[FEATURE] Support Vertex AI multi-region endpoint URL pattern (.rep.googleapis.com)

[fwuensche]

Scope check

• This is core LLM communication (not application logic)
• This benefits most users (not just my use case)
• This can't be solved in application code with current RubyLLM (only via monkey-patching internals)
• I read the Contributing Guide

Due diligence

• I searched existing issues
• I checked the documentation

What problem does this solve?

Google Vertex AI exposes generative models through three URL families:

Location kind	Host
Per-region (e.g. europe-west9)	https://<region>-aiplatform.googleapis.com
Multi-region (eu, us)	https://aiplatform.<eu|us>.rep.googleapis.com
Global	https://aiplatform.googleapis.com

RubyLLM::Providers::VertexAI#api_base handles per-region and global but not multi-region. Setting vertexai_location = "eu" currently builds https://eu-aiplatform.googleapis.com/..., which does not resolve, producing an HTML 404 page from Google.

This matters today because Gemini 3.5 Flash (GA on May 19, 2026) is only published to multi-region and global endpoints — no individual region serves it (verified empirically across europe-west1/3/4/8/9, europe-central2, europe-north1, europe-southwest1, us-central1 — all return Publisher Model not found). Any RubyLLM user who needs EU data residency for 3.5 Flash (GDPR / HDS / French health hosting) currently has no path forward without patching the gem.

Proposed solution

Either of these would solve it; option A is more transparent, option B is more general:

Option A — Recognize eu/us as multi-region in the URL builder. Roughly:

def api_base
  case @config.vertexai_location.to_s
  when 'global'  then 'https://aiplatform.googleapis.com/v1beta1'
  when 'eu', 'us' then "https://aiplatform.#{@config.vertexai_location}.rep.googleapis.com/v1beta1"
  else "https://#{@config.vertexai_location}-aiplatform.googleapis.com/v1beta1"
  end
end

Option B — Add a vertexai_api_base config option mirroring openai_api_base, gemini_api_base, perplexity_api_base, etc. The Vertex provider is currently the only Google provider without an api_base override (see also #774 for Perplexity).

For reference, our current monkey-patch (an Ai:: module we prepend on RubyLLM::Providers::VertexAI):

module VertexAiMultiregion
  MULTI_REGION_LOCATIONS = %w[eu us].freeze

  def api_base
    location = @config.vertexai_location.to_s
    if MULTI_REGION_LOCATIONS.include?(location)
      "https://aiplatform.#{location}.rep.googleapis.com/v1beta1"
    else
      super
    end
  end
end

RubyLLM::Providers::VertexAI.prepend(VertexAiMultiregion)

Why this belongs in RubyLLM

URL construction for a provider's published endpoints is core to LLM communication. Multi-region endpoints are documented public Vertex AI infrastructure (not a niche or experimental feature), and they are the only way to access certain Gemini models from EU-residency-sensitive workloads. The fix is provider-internal — no application-level workaround is possible without monkey-patching api_base. Other providers in RubyLLM already expose <provider>_api_base overrides; aligning Vertex with that pattern (or auto-handling the standard multi-region locations) would remove the divergence.