From 7ddc82f25da34a2648a27c2b5f892c7f7916d5ad Mon Sep 17 00:00:00 2001
From: Prakhar Pandey <prakhar9999pandey@gmail.com>
Date: Tue, 14 Jul 2026 13:10:52 +0530
Subject: [PATCH] feat: remove model-name entry from onboarding to match
 connect-only settings

Onboarding still asked for a model name for openrouter/aigateway/ollama/
openai-compatible. It now mirrors the settings dialog: entering a key (or
base URL) is enough and the model is resolved silently from the fetched
list (same precedence), with openai-compatible keeping a visible Model
field as its escape hatch. Save payload shape and multi-provider flow
unchanged.
---
 .../onboarding/steps/llm-setup-step.tsx       | 21 +++++++---------
 .../onboarding/use-onboarding-state.ts        | 25 +++++++++++++------
 2 files changed, 27 insertions(+), 19 deletions(-)

diff --git a/apps/x/apps/renderer/src/components/onboarding/steps/llm-setup-step.tsx b/apps/x/apps/renderer/src/components/onboarding/steps/llm-setup-step.tsx
index 4ece02c9..86787753 100644
--- a/apps/x/apps/renderer/src/components/onboarding/steps/llm-setup-step.tsx
+++ b/apps/x/apps/renderer/src/components/onboarding/steps/llm-setup-step.tsx
@@ -42,14 +42,11 @@ export function LlmSetupStep({ state }: LlmSetupStepProps) {
   } = state
 
   const isMoreProvider = moreProviders.some(p => p.id === llmProvider)
-  // Hosted providers (openai/anthropic/google) get a default model, so we only
-  // ask for a model on providers that truly need one (local/custom/gateway),
-  // or as a fallback if no model is set yet.
-  // Hosted providers (openai/anthropic/google) fetch their models from the API
-  // key on test, so they never need a manual model field. Only local/custom/
-  // gateway providers, where the user must specify a model, show the input.
-  const hostedProviders: LlmProviderFlavor[] = ["openai", "anthropic", "google"]
-  const showModelInput = !hostedProviders.includes(llmProvider)
+  // Connect-only, mirroring Settings: entering a key (or base URL) is enough
+  // and the model is resolved silently at save. openai-compatible is the sole
+  // exception with a visible Model field — its /models endpoint often doesn't
+  // exist, so a typed value must be able to win.
+  const showModelInput = llmProvider === "openai-compatible"
 
   const renderProviderCard = (provider: typeof primaryProviders[0], index: number) => {
     const isSelected = llmProvider === provider.id
@@ -150,9 +147,9 @@ export function LlmSetupStep({ state }: LlmSetupStepProps) {
 
       {/* Provider configuration */}
       <div className="space-y-4">
-        {/* Cloud providers get a default model auto-selected; only local/custom
-            providers (no catalog) need a model here. Users can pick any of the
-            provider's models later in the chat view. */}
+        {/* Every provider resolves its model silently at save. openai-compatible
+            alone keeps this field, since its /models is unreliable and a typed
+            value must win; leaving it blank auto-selects from the fetched list. */}
         {showModelInput && (
           <div className="space-y-2">
             <label className="text-xs font-medium text-muted-foreground">
@@ -167,7 +164,7 @@ export function LlmSetupStep({ state }: LlmSetupStepProps) {
               <Input
                 value={activeConfig.model}
                 onChange={(e) => updateProviderConfig(llmProvider, { model: e.target.value })}
-                placeholder="Enter model"
+                placeholder="Model ID (leave empty to auto-select)"
               />
             )}
             {modelsError && (
diff --git a/apps/x/apps/renderer/src/components/onboarding/use-onboarding-state.ts b/apps/x/apps/renderer/src/components/onboarding/use-onboarding-state.ts
index 8a9679be..a0458bac 100644
--- a/apps/x/apps/renderer/src/components/onboarding/use-onboarding-state.ts
+++ b/apps/x/apps/renderer/src/components/onboarding/use-onboarding-state.ts
@@ -457,13 +457,24 @@ export function useOnboardingState(open: boolean, onComplete: (opts?: { startTou
         return false
       }
 
-      const preferred = preferredDefaults[llmProvider]
-      // A model the user explicitly entered always wins — this used to prefer
-      // catalog[0], which silently replaced the user's Ollama model with
-      // whatever model the local server happened to list first.
-      const model = modelInputShown
-        ? (typed || catalog[0] || "")
-        : ((preferred && catalog.includes(preferred) && preferred) || catalog[0] || typed || "")
+      // Resolve the model silently — same precedence as the settings dialog's
+      // resolvedModel, so onboarding no longer asks users to type one.
+      // openai-compatible keeps a visible Model field (its /models is
+      // unreliable), so a typed value there wins; otherwise the typed/saved
+      // model if the fetched list still has it, else the flavor's preferred
+      // default, else the first fetched id. An empty list falls back to
+      // whatever was typed/seeded.
+      let model: string
+      if (llmProvider === "openai-compatible" && typed) {
+        model = typed
+      } else if (catalog.length > 0) {
+        const preferred = preferredDefaults[llmProvider]
+        model = (typed && catalog.includes(typed))
+          ? typed
+          : ((preferred && catalog.includes(preferred)) ? preferred : catalog[0])
+      } else {
+        model = typed
+      }
 
       // `models` is the user's curated assistant-model list (shown in Settings),
       // NOT the full provider catalog. Onboarding seeds it with just the selected