Disclosure: I maintain Lynkr, the self-hosted gateway discussed in the second half. OpenRouter and Requesty are good products — this post is about understanding what you're paying for so you can decide whether you need to.
Hosted LLM routers had a huge 2026 — OpenRouter alone pushes 25 trillion tokens a week. The pitch is real: one API key, 400+ models, automatic failover. The price is a ~5% fee on every token you route (5.5% on OpenRouter credits, 5% on Requesty), plus a subtler cost: every prompt, every file your coding agent reads, every secret that leaks into a context window transits their infrastructure.
For a hobby project, 5% of a small bill is nothing and the convenience wins. For an agentic coding workload — where teams routinely spend $500–$2,000 per engineer per month — 5% is real money, and the data-transit question stops being academic. So it's worth asking precisely: what does the hosted router actually do for that fee, and which parts can you self-host?
What the fee buys
Unified API across providers — one format in, translated per-provider out.
