Support multiple AI LLM providers and/or multiple AI LLM model names

Feature requests
1

Title: Support multiple AI LLM providers and/or multiple AI LLM model names

  1. What is your original issue/pain point you want to solve?
    Deterministically control configuration which Zammad AI services send prompts to which provider and/or provider model

  2. Which are one or two concrete situations where this problem hurts the most?
    Zammad’s single provider and single model configuration for everything forces compromise.

No single model wins across all modalities, speeds, and cost profiles.
User-perceived latency and Synchronous / Blocking Experience for Zammad user agent when performing AI text writing assistant during highlighting draft text message composition editing requires different configuration optimization for LLM provider & LLM model versus Hidden Tasks, Asynchronous / Non-Blocking Background Worker run by automated triggers where Zammad user agent does not perceive LLM latency.

Different LLM parameter models have different prompt processing and token generation speeds.

LLM for AI Agent ticket tagger or title rewriter may require different capabilities, specialization, resources than AI writing assistant.
LLM for AI text recognition OCR requires different model capability than ticket Summary Services generation. ( I realize that Zammad supports specifying OCR Model Provider — Zammad Admin Documentation documentation which probably works with the large providers. When testing on 127.0.0.1 localhost in order to run a different llamacpp service for OCR model functions on different network ip:port number which would require Zammad recognize the other port number as a separate provider. Immediate solution for localhost are using LLM sidecar llamacpp model hotswap service or a gateway service which would allow localhost LLM gateway to proxy both non-OCR and OCR context to correct local LLM port.)

  1. Why is it not solvable with the Zammad standard?
    page /#ai/provider only support single service provider bound to single model name.

update: I just learned today that ollama supports running multiple models spawned by single service ollama/docs/faq.mdx at main · ollama/ollama · GitHub . asap i will try to test this. nevertheless, ollama is sub-optimal choice for LLM because of numerous reasons.

  1. What is your expectation/what do you want to achieve?
    Support specifying at least more than one particular model name from one or more providers. If supporting more than one provider cannot be easily accomplished, perhaps at least support more than one model from a provider to allow Zammad admin to specify different LLM model for particular task, ie a small purpose-built OCR vision LLM model that excels for recognizing image text, a different medium model for performing trivial rote tasks ( categorization, tagging etc) , a different larger model performing more complex tasks like writing assistance etc.

Your Zammad environment:

  • Average concurrent agent count: 2
  • Average tickets a day: 20
  • What roles/people are involved: tbd

Entire Zammad community and Zammad SaaS would benefit by intelligently routing different prompt context requests to most appropriate efficient provider and model thereby increasing speed, increasing response output quality, and control or reduce costs.

Most importantly, v7.1 is really excellent!