImpl, modelId);
 }
 async function restoreDisplacedPinned(input, fetchImpl, before) {
@@ -6013,7 +2776,7 @@ async function ensureLlamaCppResidency(input) {
   const fetchImpl = input.fetchImpl ?? fetch;
   let snapshot = [];
   const adapter = {
-    targetKey: `llamacpp|${input.baseUrl}`,
+    targetKey: residencyTargetKey("llamacpp", input.baseUrl),
     targetId: input.targetId,
     runtimeId: input.runtimeId,
     keepModelId: input.keepModelId,
@@ -6112,7 +2875,7 @@ function targetForModel(model) {
     ...info.lmstudio ? { lmstudio: info.lmstudio } : {}
   };
 }
-function resolveModel2(catalog, id) {
+function resolveModel(catalog, id) {
   for (const model of catalog.models) {
     if (model.key === id) return { model, ...model.loadedInstances[0] ? { instance: model.loadedInstances[0] } : {} };
     const instance = model.loadedInstances.find((entry) => entry.id === id);
@@ -6207,13 +2970,13 @@ async function ensureLmStudioResidency(model, options = {}) {
   }
   if (!load || Object.keys(load).length === 0 || resolution.instance) return resolution.wireModelId;
   if (catalog.tier !== "0.4+") return resolution.wireModelId;
-  const selected = resolveModel2(catalog, model.id);
+  const selected = resolveModel(catalog, model.id);
   const modelKey = selected?.model.key ?? model.id;
   let instances = catalog.models.flatMap(
     (entry) => entry.loadedInstances.map((instance) => ({ modelKey: entry.key, identifier: instance.id, instance }))
   );
   const managed = residencyManagedFor(info.lifecycle);
-  const targetKey = `lmstudio|${lmStudioRootUrl(target.url ?? model.baseUrl)}`;
+  const targetKey = residencyTargetKey("lmstudio", target.url ?? model.baseUrl);
   const contextLength = target.lmstudio?.load?.contextLength;
   const plan = await reconcileResidency({
     targetKey,
@@ -6256,17 +3019,36 @@ async function ensureLmStudioResidency(model, options = {}) {
       });
     }
   }
+  const loadAndReport = async () => {
+    const instanceId = await loadOwnedInstance(target, targetKey, body, options.apiKey, options.signal);
+    emitResidencyMutation({
+      targetKey,
+      targetId: info.targetId,
+      runtimeId: "lmstudio",
+      model: modelKey,
+      operation: "load"
+    });
+    return instanceId ?? model.id;
+  };
   try {
-    return await loadOwnedInstance(target, targetKey, body, options.apiKey, options.signal) ?? model.id;
+    return await loadAndReport();
   } catch (error) {
     if (plan.decision === "observe" || plan.fallbackEvict.length === 0) throw error;
     return withResidencyLock(targetKey, async () => {
       for (const candidate of plan.fallbackEvict) {
         for (const entry of instances.filter((resident) => resident.modelKey === candidate.modelId)) {
-          await unloadOwnedInstance(target, targetKey, entry.identifier, options.apiKey, options.signal);
+          if (await unloadOwnedInstance(target, targetKey, entry.identifier, options.apiKey, options.signal)) {
+            emitResidencyMutation({
+              targetKey,
+              targetId: info.targetId,
+              runtimeId: "lmstudio",
+              model: candidate.modelId,
+              operation: "evict"
+            });
+          }
         }
       }
-      return await loadOwnedInstance(target, targetKey, body, options.apiKey, options.signal) ?? model.id;
+      return loadAndReport();
     });
   }
 }
@@ -6288,8 +3070,36 @@ function pickSamplingProfile2(quirks, thinkingActive) {
 function isPlainRecord(value) {
   return value !== null && typeof value === "object" && !Array.isArray(value);
 }
-function observeResponseModelIdLine(line, capture) {
-  if (capture.done) return;
+function nonnegativeFiniteNumber(value) {
+  return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : null;
+}
+function backendCompletionTimings(value, source) {
+  if (!isPlainRecord(value)) return null;
+  const promptN = nonnegativeFiniteNumber(value.prompt_n);
+  const predictedN = nonnegativeFiniteNumber(value.predicted_n);
+  const promptMs = nonnegativeFiniteNumber(value.prompt_ms);
+  const predictedMs = nonnegativeFiniteNumber(value.predicted_ms);
+  if (promptN === null || predictedN === null || promptMs === null || predictedMs === null) return null;
+  const cacheN = nonnegativeFiniteNumber(value.cache_n);
+  return {
+    promptTokens: promptN + (cacheN ?? 0),
+    cachedTokens: cacheN,
+    predictedTokens: predictedN,
+    promptMs,
+    predictedMs,
+    source
+  };
+}
+function backendTimingsSourceForModel(model) {
+  const metadata2 = model.clio;
+  if (model.provider === "llamacpp" && metadata2?.runtimeId === "llamacpp") return "llamacpp-timings";
+  if (model.provider === "lmstudio" && metadata2?.runtimeId === "lmstudio") return "lmstudio-timings";
+  return null;
+}
+function captureCanStopEarly(capture) {
+  return capture.modelIdDone && capture.backendTimingsSource === null;
+}
+function observeResponseMetadataLine(line, capture) {
   const normalized = line.endsWith("\r") ? line.slice(0, -1) : line;
   if (!normalized.startsWith("data:")) return;
   const data = normalized.slice("data:".length).trimStart();
@@ -6297,23 +3107,29 @@ function observeResponseModelIdLine(line, capture) {
   try {
     const payload = JSON.parse(data);
     if (!isPlainRecord(payload)) return;
-    const model = payload.model;
-    if (typeof model === "string" && model.trim().length > 0) {
-      capture.reportedModelId = model.trim();
-      capture.done = true;
+    if (!capture.modelIdDone) {
+      const model = payload.model;
+      if (typeof model === "string" && model.trim().length > 0) {
+        capture.reportedModelId = model.trim();
+        capture.modelIdDone = true;
+      }
+    }
+    if (capture.backendTimingsSource !== null) {
+      const timings = backendCompletionTimings(payload.timings, capture.backendTimingsSource);
+      if (timings !== null) capture.backendTimings = timings;
     }
   } catch {
   }
 }
 function observeResponseModelIdBytes(chunk, capture, flush = false) {
-  if (capture.done || !capture.decoder) return;
+  if (!capture.decoder) return;
   capture.buffer += chunk ? capture.decoder.decode(chunk, { stream: !flush }) : capture.decoder.decode();
   let newline = capture.buffer.indexOf("\n");
   while (newline >= 0) {
     const line = capture.buffer.slice(0, newline);
     capture.buffer = capture.buffer.slice(newline + 1);
-    observeResponseModelIdLine(line, capture);
-    if (capture.done) {
+    observeResponseMetadataLine(line, capture);
+    if (captureCanStopEarly(capture)) {
       capture.buffer = "";
       capture.decoder = null;
       return;
@@ -6321,10 +3137,10 @@ function observeResponseModelIdBytes(chunk, capture, flush = false) {
     newline = capture.buffer.indexOf("\n");
   }
   if (flush && capture.buffer.length > 0) {
-    observeResponseModelIdLine(capture.buffer, capture);
+    observeResponseMetadataLine(capture.buffer, capture);
     capture.buffer = "";
   }
-  if (capture.done || flush) capture.decoder = null;
+  if (captureCanStopEarly(capture) || flush) capture.decoder = null;
 }
 function captureResponseModelId(response, capture) {
   if (!response.body || !response.headers.get("content-type")?.toLowerCase().includes("text/event-stream")) {
@@ -6334,12 +3150,12 @@ function captureResponseModelId(response, capture) {
     new TransformStream({
       transform(chunk, controller) {
         capture.observed = true;
-        if (!capture.done) observeResponseModelIdBytes(chunk, capture);
+        observeResponseModelIdBytes(chunk, capture);
         controller.enqueue(chunk);
       },
       flush() {
         capture.observed = true;
-        if (!capture.done) observeResponseModelIdBytes(void 0, capture, true);
+        observeResponseModelIdBytes(void 0, capture, true);
       }
     })
   );
@@ -6349,13 +3165,15 @@ function captureResponseModelId(response, capture) {
     headers: response.headers
   });
 }
-function withResponseModelIdCapture(options, sourceFactory) {
+function withResponseModelIdCapture(model, options, sourceFactory) {
   const capture = {
     reportedModelId: null,
     observed: false,
-    done: false,
+    modelIdDone: false,
+    backendTimings: null,
+    backendTimingsSource: backendTimingsSourceForModel(model),
     buffer: "",
-    decoder: new TextDecoder2()
+    decoder: new TextDecoder()
   };
   const fetchImpl = options.fetch ?? ((input, init) => globalThis.fetch(input, init));
   const capturedOptions = {
@@ -6368,8 +3186,13 @@ function withResponseModelIdCapture(options, sourceFactory) {
     try {
       for await (const event of source) {
         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;
-        else if (event.type === "error") event.error.responseModelIdObservation = observation;
+        if (event.type === "done") {
+          event.message.responseModelIdObservation = observation;
+          if (capture.backendTimings !== null) event.message.backendTimings = capture.backendTimings;
+        } else if (event.type === "error") {
+          event.error.responseModelIdObservation = observation;
+          if (capture.backendTimings !== null) event.error.backendTimings = capture.backendTimings;
+        }
         annotated.push(event);
       }
       annotated.end();
@@ -6652,12 +3475,12 @@ function estimateReasoningTokens(content) {
   if (chars === 0) return 0;
   return Math.max(1, Math.round(chars / REASONING_CHARS_PER_TOKEN2));
 }
-function positiveNumber4(value) {
+function positiveNumber(value) {
   return typeof value === "number" && Number.isFinite(value) && value > 0;
 }
 function hasReportedReasoningUsage(usage) {
   const aliases = usage;
-  return positiveNumber4(aliases.reasoning) || positiveNumber4(aliases.reasoningTokens) || positiveNumber4(aliases.reasoning_tokens);
+  return positiveNumber(aliases.reasoning) || positiveNumber(aliases.reasoningTokens) || positiveNumber(aliases.reasoning_tokens);
 }
 function applyOpenAICompatReasoningEstimate(message) {
   if (hasReportedReasoningUsage(message.usage)) return;
@@ -6864,6 +3687,7 @@ var openAICompletionsApiProvider = {
           stripNeverReasoningFromStream(
             filterGemmaChannelStream(
               withResponseModelIdCapture(
+                model,
                 withRemainingContextBudget(model, effectiveContext, withSamplers),
                 (capturedOptions) => withLocalResidency(
                   model,
@@ -6895,6 +3719,7 @@ var openAICompletionsApiProvider = {
           stripNeverReasoningFromStream(
             filterGemmaChannelStream(
               withResponseModelIdCapture(
+                model,
                 withRemainingContextBudget(model, effectiveContext, withSamplers),
                 (capturedOptions) => withLocalResidency(
                   model,
@@ -6929,14 +3754,14 @@ function registerClioApiProviders() {
 
 // src/domains/providers/knowledge-base-path.ts
 init_esm_shims();
-import { existsSync, statSync as statSync2 } from "node:fs";
-import { delimiter, dirname, join as join3 } from "node:path";
+import { existsSync, statSync } from "node:fs";
+import { delimiter, dirname, join as join2 } from "node:path";
 import { fileURLToPath } from "node:url";
 var MODEL_CATALOG_OVERLAY_DIR = "model-catalog.d";
 var MODEL_CATALOG_DIRS_ENV = "CLIO_CODER_MODEL_CATALOG_DIRS";
 function isDirectory(path) {
   try {
-    return statSync2(path).isDirectory();
+    return statSync(path).isDirectory();
   } catch {
     return false;
   }
@@ -6944,18 +3769,18 @@ function isDirectory(path) {
 function resolveProvidersModelsDir(importMetaUrl) {
   const start = dirname(fileURLToPath(importMetaUrl));
   const directCandidates = [
-    join3(start, "models"),
-    join3(start, "..", "domains", "providers", "models"),
-    join3(start, "..", "providers-models")
+    join2(start, "models"),
+    join2(start, "..", "domains", "providers", "models"),
+    join2(start, "..", "providers-models")
   ];
   for (const candidate of directCandidates) {
     if (isDirectory(candidate)) return candidate;
   }
   let cursor = start;
   for (let i = 0; i < 8; i++) {
-    const packageJson = join3(cursor, "package.json");
-    const sourceModels = join3(cursor, "src", "domains", "providers", "models");
-    const distModels = join3(cursor, "dist", "providers-models");
+    const packageJson = join2(cursor, "package.json");
+    const sourceModels = join2(cursor, "src", "domains", "providers", "models");
+    const distModels = join2(cursor, "dist", "providers-models");
     if (existsSync(packageJson)) {
       if (isDirectory(sourceModels)) return sourceModels;
       if (isDirectory(distModels)) return distModels;
@@ -6986,8 +3811,8 @@ function resolveProviderModelCatalogDirs(importMetaUrl, options = {}) {
   const bundled = resolveProvidersModelsDir(importMetaUrl);
   const cwd = options.cwd ?? process.cwd();
   const overlays = uniqueExistingDirs([
-    join3(resolveClioDirs().config, MODEL_CATALOG_OVERLAY_DIR),
-    join3(cwd, ".clio-coder", MODEL_CATALOG_OVERLAY_DIR),
+    join2(resolveClioDirs().config, MODEL_CATALOG_OVERLAY_DIR),
+    join2(cwd, ".clio-coder", MODEL_CATALOG_OVERLAY_DIR),
     ...envOverlayDirs()
   ]);
   return {
@@ -7006,7 +3831,7 @@ function resolveProviderKnowledgeBaseRoots(importMetaUrl, options = {}) {
 
 // src/domains/providers/plugins.ts
 init_esm_shims();
-import { join as join4 } from "node:path";
+import { join as join3 } from "node:path";
 function extractPluginPackages(settings) {
   if (!settings || typeof settings !== "object") return [];
   const raw = settings.runtimePlugins;
@@ -7015,7 +3840,7 @@ function extractPluginPackages(settings) {
 }
 async function loadPluginRuntimes(registry, settings) {
   const loaded = [];
-  const pluginDir = join4(clioConfigDir(), "runtimes");
+  const pluginDir = join3(clioConfigDir(), "runtimes");
   const packages = extractPluginPackages(settings);
   try {
     const ids = await registry.loadFromDir(pluginDir, activateExternalPluginApiBridge);
@@ -7043,8 +3868,8 @@ async function loadPluginRuntimes(registry, settings) {
 // src/domains/providers/types/knowledge-base.ts
 init_esm_shims();
 var import_yaml = __toESM(require_dist(), 1);
-import { readdirSync as readdirSync2, readFileSync, statSync as statSync3 } from "node:fs";
-import { join as join5 } from "node:path";
+import { readdirSync, readFileSync, statSync as statSync2 } from "node:fs";
+import { join as join4 } from "node:path";
 var FileKnowledgeBase = class {
   roots;
   loaded = [];
@@ -7098,7 +3923,7 @@ function normalizeRoots(root) {
     if (dir.length === 0 || seen.has(dir)) continue;
     let isRootDir = false;
     try {
-      isRootDir = statSync3(dir).isDirectory();
+      isRootDir = statSync2(dir).isDirectory();
     } catch (err) {
       if (raw.optional === true) continue;
       throw err;
@@ -7114,9 +3939,9 @@ function normalizeRoots(root) {
 }
 function collectYamlFiles(dir, prefix = "") {
   const out = [];
-  const entries = readdirSync2(dir, { withFileTypes: true }).sort((a, b) => a.name.localeCompare(b.name));
+  const entries = readdirSync(dir, { withFileTypes: true }).sort((a, b) => a.name.localeCompare(b.name));
   for (const entry of entries) {
-    const path = join5(dir, entry.name);
+    const path = join4(dir, entry.name);
     const name = prefix ? `${prefix}/${entry.name}` : entry.name;
     if (entry.isDirectory()) {
       out.push(...collectYamlFiles(path, name));
@@ -7135,20 +3960,20 @@ function normalizeEntry(raw, file) {
   const candidate = raw;
   const family = candidate.family;
   const patterns = candidate.matchPatterns;
-  const capabilities2 = candidate.capabilities;
+  const capabilities = candidate.capabilities;
   if (typeof family !== "string" || family.length === 0) {
     throw new Error(`knowledge base file ${file}: entry is missing 'family' string`);
   }
   if (!Array.isArray(patterns) || patterns.some((p) => typeof p !== "string")) {
     throw new Error(`knowledge base file ${file}: entry '${family}' needs matchPatterns: string[]`);
   }
-  if (typeof capabilities2 !== "object" || capabilities2 === null || Array.isArray(capabilities2)) {
+  if (typeof capabilities !== "object" || capabilities === null || Array.isArray(capabilities)) {
     throw new Error(`knowledge base file ${file}: entry '${family}' needs capabilities object`);
   }
   const entry = {
     family,
     matchPatterns: patterns,
-    capabilities: capabilities2
+    capabilities
   };
   if (candidate.quirks !== void 0) {
     if (typeof candidate.quirks !== "object" || candidate.quirks === null || Array.isArray(candidate.quirks)) {
@@ -7198,9 +4023,9 @@ function capabilitiesFor(desc, target, probe, kb) {
   const base = capabilitiesFromCatalogModel(desc.defaultCapabilities, catalogModel);
   const probeCaps = probeCapabilitiesForModel({ target, ...probe }, target.defaultModel);
   const userOverride = target.capabilities ?? null;
-  const capabilities2 = mergeCapabilities(base, kbHit?.entry.capabilities ?? null, probeCaps, userOverride);
+  const capabilities = mergeCapabilities(base, kbHit?.entry.capabilities ?? null, probeCaps, userOverride);
   return {
-    capabilities: capabilities2,
+    capabilities,
     contextWindowProvenance: contextWindowProvenanceOf(kbHit, catalogModel, probeCaps, userOverride)
   };
 }
@@ -7255,6 +4080,15 @@ function mergeProbeResult(desc, target, probe, previous) {
   if (probeSurfaces && Object.keys(probeSurfaces).length > 0) merge.probeSurfaces = probeSurfaces;
   return merge;
 }
+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 (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`;
+}
 function createProvidersBundle(context) {
   const registry = getRuntimeRegistry();
   const authStore = openAuthStorage();
@@ -7330,12 +4164,13 @@ function createProvidersBundle(context) {
     }
     const availability = availabilityFor(desc, target, authStatusFor);
     const merge = mergeProbeResult(desc, target, probe, previous);
-    const { capabilities: capabilities2, contextWindowProvenance } = capabilitiesFor(desc, target, merge, kb);
+    const { capabilities, contextWindowProvenance } = capabilitiesFor(desc, target, merge, kb);
     const healthy = probe !== null ? probe.ok : null;
+    const unservedDefault = probe?.ok ? unservedDefaultModelReason(desc, target, merge) : null;
     const health = probe === null ? previous?.health ?? emptyHealth() : {
-      status: healthy ? "healthy" : "down",
+      status: healthy ? unservedDefault === null ? "healthy" : "degraded" : "down",
       lastCheckAt: (/* @__PURE__ */ new Date()).toISOString(),
-      lastError: probe.error ?? null,
+      lastError: probe.error ?? unservedDefault,
       latencyMs: probe.latencyMs ?? null
     };
     const available = availability.available && (probe === null || probe.ok);
@@ -7346,7 +4181,7 @@ function createProvidersBundle(context) {
       available,
       reason,
       health,
-      capabilities: capabilities2,
+      capabilities,
       contextWindowProvenance,
       probeCapabilities: merge.probeCapabilities,
       probeModelCapabilities: merge.probeModelCapabilities,
@@ -7677,12 +4512,6 @@ function resolveModelReference(rawPattern, providers) {
   return buildResult(base, pickFirst(partial), thinkingLevel);
 }
 
-// src/domains/providers/types/cost-provenance.ts
-init_esm_shims();
-function normalizeCostProvenance(value) {
-  return value ?? "unknown";
-}
-
 // src/domains/providers/index.ts
 var ProvidersDomainModule = {
   manifest: ProvidersManifest,
@@ -7690,14 +4519,15 @@ var ProvidersDomainModule = {
 };
 
 export {
-  supportsAgentRoleTools,
-  AGENT_ROLE_TOOLS_REQUIRED_REASON,
-  resolveEffectivePricing,
-  getCatalogModelForRuntime,
-  getRuntimeRegistry,
-  credentialsPresent,
+  residencyTargetKey,
   setGlobalDefaultMaxOutputTokens,
   resolveReservedOutputTokens,
+  canonicalEndpointKey,
+  endpointLabel,
+  endpointCapacityForStatus,
+  endpointCapacitiesForStatuses,
+  registerForegroundStream,
+  foregroundStreamUsage,
   setResidencyNoticeSink,
   setProtectedModelsProvider,
   withRunOverrides,
@@ -7710,22 +4540,12 @@ export {
   thinkingLevelFromChoiceLabel,
   resolveModelRuntimeCapabilitiesForProviders,
   resolveModelRuntimeCapabilitiesForModel,
-  greetLmStudio,
   registerClioApiProviders,
   resolveProviderKnowledgeBaseRoots,
   loadPluginRuntimes,
-  formatContextWindowSlots,
-  registerBuiltinRuntimes,
   FileKnowledgeBase,
   isOrchestratorEligibleRuntime,
   isDispatchEligibleRuntime,
-  supportGroupLabel,
-  listKnownModelsForRuntime,
-  describeRuntimeModels,
-  buildProviderSupportEntry,
-  listProviderSupportEntries,
-  configuredTargetsForRuntime,
-  resolveProviderReference,
   modelResidencyForStatus,
   modelCandidatesForStatus,
   modelIdsForStatus,
@@ -7740,4 +4560,4 @@ export {
   normalizeCostProvenance,
   ProvidersDomainModule
 };
-//# sourceMappingURL=chunk-VQNODYQ4.js.map
+//# sourceMappingURL=chunk-MIX5N5AC.js.map