Why Multi-Provider?
Every AI provider has strengths. Midjourney handles aesthetics differently than DALL-E. Claude reasons differently than GPT-4. Building on a single provider means inheriting their blind spots.
HobFarm routes through multiple providers using Cloudflare’s AI Gateway. One request, multiple responses, structured comparison. The architecture is simple: a Workers-based routing layer that normalizes inputs and outputs across providers.
The Stack
- Cloudflare Workers for compute at the edge
- AI Gateway for provider routing, logging, and fallback chains
- R2 for media storage with CDN delivery
- D1 for structured data (Grimoire knowledge base, analytics)
- KV for ephemeral state (session data, rate limiting, provider health)
Lessons Learned
Provider APIs change constantly. Abstractions that work today break tomorrow. The key insight: keep the routing layer thin. Don’t try to normalize everything. Let provider-specific features through when they matter, abstract only the common patterns.
The second insight: cost awareness needs to be a first-class concern, not an afterthought. Every request has a price. Route based on value, not just capability.
What’s Next
StyleFusion already uses this architecture for image generation. Grimoire and HobBot are next. The goal: every HobFarm tool shares the same provider routing, the same cost tracking, the same fallback chains.
