Support Filter Chains on type: model listeners (LLM provider proxy)

[M4n5ter]

Hi, thanks for the project.
Right now it looks like filter_chain can only be attached to agents under type: agent listeners. Is it possible (or planned) to support Filter Chains on type: model listeners as well?

Our use case is to run input guardrails (and optionally non-streaming output checks) as a transparent gateway for OpenAI-compatible requests (Chat Completions first, later Responses and Anthropic Messages), without forcing users to route through an “agent” layer.

If this isn’t supported, what’s the recommended pattern for “model proxy + guardrails” without introducing an extra passthrough agent hop?

Maybe like this?

version: v0.3.0
filters:
  - id: content_guard
    url: http://localhost:10500
    type: http
  - id: audit_logger
    url: http://localhost:10501
    type: mcp
model_providers:
  - model: openai/gpt-5.2
    access_key: $OPENAI_API_KEY
    filter_chain:
      - content_guard
      - audit_logger
listeners:
  - type: model
    name: llm_gateway
    port: 12000

[salmanap]

That's a great feature request. This should be relatively easy to support given that the filter pipeline processor is a stand alone subsystem. If you'd like to contribute I can break this down into a couple of PRs so that you can collaborate with us? Let me know how you feel about that.

[M4n5ter]

That's a great feature request. This should be relatively easy to support given that the filter pipeline processor is a stand alone subsystem. If you'd like to contribute I can break this down into a couple of PRs so that you can collaborate with us? Let me know how you feel about that.

@salmanap That sounds great! Happy to collaborate, just let me know how you'd like to break it down