fix: unify context limit resolution behind provider API (#11213)

Signed-off-by: Jack Amadeo <jackamadeo@squareup.com>

ui/desktop/src/acp/__tests__/gooseSessionNotifications.test.ts

@@ -160,6 +160,7 @@ describe('applyGooseSessionNotification', () => {
           type: 'tokenState',
           tokenState: {
             totalTokens: 42,
+            contextLimit: 200,
             accumulatedInputTokens: 10,
             accumulatedOutputTokens: 15,
             accumulatedTotalTokens: 25,

ui/desktop/src/acp/__tests__/providers.test.ts

@@ -219,7 +219,7 @@ describe('ACP providers', () => {
 
     expect(result.connectionChecked).toBe(true);
     expect(result.provider.metadata.known_models).toEqual([
-      { name: 'claude-sonnet', context_limit: 0, reasoning: undefined },
+      { name: 'claude-sonnet', context_limit: undefined, reasoning: undefined },
     ]);
   });
 
@@ -268,7 +268,7 @@ describe('ACP providers', () => {
     });
     expect(enabled.is_configured).toBe(true);
     expect(enabled.metadata.known_models).toEqual([
-      { name: 'claude-sonnet', context_limit: 0, reasoning: undefined },
+      { name: 'claude-sonnet', context_limit: undefined, reasoning: undefined },
     ]);
   });
 

ui/desktop/src/acp/__tests__/sessionNotificationAdapter.test.ts

@@ -872,6 +872,7 @@ describe('createAcpSessionNotificationAdapter', () => {
           type: 'tokenState',
           tokenState: {
             totalTokens: 42,
+            contextLimit: 200,
             accumulatedInputTokens: 10,
             accumulatedOutputTokens: 15,
             accumulatedTotalTokens: 25,

ui/desktop/src/acp/adapter/gooseSessionNotifications.ts

@@ -15,6 +15,7 @@ export function applyGooseSessionNotification(
           type: 'tokenState',
           tokenState: {
             totalTokens: update.used,
+            contextLimit: update.contextLimit,
             accumulatedInputTokens: update.accumulatedInputTokens,
             accumulatedOutputTokens: update.accumulatedOutputTokens,
             accumulatedTotalTokens: update.accumulatedInputTokens + update.accumulatedOutputTokens,

ui/desktop/src/acp/providers.ts

@@ -71,7 +71,7 @@ function providerEntryToDetails(entry: ProviderInventoryEntryDto): ProviderDetai
       })),
       known_models: entry.models.map((model) => ({
         name: model.id,
-        context_limit: model.contextLimit ?? 0,
+        context_limit: model.contextLimit ?? undefined,
         reasoning: model.reasoning ?? undefined,
       })),
       setup_steps: entry.setupSteps,

ui/desktop/src/components/BaseChat.tsx

@@ -522,6 +522,7 @@ export default function BaseChat({
             initialValue={initialPrompt}
             setView={setView}
             totalTokens={tokenState?.totalTokens ?? session?.usage?.total_tokens ?? undefined}
+            contextLimit={tokenState?.contextLimit}
             accumulatedInputTokens={
               tokenState?.accumulatedInputTokens ??
               session?.accumulated_usage?.input_tokens ??

ui/desktop/src/components/ChatInput.tsx

@@ -82,7 +82,7 @@ const removeQueuedMessage = (messages: QueuedMessage[], messageId: string): Queu
 
 const MAX_IMAGES_PER_MESSAGE = 10;
 
-const TOKEN_LIMIT_DEFAULT = 128000; // fallback for custom models that the backend doesn't know about
+const TOKEN_LIMIT_DEFAULT = 128000; // used before a session has a backend-resolved limit
 
 const getContextAlertType = (totalTokens: number, tokenLimit: number): AlertType => {
   const percentage = tokenLimit ? (totalTokens / tokenLimit) * 100 : 0;
@@ -172,6 +172,7 @@ interface ChatInputProps {
   onFilesProcessed?: () => void;
   setView: (view: View) => void;
   totalTokens?: number;
+  contextLimit?: number;
   accumulatedInputTokens?: number;
   accumulatedOutputTokens?: number;
   accumulatedCost?: number | null;
@@ -207,6 +208,7 @@ export default function ChatInput({
   onFilesProcessed,
   setView,
   totalTokens,
+  contextLimit,
   accumulatedInputTokens,
   accumulatedOutputTokens,
   accumulatedCost,
@@ -581,7 +583,11 @@ export default function ChatInput({
   // Load providers and get current model's token limit
   const loadProviderDetails = async () => {
     try {
-      // Reset token limit loaded state
+      if (sessionId) {
+        setTokenLimit(0);
+        setIsTokenLimitLoaded(false);
+        return;
+      }
       setIsTokenLimitLoaded(false);
 
       // Use effective model/provider (includes overrides from in-session model changes),
@@ -642,7 +648,20 @@ export default function ChatInput({
   useEffect(() => {
     loadProviderDetails();
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [effectiveModel, effectiveProvider, configModel, configProvider]);
+  }, [effectiveModel, effectiveProvider, configModel, configProvider, sessionId]);
+
+  useEffect(() => {
+    if (contextLimit === undefined) {
+      if (sessionId) {
+        setTokenLimit(0);
+        setIsTokenLimitLoaded(false);
+      }
+      return;
+    }
+
+    setTokenLimit(contextLimit);
+    setIsTokenLimitLoaded(true);
+  }, [contextLimit, sessionId]);
 
   // Handle token usage alerts
   useEffect(() => {

ui/desktop/src/components/settings/providers/modal/subcomponents/forms/CustomProviderForm.tsx

@@ -459,8 +459,8 @@ export default function CustomProviderForm({
 
     const modelList = models
       .split(',')
-      .map((m) => m.trim())
-      .filter((m) => m);
+      .map((name) => name.trim())
+      .filter(Boolean);
 
     let allHeaders = [...headers];
 

ui/desktop/src/types/chat.ts

@@ -13,6 +13,7 @@ export type TokenState = {
   inputTokens: number;
   outputTokens: number;
   totalTokens: number;
+  contextLimit?: number;
 };
 
 export interface ChatType {

ui/desktop/src/types/providers.ts

@@ -13,7 +13,7 @@ export type ConfigKey = {
 };
 
 export type ModelInfo = {
-  context_limit: number;
+  context_limit?: number | null;
   currency?: string | null;
   input_token_cost?: number | null;
   name: string;