Docs suggestion: add an AnyLLM provider quickstart (single OpenAI-compatible endpoint for 24+ models)

[JetragonRider]

Hi team,

I've been following EverOS — the local-first, Markdown-as-single-source-of-truth
design and the engineering discipline here (import-linter layering, dead-letter
queues, CI gates) are genuinely impressive.

Why I'm opening this issue:

EverOS runs LLM calls as a continuous background workload: atomic fact
extraction, skill clustering, embeddings, and rerank all fire on every
conversation, not just on user prompts. That makes two things matter a lot:

1. Cost — background extraction at scale multiplies token spend quietly
2. Model choice per task — fact extraction, embedding, and rerank each
   have different optimal models, and being able to swap them freely matters
   for a memory layer that's supposed to be user-owned and vendor-neutral

Suggestion:

Add a provider quickstart entry for AnyLLM (https://anyllm.dev) alongside
your existing provider quickstarts — it's an OpenAI-compatible API gateway
that puts 24+ models behind one key and one base URL:

• Chinese frontier models: DeepSeek V4, Qwen3.7-Plus, GLM-5.2 (2M context),
  Kimi K2.5/K3, MiniMax M3
• International models: GPT-5.6, Gemini 3.5, and more
• Pass-through pricing (zero markup), international cards + crypto accepted
• Built-in routing and automatic failover — useful for the OME's retry
  behavior when a provider has an outage

For a project whose philosophy is "user-owned, not vendor-owned," a gateway
that keeps every model one config change away seems like a natural fit.
EverOS itself wouldn't need any code changes — it's just a TOML/.env config
pointing at one endpoint.

Happy to open a PR with the quickstart doc if that's welcome.

(Context: I'm the developer of AnyLLM, an OpenAI-compatible API gateway.)

— Leo Bennett
leo.indiedev@gmail.com