Spaces:
Sleeping
Sleeping
Commit History
-
feat: Wrap Flutter apps in iPhone 15 Pro device frame for DartPad preview
AI Software Factoryd535ad0 -
fix: Add GitHub token authentication for Gist creation
AI Software FactoryThe Simple Generator's auto-load DartPad feature requires GitHub authentication: - Add GITHUB_TOKEN environment variable support - Update create_anonymous_gist() to use Bearer token auth - Improve error messages with clear setup instructions - Add fallback UI when token is not configured - Document GITHUB_TOKEN setup in README Without token, users get helpful message explaining how to configure it. With token, Gist creation works and DartPad auto-loads code. Setup: Add GITHUB_TOKEN secret in HF Space settings with 'gist' scope.be2dec1 -
feat: Auto-load Flutter code in DartPad via Gist
AI Software Factory- Create anonymous GitHub Gist automatically with generated code - Embed DartPad with code pre-loaded (no manual copy-paste) - Hide source code in collapsible Accordion - Add 'Open in DartPad Full Screen' button - Update UI with success/fallback states - Improve user experience with automatic workflow User can now just click 'Run' in DartPad - code is already there!9b549f2 -
Add DartPad embedded iframe to Simple Generator for live testing
Claude17d3586 -
Add Simple Flutter Generator: direct prompt to DeepSeek, no presets
Claude71f2499 -
Improve custom prompt usage: AI now generates based on user task description
Claude6dff68d -
Add custom prompt field to AC3 for user-defined screen generation
Claudea740077 -
Fix AC3: Force AI to generate cards in variants + add debug logging
Claudefe97795 -
Simplify AC3 prompt: generate ONE home page in 4 different visual versions
Claudee2cbbd9 -
Add layout_type support for generating 4 DIFFERENT home screen structures
ClaudeThe problem: All 4 variants had the same vertical list structure The solution: AI now generates different layout_type for each variant Changes: - Added layout_type field to AI prompt (grid_2x2, vertical_list, hero_card, timeline, featured_grid) - Updated JSON output to include layout_type - Added layout_type to fitness fallback variants - Created _generate_body_content() method that generates different Dart layouts: * grid_2x2: GridView with 2 columns for dashboards/metrics * hero_card: Large featured card + smaller cards below * timeline: ListView for chronological events * vertical_list: Classic scrollable column (default) - Replaced hardcoded Expanded(SingleChildScrollView(...)) with dynamic body_content Now each variant has a DIFFERENT visual structure, not just different colors! Example for fitness app: 1. "Dashboard" → Grid 2x2 with metrics 2. "Focus Mode" → Hero card with large workout info 3. "Progress" → Timeline of achievements 4. "Social" → Vertical list of activi…67f1922 -
Add detailed logging to debug DeepSeek API calls
Claudeb5af2c0 -
Fix AC3: Remove leftover hf_token reference, use deepseek_api_key
Claude9a7bb3a -
Fix AC3: Integrate DeepSeek API to replace HuggingFace Inference
ClaudeHuggingFace Serverless Inference API no longer supports most LLMs for free. Switched to DeepSeek Cloud API which offers 500k free tokens. Changes: - Removed huggingface_hub.InferenceClient dependency - Added requests library for HTTP calls - Created _call_deepseek_api() method for DeepSeek API calls - Changed model to "deepseek-coder" (DeepSeek Coder V2) - Updated all 3 API calls to use DeepSeek instead of HuggingFace - API endpoint: https://api.deepseek.com/v1/chat/completions - Uses DEEPSEEK_API_KEY environment variable - OpenAI-compatible API format DeepSeek Coder is specifically designed for code generation tasks, making it perfect for generating Flutter/Dart code and JSON schemas.51cc484 -
Fix AC3: Switch to chat_completion API with Zephyr-7B-Beta
ClaudeHuggingFace changed their Serverless Inference API in July 2025: - text_generation endpoint no longer supports generative models - Must use chat_completion endpoint instead Changes: - Changed model to "HuggingFaceH4/zephyr-7b-beta" (supports chat_completion) - Replaced all text_generation() calls with chat_completion() - Updated message format to use messages=[{"role": "user", "content": prompt}] - Changed max_new_tokens to max_tokens (chat_completion parameter) - Extract response from response.choices[0].message.content - Updated docstrings to reflect Zephyr 7B model This should fix the 404 errors and enable AI-powered variant generation.13322d6 -
Fix AC3: Switch from Llama 3.1 70B to Llama 3.2 3B Instruct
ClaudeLlama 3.1 70B (70 billion parameters) is too large for HuggingFace's free Serverless Inference API (limit ~10GB). This was causing 404 errors. Changes: - Updated model from "meta-llama/Llama-3.1-70B-Instruct" to "meta-llama/Llama-3.2-3B-Instruct" - Llama 3.2 3B Instruct is available for FREE on HF Inference API - Updated documentation comments to reflect the new model - Model is still capable of generating JSON for UI variants This should fix the 404 errors and enable AI-powered variant generation.ea6878f -
Fix AC3: Replace all invalid card IDs with valid CARD_CATALOG entries
Claude- Updated AI prompt card suggestions to use only valid card IDs - Fixed fitness fallback variants (workout_exercise, calorie_goal, etc.) - Fixed banking fallback variants (transfer_form, investment_item, etc.) - Fixed rider fallback variants (delivery_tracker, map_location, etc.) - Fixed ecommerce fallback variants (brand_showcase, category_grid, etc.) - Fixed social fallback variants (social_post_card, reel_preview, etc.) - Fixed generic default variants with valid card types This ensures _generate_cards_code() can find all cards in CARD_CATALOG and generate proper Flutter widgets instead of falling back to generic cards.37d0b8b -
Fix AC3: Correct Llama model ID from Meta-Llama to Llama
Claude- Changed model from "meta-llama/Meta-Llama-3.1-70B-Instruct" to "meta-llama/Llama-3.1-70B-Instruct" - Fixes 404 error when calling HuggingFace Inference API7bb506c -
Fix AC3: Generate dynamic cards from AI selection instead of hardcoded cards
Claude- Added _generate_cards_code() method to dynamically generate card widgets - Added _StatsCard, _ActionCard, and _ProductCard Flutter widgets - Modified _generate_complete_screen() to use AI-selected cards from variant - Cards now personalized based on app_type (fitness, banking, rider, ecommerce, social) - Replaced hardcoded "Feature 1/2/3" cards with contextual content2489e19 -
Upload folder using huggingface_hub
Verified signatureb57ecd3 -
Upload app.py with huggingface_hub
Verified signature77de16a -
Upload folder using huggingface_hub
Verified signaturece3921a -
Upload app.py with huggingface_hub
Verified signaturef5d0371 -
Upload app.py with huggingface_hub
Verified signaturef60e4fa -
Upload folder using huggingface_hub
Verified signature6838a74 -
Upload app.py with huggingface_hub
Verified signaturebb153f4 -
Upload folder using huggingface_hub
Verified signature88d5299 -
Upload app.py with huggingface_hub
Verified signaturec9d8d7a -
Upload folder using huggingface_hub
Verified signaturec2799fb -
Upload app.py with huggingface_hub
Verified signaturee99c86d -
Upload app.py with huggingface_hub
Verified signature37ce01e -
Upload folder using huggingface_hub
Verified signature520e139 -
Upload app.py with huggingface_hub
Verified signatureadb373e -
Upload app.py with huggingface_hub
Verified signature0eec56c -
Upload app.py with huggingface_hub
Verified signatureb2c144e -
Deploy AC3 Complete Refactor - Deterministic Styling + AI Composition
Claude42f73ba