Cloudflare Workers — Edge Compute for Agent API Proxies

Pros

• –Sub-5ms cold start globally — effectively no cold start penalty vs AWS Lambda
• –Workers AI provides GPU inference at edge — useful for embedding generation near user
• –KV, Durable Objects, and R2 provide consistent primitives for stateful edge agents
• –Free tier includes 100k requests/day — adequate for prototyping and low-traffic APIs

Cons

• –128MB memory limit and 10ms CPU time limit (50ms on paid) constrain heavy compute tasks
• –No Node.js built-in modules — many npm packages require polyfills or rewrites
• –Durable Objects have regional affinity — global consistency patterns require careful design
• –Wrangler CLI tooling has occasional breaking changes between major versions