Feature roadmap: top three needed features for llmfit

[three-foxes-in-a-trenchcoat]

We need to prioritize the next set of core capabilities for llmfit. Based on community feedback and current gaps, the top three features to implement are:

1. Kubernetes deployment support – Enable llmfit to be deployed as a scalable service on Kubernetes clusters, with Helm chart and operator integration. This will simplify production deployments and align with the project's containerization goals.
2. Built‑in automatic benchmark module – Provide an out‑of‑the‑box benchmark suite that runs common model workloads, captures latency/TPS, and stores results for later recommendation/calibration. This feature underpins the "best model" scoring and helps users evaluate hardware choices.
3. ONNX model format support – Add import and inference pipelines for ONNX models, broadening the model ecosystem beyond llama.cpp and GGUF. This will attract users with existing ONNX assets and improve interoperability.

These features address the most requested use‑cases and lay the groundwork for future extensions such as hardware‑aware model selection and advanced UI integrations.

Please add any additional high‑priority items or adjust the ordering as needed.

[AlexsJones]

I dont think the deployment method is crucial, sympozium already deploys llmfit as a skill onto a container and mounts it against the node.