Phase 2: Templates + Agent Catalog¶
Goal¶
Expand from 3 to 15+ agents. Build out all template types. Enable community contributions.
New templates¶
template-agent-translator¶
- NLLB-200 (200 languages)
- Language detection
- Side-by-side input/output
- Clipboard integration
template-agent-summarizer¶
- BART-small or T5-small
- URL fetcher (via CORS proxy) or text paste
- Adjustable length
- Key points extraction
template-agent-code¶
- WebContainers for tooling (ESLint, tsc, Prettier)
- Optional Ollama for LLM review
- Syntax highlighting (CodeMirror)
- Multi-file support
template-agent-chat¶
- WebLLM (Llama 3.2 1B or Phi-3 mini)
- Conversation UI
- System prompt customization
- Optional Ollama backend
template-agent-ocr¶
- Florence-2 or Tesseract.js
- Camera capture + image upload
- Structured output (JSON, text, table)
- PDF support via PDF.js
template-agent-diffusion¶
- Stable Diffusion (WebGPU)
- Text prompt input
- Generation progress
- Gallery of results
- Download/share
template-agent-embeddings¶
- all-MiniLM-L6 (23MB)
- Upload documents → build local search index
- Semantic search across documents
- IndexedDB persistence
template-agent-data¶
- CSV/JSON parser
- Chart.js for visualization
- Small LLM for natural language queries
- Export results
New agents (built from templates)¶
| Agent | Template | Unique features |
|---|---|---|
| Translator | agent-translator | Saved translations history |
| Article Summarizer | agent-summarizer | URL input, RSS feed support |
| Code Linter | agent-code | ESLint + Biome profiles |
| Local ChatBot | agent-chat | System prompt library |
| Receipt Scanner | agent-ocr | Expense categorization |
| Image Creator | agent-diffusion | Prompt suggestions |
| Document Search | agent-embeddings | Multi-document support |
| CSV Analyzer | agent-data | Auto-chart suggestions |
| Sentiment Tracker | Custom (DistilBERT) | Social media monitoring |
| Flashcard Maker | Custom (T5) | Spaced repetition |
| Font Previewer | Custom (no model) | Google Fonts integration |
| Color Palette | Custom (CLIP) | Image-to-palette |
Community contribution flow¶
- Developer runs
fags init my-agent --template agent-whisper - Builds their agent locally
- Runs
fags check(compliance) - Runs
fags publish(creates repo, provisions hosting) - Agent appears on store after admin approval
Shared model cache (Phase 2 feature)¶
Service Worker at the store level that caches models across agents:
User opens Transcriber → downloads Whisper (240MB)
User opens Meeting Notes → Whisper already cached!
User opens Podcast Summarizer → Whisper already cached!
Reduces bandwidth and improves UX across the agent catalog.
Definition of done (Phase 2)¶
- [ ] 8 templates available
- [ ] 15+ agents on store
- [ ] Shared model cache working
- [ ] Community publish flow documented
- [ ] VibeCode agent can build new agents from description
- [ ] Categories and search on store site