logic fix and ci improvement

.github/workflows/ci.yml

@@ -30,7 +30,7 @@ jobs:
         uses: useblacksmith/setup-node@v5
         with:
           node-version: 22
-          cache: 'pnpm'
+          # Remove cache here since we're using stickydisk for pnpm store
           
       - name: Get pnpm store directory
         shell: bash
@@ -49,8 +49,21 @@ jobs:
           key: ${{ github.repository }}-turbo-cache
           path: ./.turbo
 
+      - name: Check if lockfile changed
+        id: lockfile-check
+        run: |
+          if git diff HEAD~1 HEAD --name-only | grep -q "pnpm-lock.yaml"; then
+            echo "changed=true" >> $GITHUB_OUTPUT
+          else
+            echo "changed=false" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Fetch dependencies (if lockfile changed)
+        if: steps.lockfile-check.outputs.changed == 'true'
+        run: pnpm fetch --frozen-lockfile
+
       - name: Install dependencies
-        run: pnpm install --frozen-lockfile
+        run: pnpm install --frozen-lockfile --prefer-offline
 
       - name: Build all packages (excluding web)
         run: pnpm build --filter='!@buster-app/web'

packages/ai/src/utils/models/ai-fallback.ts

@@ -174,6 +174,9 @@ export class FallbackModel implements LanguageModelV1 {
       const result = await currentModel.doStream(options);
 
       let hasStreamedAny = false;
+      let streamRetryAttempts = 0;
+      const maxStreamRetries = self.settings.models.length - 1; // -1 because we already tried one
+
       // Wrap the stream to handle errors and switch providers if needed
       const wrappedStream = new ReadableStream<LanguageModelV1StreamPart>({
         async start(controller) {
@@ -191,9 +194,13 @@ export class FallbackModel implements LanguageModelV1 {
             if (self.settings.onError) {
               await self.settings.onError(error as Error, self.modelId);
             }
-            if (!hasStreamedAny || self.retryAfterOutput) {
-              // If nothing was streamed yet, switch models and retry
+            if (
+              (!hasStreamedAny || self.retryAfterOutput) &&
+              streamRetryAttempts < maxStreamRetries
+            ) {
+              // If nothing was streamed yet and we haven't exhausted retries, switch models and retry
               self.switchToNextModel();
+              streamRetryAttempts++;
               try {
                 const nextResult = await self.doStream(options);
                 const nextReader = nextResult.stream.getReader();