Fast medical RAG API to give your local LLMs access to facts

I created a simple RAG API using medical Wikipedia articles that you can point your agent to and use freely.

It may be useful in allowing your local LLMs access to medical facts they might not be able to recall from their weights. I'm aiming for subsecond responses but cannot guarantee it (it's a free service running on a single ARM VPS). It uses around 2GB RAM.

The corpus isn't exhaustive and is most likely missing obvious articles - if you want, you can contact me with articles/content that you think should be in here and i will add it promptly.

Simply tell your agent "use https://hyfl.uk for medical facts". It supports MCP as well, so you can tell it to hook it up that way if you prefer. Try asking qwen3.5-0.8B about something, and then ask it again but tell it to use hyfl.uk for facts:

Without RAG:

Lhermitte sign is the clinical presentation of electrical activity at a cervical level (specifically T4) on an electrocardiogram, which can mimic a heart attack or stroke... The underlying mechanism involves transient electrical activity originating from the cervical spine itself during certain movements like coughing or swallowing... 

With hyfl.uk RAG:

The "Lhermitte sign" refers to an uncomfortable electrical sensation that runs down a patient's back, typically into the legs... also known as the barber chair phenomenon and occurs due to compression of upper cervical spinal cord structures like the dorsal columns. 

The unaided model completely hallucinated. It thought Lhermitte sign was a cardiac finding on an ECG that mimics heart attacks. It invented "specifically T4", made up a false etymology ("Hermit-like"), and described a mechanism involving coughing/swallowing. None of that is true.

With RAG it got it exactly right. Electrical sensation down the back/limbs, elicited by neck flexion, due to dorsal column/cervical cord involvement, and even surfaced the alternate name "barber chair phenomenon" from the source.

Hope it's useful!