Cache as L1
Stable prompt fingerprints turn repeated AI requests into edge reads.
Edge AI Assistant
A Cloudflare-first product blueprint for routing, caching, and governing GPU-backed AI inference without exposing private API implementation code.
edge-control-plane
prompt fingerprint
cache hit
context window
4.1k chars
route decision
small model
Public surface
No API source
Runtime target
Cloudflare Pages
Core thesis
Manage AI, not models
Architecture without leaking implementation.
The public project describes the control plane: prompt classification, cache-first execution, model dispatch, memory tiers, and cost gates. Private deployment details stay outside the repository.
Designed as a low-cost inference router.
The project positions Cloudflare as the fast decision layer and external GPU inference as the expensive execution layer.
Routing as control logic
Prompt length, intent, and context pressure select the cheapest viable model.
Memory as hierarchy
Short-term state, durable history, and object storage remain separate tiers.
Launch path
The repository is intentionally presentation-grade: it can be shared, starred, and connected to Cloudflare Pages without publishing private service code.
- Public blueprint Architecture, positioning, and product narrative.
- Private control plane Worker logic, tokens, routing policy, and billing rules remain internal.
- Hosted demo shell Cloudflare Pages publishes the polished surface globally.