diff --git a/plugins/web-ui/src/model-options.ts b/plugins/web-ui/src/model-options.ts
index 5ee2c799..a0a0975c 100644
--- a/plugins/web-ui/src/model-options.ts
+++ b/plugins/web-ui/src/model-options.ts
@@ -82,9 +82,8 @@ function buildOption(
 ): ModelOption | null {
   try {
     const dynamic = catalog[id];
-    const meta = MODEL_CATALOG[id] ?? (dynamic ? { label: dynamic.name, buttonLabel: dynamic.name } : null);
-    if (!meta) return null;
-    const model = getBaseModel(id, dynamic);
+    const meta = MODEL_CATALOG[id] ?? (dynamic ? { label: dynamic.name, buttonLabel: dynamic.name } : { label: id, buttonLabel: id });
+    const model = getBaseModel(id, dynamic ?? { name: id, provider: "custom" });
     return {
       value: qualified ? `${harnessId}:${id}` : id,
       harnessId,
@@ -157,10 +156,11 @@ export function runtimeModelOptions(
   catalog: Readonly<Record<string, { name: string; provider: string }>> = {},
 ): ModelOption[] {
   const options = approvedHarnesses.flatMap((harnessId) => {
-    const configured = buildOptions(modelsByHarness[harnessId] ?? [], harnessId, true, catalog);
-    return configured.length
-      ? configured
-      : buildOptions(defaultModelIdsForHarness(harnessId), harnessId, true, catalog);
+    const serverModels = modelsByHarness[harnessId];
+    if (serverModels) {
+      return buildOptions(serverModels, harnessId, true, catalog);
+    }
+    return buildOptions(defaultModelIdsForHarness(harnessId), harnessId, true, catalog);
   });
   return options.length ? options : buildOptions(DEFAULT_PICKER_MODEL_IDS);
 }
diff --git a/plugins/web-ui/src/pi-models.ts b/plugins/web-ui/src/pi-models.ts
index ca2128a3..7e47f55f 100644
--- a/plugins/web-ui/src/pi-models.ts
+++ b/plugins/web-ui/src/pi-models.ts
@@ -30,10 +30,13 @@ export function getBaseModel(id: string, fallback?: { name: string; provider: st
     const template = builtinModel(clone.template);
     if (template) return cloneModel(template, id, clone.name);
   }
-  if (fallback?.provider === "openrouter") {
-    const template = getModel("openrouter", "openrouter/auto" as Parameters<typeof getModel>[1]) as PiModel | undefined;
+  if (fallback) {
+    const providerKey = (KNOWN_PROVIDERS.includes(fallback.provider as any) ? fallback.provider : "openai") as any;
+    const template = (getModel(providerKey, "gpt-4o" as any) ?? getModel("openai", "gpt-4o" as any)) as PiModel | undefined;
     if (template) return cloneModel(template, id, fallback.name);
   }
+  const defaultTemplate = getModel("openai", "gpt-4o" as any) as PiModel | undefined;
+  if (defaultTemplate) return cloneModel(defaultTemplate, id, id);
   throw new Error(`Unsupported model: ${id}`);
 }
 
diff --git a/src/api/routes/surface.ts b/src/api/routes/surface.ts
index 3fbe270a..7aa2a3db 100644
--- a/src/api/routes/surface.ts
+++ b/src/api/routes/surface.ts
@@ -1034,11 +1034,13 @@ async function getSurfaceConfig(ctx: ApiCtx): Promise<void> {
     ...(mark ? { mark } : {}),
     ...(selfLabel ? { selfLabel } : {}),
   };
+  const customCount = deps.customProviders ? (await deps.customProviders.enabled()).length : 0;
+  const isConfigured = (managedKeys ? Object.values(managedKeys).some(Boolean) : false) || customCount > 0;
   return sendJson(res, 200, {
     webuiModels: configuredPicker.length ? configuredPicker : allowed,
     baseModel: resolvedBase,
     harnessId,
-    ...(managedKeys ? { modelProviderConfigured: Object.values(managedKeys).some(Boolean) } : {}),
+    modelProviderConfigured: isConfigured,
     externalSlackParticipants,
     ...(Object.keys(resolvedBranding).length ? { branding: resolvedBranding } : {}),
   });
diff --git a/src/core/turn-options.ts b/src/core/turn-options.ts
index 2f8990c0..88431d76 100644
--- a/src/core/turn-options.ts
+++ b/src/core/turn-options.ts
@@ -1,3 +1,4 @@
+import { customModelCatalog } from "../model/custom-providers.ts";
 import {
   DEFAULT_WEBUI_MODEL_IDS,
   THINKING_LEVELS,
@@ -49,7 +50,8 @@ export function validateWebTurnModelOptions(
   enabledModels: readonly string[] | null,
   providers: ModelProviderAvailability = ALL_PROVIDERS_AVAILABLE,
 ): string | null {
-  const enabled = enabledModels?.length ? enabledModels : DEFAULT_WEBUI_MODEL_IDS;
+  const customIds = customModelCatalog().map((m) => m.id);
+  const enabled = enabledModels?.length ? enabledModels : [...DEFAULT_WEBUI_MODEL_IDS, ...customIds];
   const allowedModels = serviceableModelIds(enabled, providers);
   if (input.model && !allowedModels.includes(input.model)) {
     return resolveModel(input.model) && !modelServiceable(input.model, providers)
diff --git a/src/model/custom-providers.ts b/src/model/custom-providers.ts
index cb2a92c2..36f0e96b 100644
--- a/src/model/custom-providers.ts
+++ b/src/model/custom-providers.ts
@@ -85,6 +85,7 @@ export interface CustomRuntimeModel {
   id: string;
   name: string;
   provider: string;
+  protocol?: CustomProviderProtocol;
   api: "openai-completions" | "anthropic-messages";
   baseUrl: string;
   reasoning: boolean;
@@ -102,6 +103,7 @@ function toRuntimeModel(provider: CustomProviderSpec, m: CustomModelSpec): Custo
     id: m.id,
     name: m.name?.trim() || m.id,
     provider: provider.id,
+    protocol: provider.protocol,
     api: provider.protocol === "anthropic" ? "anthropic-messages" : "openai-completions",
     baseUrl: provider.baseUrl,
     reasoning: false,
diff --git a/src/model/pi-models.ts b/src/model/pi-models.ts
index 7d193eec..26f0014e 100644
--- a/src/model/pi-models.ts
+++ b/src/model/pi-models.ts
@@ -175,12 +175,19 @@ export function contextTokenBudgetForModel(id: string): number | undefined {
 
 export function modelSupportedByHarness(id: string | undefined, harness: string): boolean {
   if (!id) return false;
-  if (isCustomModelId(id) && !REGISTRY_BY_ID.has(id))
-    return harness === "pi" || harness === "opencode" || harness === "mock";
+  if (isCustomModelId(id) && !REGISTRY_BY_ID.has(id)) {
+    const custom = resolveCustomModel(id);
+    if (harness === "pi" || harness === "opencode" || harness === "mock") return true;
+    if (harness === "claude") return custom?.protocol === "anthropic" || /^claude-/i.test(id);
+    if (harness === "codex") return custom?.protocol === "openai" || /^(?:gpt-|o\d|codex|openai\/)/i.test(id);
+    return true;
+  }
   if (harness === "pi" || harness === "opencode" || harness === "mock") return Boolean(resolveModel(id));
-  const provider = resolveModel(id)?.provider;
-  if (harness === "claude") return provider === "anthropic" || /^claude-/i.test(id);
-  if (harness === "codex") return provider === "openai" || /^(?:gpt-|o\d|codex|openai\/)/i.test(id);
+  const model = resolveModel(id);
+  const provider = model?.provider;
+  const protocol = (model as unknown as { protocol?: string })?.protocol;
+  if (harness === "claude") return provider === "anthropic" || protocol === "anthropic" || /^claude-/i.test(id);
+  if (harness === "codex") return provider === "openai" || protocol === "openai" || /^(?:gpt-|o\d|codex|openai\/)/i.test(id);
   return false;
 }
 
@@ -205,11 +212,12 @@ export interface ModelProviderAvailability {
 }
 
 export function modelServiceable(id: string, providers: ModelProviderAvailability): boolean {
-  const provider = resolveModel(id)?.provider;
-  if (!provider) return false;
+  const model = resolveModel(id);
+  if (!model) return false;
+  const provider = model.provider;
   if (isCustomModelId(id) && !REGISTRY_BY_ID.has(id)) return true;
-  if (provider === "openai") return providers.openai;
-  if (provider === "anthropic") return providers.anthropic;
+  if (provider === "openai") return providers.openai || Boolean(process.env.OPENAI_BASE_URL);
+  if (provider === "anthropic") return providers.anthropic || Boolean(process.env.ANTHROPIC_BASE_URL);
   if (provider === "openrouter") return providers.openrouter;
   return true;
 }
@@ -225,9 +233,20 @@ export function modelProviderAvailabilityFor(
   configKeys: ModelProviderAvailability,
   managedKeys: ModelProviderAvailability = configKeys,
 ): ModelProviderAvailability {
-  if (harness === "pi") return managedKeys;
-  if (harness === "opencode") return { ...configKeys, openrouter: false };
-  if (harness === "codex") return configKeys;
+  const effectiveConfig = {
+    ...configKeys,
+    openai: configKeys.openai || Boolean(process.env.OPENAI_BASE_URL),
+    anthropic: configKeys.anthropic || Boolean(process.env.ANTHROPIC_BASE_URL),
+  };
+  const effectiveManaged = {
+    ...managedKeys,
+    openai: managedKeys.openai || Boolean(process.env.OPENAI_BASE_URL),
+    anthropic: managedKeys.anthropic || Boolean(process.env.ANTHROPIC_BASE_URL),
+  };
+  if (harness === "pi") return effectiveManaged;
+  if (harness === "opencode") return { ...effectiveConfig, openrouter: false };
+  if (harness === "codex") return effectiveConfig;
+  if (harness === "claude") return effectiveConfig;
   return ALL_PROVIDERS_AVAILABLE;
 }
 
diff --git a/test/custom-provider-e2e.test.ts b/test/custom-provider-e2e.test.ts
index cc51db83..50cfb124 100644
--- a/test/custom-provider-e2e.test.ts
+++ b/test/custom-provider-e2e.test.ts
@@ -149,7 +149,7 @@ test("QA: full custom-provider lifecycle against a live fake upstream", async ()
     assert.equal((model as { baseUrl?: string }).baseUrl, upstreamUrl);
     assert.equal(modelSupportedByHarness("qa-chat", "pi"), true);
     assert.equal(modelSupportedByHarness("qa-chat", "opencode"), true);
-    assert.equal(modelSupportedByHarness("qa-chat", "codex"), false);
+    assert.equal(modelSupportedByHarness("qa-chat", "codex"), true);
     assert.equal(modelServiceable("qa-chat", { anthropic: false, openai: false, openrouter: false }), true);
 
     // 6. REAL model call through QM's pi path → fake upstream answers